The AI-driven memory apocalypse has made computer parts so expensive that the average buyer has stopped building PCs. Multiple research firms have already said that, but we’re now seeing it in action in the market. Uniko’s Hardware shared on X four crazy combos of RTX 5090 GPUs being sold on the Taiwanese e-commerce platform PChome24h.

Some of the GPUs that come with a lot of "freebies" include the Asus ROG Astral GeForce RTX 5090, Gigabyte Aorus RTX 5090 Master, MSI GeForce RTX 5090 Gaming Trio OC, and a Gigabyte Aorus RTX 5090 AI Box external graphics card. These items are priced in NTD, but we’ll list them below, including their retail prices in dollars (NTD 32.25 = USD $1 exchange rate) and the included items.

As we can see, even though GPU prices are quite inflated, many of them come with enough freebies that you can start building one high-end system and three to five entry-level gaming rigs. They’re still priced relatively high compared to Amazon or Newegg in the U.S., though, but we have to keep in mind that Taiwan is a different market, so there will likely be variations in prices plus VAT.

While this might be an amusing piece of trivia for some people, it’s actually evidence of a disturbing trend in the PC building industry. Multiple research firms have been saying that the market will contract by double digits this year, with the entry-level PC market being hit the hardest, and that it’s not expected to bounce back until 2029. These combos are just evidence of that, at least in Taiwan.

That’s because sellers often put slow-moving products in a combo to help get their inventory moving. Given the number of motherboards, entry-to-mid-range GPUs, and other components included in these combos, it’s safe to say that these items aren’t selling and retailers are doing all these crazy deals just to free up capital.

As the AI-driven memory crisis has pushed up the prices of memory and graphics cards, many have already paused or delayed their purchases and upgrades. Only those with deep pockets who can afford to spend thousands of dollars on parts and combos like these are still buying PC parts.

A powerful processor is at the heart of every exceptional gaming PC, and nothing rivals the performance of AMD's Ryzen 7 9800X3D. If you are building a new system or updating an existing one, this Ryzen 7 9800X3D bundle, which includes an MSI B850 Gaming Plus WiFi motherboard and an MSI MAG Coreliquid A13 240 AIO liquid cooler, is for you. This curated package is available for $458.99, a massive $160 off the regular price of purchasing the components separately.

• Check out the Ryzen 7 9800X3D bundle deal on Newegg

The Ryzen 7 9800X3D is the best gaming processor that money can buy right now. With an octa-core, 16-thread design and a massive 96MB of L3 cache, the Ryzen 7 9800X3D not only delivers top-notch gaming performance, but it also excels in productivity tasks. Whether you are assembling a system for gaming or serious work, the Ryzen 7 9800X3D provides a rock-solid platform for your build.

While the exclusive bundle does not include any DDR5 memory, it is not a big drawback. Thanks to its design, the Ryzen 7 9800X3D maintains stellar gaming performance regardless of whether you pair it with slow memory or even a single memory module. Therefore, you do not need to spend a fortune on premium memory, which is a big plus in this ongoing memory shortage.

On the motherboard side, Newegg's premium bundle includes a feature-rich MSI B860 Gaming Plus WiFi motherboard. It boasts a 12+2+1-phase power delivery subsystem that maximizes the Ryzen 7 9800X3D's performance while also offering ample headroom for overclocking for those who want to extract extra performance.

The B860 Gaming Plus WiFi comes equipped with four DDR5 memory slots that can accommodate up to 256GB of memory with blazing data rates up to DDR5-8200. Meanwhile, storage options arrive in the shape of three high-speed M.2 slots (one at PCIe 5.0 and two at PCIe 4.0) and four standard SATA III ports. The motherboard also has a PCIe 5.0 x16 expansion slot for pairing the Ryzen 7 9800X3D with one of the best graphics cards.

While the Ryzen 7 9800X3D is not a processor that runs very hot, you can never have too much cooling. The MSI MAG Coreliquid A13 240, a solid 240mm AIO liquid cooler, uses a high-performance pump, a thick radiator with wider water channels for a higher flow rate, and two pre-installed 120mm PWM fans that are quiet and performant.

The Ryzen 7 9800X3D bundle will be available on Newegg for the entire month. The massive 26% discount translates to real, tangible savings, bringing the bundle down to just $458.99. It is a rare opportunity to secure the best gaming processor with a very decent motherboard and 240mm AIO liquid cooler at a price that you probably will not find elsewhere. Given the popularity of the Ryzen 7 9800X3D, the bundle will likely sell out very fast, so do not think too much before swiping your credit card.

If you're looking for more savings, check out our Best PC Hardware deals for a range of products, or dive deeper into our specialized SSD and Storage Deals, Hard Drive Deals, Gaming Monitor Deals, Graphics Card Deals, gaming chair, or CPU Deals pages.

The community of Gilroy, California, about 30 miles southwest of San Jose, was surprised when an Amazon data center started construction in their town last month. But as many projects like these have been suffering from opposition and delays across the country, the Wall Street Journal reports that this specific site avoided scrutiny by adhering to local zoning rules set 45 years ago.

Amazon’s Gilroy project sits on a 56-acre piece of farmland between a Walmart Supercenter and the Gilroy Premium Outlets. The company and the city didn’t exactly keep the project secret when it started its application in 2020. In fact, there were disagreements between the tech giant and the city government during the process. The publication noted that some Amazon officials were frustrated by slow permitting progress, which was delayed by at least one year. At the same time, the city pushed back on some of the company’s requests, especially if they were no longer up for discussion.

Roger Wehner, Amazon Web Services vice president of economic development, told WSJ that the site had gone through a long approval process, which required public notices and public-comment periods. However, when Gilroy resident Rosa Rodriguez wanted to state her concern about the damage a data center can do to a region suffering from drought, she was told that the comment period had already lapsed in 2024, long before data centers became a hot topic across the country.

Gilroy City Mayor Greg Bozzo said that other industrial projects that had a similar scale to the Amazon data center have previously been approved without resistance.

“It didn’t bother them that a food-distribution center came to Gilroy without their engagement, but now it bothers them that a data center is coming to Gilroy without them having the possibility of engagement,” Bozzo said.

Still, he conceded the people are now more aware about data center developments and added, “We…have an opportunity to change the way that we do business in Gilroy. We’re adapting to the times.”

The tech giant also said that it’s engaging more with the community, with Amazon officials claiming that they met residents directly and even hosted an open house where people can ask questions and leave comments. It also addressed fears that the development would cause a water shortage, which other data centers have already caused, saying the company is constructing a wastewater recovery system to supply its water needs and will even build a pipeline to bring wastewater from the source to its plant.

Bozzo also highlighted the benefits that Amazon will bring to the city, especially increased tax revenue, construction jobs, and a million-dollar donation towards a new fire truck. He also pointed out that the tech giant has been giving grants to non-profit organizations in the city, as well as sponsoring the annual Gilroy Garlic Festival.

But at this point, it’s not the Amazon data center the people of Gilroy are having an issue with. Instead, it’s the fact that a major project that has the potential to have a major impact on the community has been approved without the knowledge of many residents.

There are plenty of ultra-lightweight gaming mice in our best gaming mice picks — but there aren't any that have a built-in hand cooler.

Pulsar recently released a collaboration with infamous air-cooling advocate Noctua — a beige and brown mouse full of top-tier hardware and holes and a built-in 40mm NF-A4x10 fan. For cooling your hand, not the hardware. This eSports-oriented gaming mouse features Pulsar's XS-2 sensor, which has a maximum resolution of 42,000 DPI and a maximum speed of 800 IPS, and can handle up to 70 G's of force. It also features up to an 8,000 Hz polling rate, and, of course, the built-in fan. It retails for the same price as the original Feinmann F01 gaming mouse, which is to say, it's pretty pricey at $179.99.

Design and Comfort of the F01 Noctua Edition

The F01 Noctua Edition is a lightweight wireless eSports gaming mouse with an ergonomic right-handed shape and a hole-filled chassis with a built-in fan. It comes in Noctua's signature beige-and-brown color scheme: The top part of the chassis is light beige, while the bottom and accents (side buttons, scroll wheel, Noctua's logo plate, and some interior elements) are brown.

The F01 Noctua Edition has a right-handed, ergonomic shape with contoured primary mouse buttons and an off-center hump. It's a little smaller than most mainstream gaming mice, measuring 4.79 inches (121.7mm) long by 2.62 inches (66.5mm) wide and 1.7 inches (43.3mm) high at its thickest point. It weighs 2.57 ounces (73g), which is on the heavier side for what is supposed to be an ultra-lightweight eSports mouse. It's about an ounce heavier than the regular Pulsar Feinmann F01, which weighs 1.62 ounces (46g), thanks to the built-in fan.

The mouse features a matte chassis made of a carbon composite material for durability. The chassis is full of holes, but at least they serve a purpose (sort of) other than reducing weight and triggering your trypophobia: they're also there to keep your hand cool. In the center of the mouse you'll see a mini Noctua fan that directs air upward — toward your palm. You can activate this fan by holding down both side buttons and scrolling up. There are four different fan speeds, indicated by the small LED indicator light on the left side of the mouse, which you can cycle through by scrolling up or down with both. Holding down both buttons and scrolling down turns the fan off.

Button-wise, it's pretty standard: primary mouse buttons, two side buttons, a power button / DPI switch on the bottom, and a ridged, notched scroll wheel made of aluminum alloy. The mouse feels sturdy and the metal scroll wheel makes it seem a little fancier — it definitely looks and feels pretty premium. I did think the side buttons felt like they were placed a little too far back, but after comparing the F01 Noctua Edition with other mice, the actual issue is that the side buttons are larger than average — and this results in the back button sitting further back on the chassis. It wasn't a dealbreaker, but I found it mildly uncomfortable.

The bottom of the mouse features two large PTFE mouse feet at the top and bottom, as well as a strip of PTFE around the sensor. There are five geometric cutouts surrounding the sensor, as well as the power button / DPI switch.

The mouse comes with a separate 8K polling dongle, which has two buttons (a programmable profile switch button and a DPI switch), and a 6-foot (1.8m) detachable, braided USB-C to USB-A cable. Both the dongle and the cable come in Noctua's signature colors (the dongle is brown and the cable has a beige fabric covering with brown accents).

Specs

Performance of the F01 Noctua Edition

The F01 Noctua Edition mouse features Pulsar's XS-2 optical sensor, which has a maximum resolution of 42,000 DPI and a maximum speed of 800 IPS, and can handle up to 70 G's of force. The mouse also features Pulsar's optical switches in its primary buttons (rated for up to 100 million clicks), and can get up to an 8,000 Hz polling rate. The sensor doesn't disappoint — its tracking was accurate and speedy in my tests, and it worked perfectly across a variety of surfaces (including glass). The only thing I can see affecting performance in this mouse is its weight — it's not heavy, but it's heavier than most of its competition.

As I mentioned earlier, at 73g the mouse is on the heavy side for an ultra-lightweight eSports mouse — and it's an entire ounce heavier than the fan-less version. I guess that makes sense considering... there's a fan inside it, but still. If you're looking for an ultra-lightweight gaming mouse and you're not particularly intrigued by the fan, there are lighter, cheaper mice out there, such as my favorite Razer DeathAdder V4 Pro, which weighs 56g and costs $169.99.

I'll admit I was a bit skeptical about the fan — I'm not really a person who suffers from sweaty palms while I'm gaming, however, so I'm aware I'm not the target audience. But as someone who is not the target audience, I have to say that the fan is actually pretty cool (pun intended). It's small, yes, but you can definitely feel it — and even though I don't sweat much, it still felt pretty good. The fan has four speeds, and the first two are practically silent. You can hear the two fastest speeds, but it's not particularly loud — you'll notice it, but those around you probably won't.

Activating the fan using the two side buttons and scroll wheel was easy enough, though the side buttons will activate when you press them, so it's not perfect. You can also turn on and adjust the fan speed in the software and/or you can program one of the buttons on the dongle to cycle through speeds.

If you do get sweaty palms while gaming, or you just happen to live and game in a hot environment, the fan in this mouse definitely works. It takes a toll on battery life, of course, but you can always plug it in. That said, that part of the mouse is practically non-existent, the holes are so big (it's even more full of holes than the Corsair Sabre v2 Pro Wireless MG). The structure of the mouse honestly keeps your hand — or at least your palm — pretty cool, even without a fan.

Features and Software of the F01 Noctua Edition

The F01 Noctua Edition is largely controllable via the hardware, which is nice. The fan can be activated and adjusted by holding down the two side buttons and scrolling, and DPI can be cycled through using the DPI switch button on the bottom of the mouse or the DPI switch button on the wireless dongle. The dongle also has a function button that cycles through onboard profiles by default — so you will need to connect to Pulsar's web-based peripheral software, Bibimbap, if you want to take advantage of this.

With Bibimbap — which you can connect to wirelessly — lets you remap keys (there are five programmable buttons, including the scroll wheel click) and has a built-in macro recorder. You can also adjust various settings, including DPI steps (up to 6), polling rate (ships with 1,000 Hz and goes up to 8,000 Hz), LOD or liftoff distance, and debounce time. You can also turn on the fan from this app, if you want. You can save up to five profiles directly on the mouse and cycle through them with the function button, or you can remap the function button to cycle through polling rate, LOD, debounce time, motion sync, turbo mode, and fan mode. That last one just does the same thing that holding down both mouse buttons and scrolling does, but it doesn't accidentally activate the side button functions.

Wireless Experience and Battery Life of the F01 Noctua Edition

The F01 Noctua Edition features two forms of connectivity: 2.4GHz wireless and wired (via USB-C). It doesn't have any RGB lighting, but it does have a built-in 40mm 5V PWM NF-A4x10 fan with four different speed settings. Oh, and an 8,000 Hz polling rate. Needless to say, its battery life varies quite a bit, depending on how you have it set up. Its built-in battery is 500mAh, which is... well, don't expect much when it comes to battery life. Based on my observations as I tested the mouse, I think you'll still be able to eke out a full day's worth of usage if you have the fan running at top speed, which is not a lot compared to most wireless mice, but not bad, considering.

Bottom Line

The F01 Noctua Edition is a very well-built lightweight gaming mouse, but its main selling point is definitely the built-in fan. And the built-in fan works pretty well — I was definitely impressed, even if I don't see myself using it very often (or at all, really). But if you don't need a mouse with a built-in fan, it's hard to recommend the F01 Noctua Edition over just about any other lighter, more affordable mouse, including Pulsar's own fan-free Feinmann F01. The Corsair Sabre v2 Pro Wireless, for example, weighs just 56g and has a shell full of holes, gets up to an 8,000 Hz polling rate and retails for $149.99.

The best RAM on the market generally utilizes integrated circuits (ICs) from industry giants such as SK hynix, Samsung, or Micron. Recently, however, Chinese chipmaker ChangXin Memory Technologies (CXMT) has emerged as a strong competitor amid the ongoing global memory shortage. In Colorful's latest overclocking exhibition, CXMT's ICs achieved speeds up to DDR5-8800 on AMD's X870E platform, keeping pace with the best overclocking ICs, such as SK hynix's A-die and M-die.

Many often regard CXMT as an underdog in the memory industry, particularly when compared to the "Big Three." Founded only a decade ago, the ambitious Chinese chipmaker eventually became China's largest domestic memory manufacturer and is now the fourth-largest memory maker globally.

The recent memory shortage has accelerated CXMT’s rise to fame. The chipmaker has been silently gaining ground in the memory market. Its momentum has been especially evident in the DDR5 segment despite the company having only unveiled its first DDR5 ICs in November 2025. Many top Chinese memory brands, including Asgard, Colorful, Gloway, and KingBank, are widely using CXMT ICs, and even Corsair has done so to a certain extent.

Colorful showed off an incredible overclock with its iGame Shadow II DDR5-6000 C34 memory kits, pushing both the 32GB (2x16GB) and 48GB (2x24GB) configurations far beyond their rated speeds. These kits reached DDR5-8812 and DDR5-8817, respectively, on Colorful's iGame X870E Vulcan W OC motherboard. Unfortunately, the screenshots provided by Colorful were too blurry to discern detailed specifications such as memory timings or DRAM voltage. For context, some high-end DDR5-8800 memory kits typically operate at timings around 42-55-55-140 with a voltage of 1.45V.

The noteworthy aspect is that Colorful achieved the overclocking feat on 16GB memory modules. Most, if not all, DDR5-8600 and faster memory kits currently on the market come with 24GB memory modules. The reason is that many use SK hynix’s 24Gb (3GB) M-die ICs, produced on the chipmaker's advanced 1a-nm (~14nm) process node with Extreme Ultraviolet (EUV) lithography. These M-die ICs are the go-to choice for memory vendors thanks to their superior density, power efficiency, and ability to scale to higher frequencies.

In comparison, CXMT’s accomplishments are notable when considering the technological challenges. Due to U.S. export restrictions on China, CXMT does not have access to cutting-edge EUV equipment, so its 16nm node, which it uses for DDR5 ICs, leverages Deep Ultraviolet (DUV) machines instead. CXMT’s 16nm process is roughly in the same alley as the 1z nm (~14–15nm) nodes previously used by the Big Three, so CXMT is easily a few years behind the top-tier process nodes.

The reality is that CXMT will have to make do with what it has. The chipmaker is already developing a 1a-nm node but on DUV. It is far from prime time, though.

In addition to seeing CXMT ICs in mainstream memory kits, companies such as HP, Asus, and Acer have already started selling consumer devices, such as laptops or desktops with CXMT memory. Even Apple reportedly wants to buy memory chips from CXMT.

CXMT recently went public via IPO and saw its stock price skyrocket by 466%, raising over $8.6 billion. Without access to EUV machinery, CXMT will focus primarily on memory production and is rapidly building factories across China. The ambitious Chinese chipmaker hopes to steal 30% of the market share from the Big Three by 2030. It is just a matter of time before we start seeing CXMT ICs in all the prominent consumer memory brands.

An extensive Atari ST floppy disk database can now easily be accessed by all netizens. Originally created to serve floppy images to the owners of the SidecarTridge multi‑function hardware add‑on for the Atari ST family (ST, STE, Mega ST, TT, Falcon), anyone can now browse, search, and download the .ST files to use with other hardware, emulators, or even to rewrite the disk images to actual floppies.

After reading the announcement, I decided to check out the online repository. My favorite old platformer series was there, with more than one disk featuring the legendary Turrican and Turrican II releases. I also checked and confirmed public domain/shareware favorites like Gravity Force were available, as was Llamatron, and the mono high-resolution Bolo. Some things I looked for, like the Stupendous Demo, seemed to be missing, as were serious art and DTP tools I used to dabble with in the 90s.

The database is pleasingly new and regularly maintained, though. For example, it has 41 Atari ST disks flagged as recent arrivals. These are a mix of old releases being added, and some extremely new titles like new homebrew games, and a collection of demos released at Sommerhack 2026 just last month.

As indicated in the intro, this disk database was originally collated for direct access via hardware designed to bring several modern conveniences to those who still cling to original Atari ST-family computers. The SidecarTridge slots into the computer’s cartridge port, hence the name. It is much more than a ‘Gotek-style’ flash card floppy loader, though. This Raspberry Pi Pico W RP2040 microcontroller-powered device can also load different ROMs (the Atari ST’s entire OS and GEM desktop were loaded from ROM for super-fast,t no-disk-needed booting) via its microSD card slot. It also provides hard disk emulation, networking features (Wi-Fi for downloading Atari ST Public Floppy Database disks), access to RTC, and modern keyboards, mice, and other peripherals.

We tested gaming performance with both NVMe and SATA SSDs across 11 games to determine if upgrading to a faster drive is worth it. Solid-state drives (SSDs) have been a staple of PC gaming for years, but their importance grew significantly with the launch of the PlayStation 5 and Xbox Series X|S in 2020. Both consoles feature high-speed NVMe SSDs and storage architectures designed to fully leverage their capabilities, enabling faster asset streaming and reducing traditional storage bottlenecks.

As a result, game developers have increasingly developed their titles for SSDs, with many modern games now listing them as a minimum system requirement. But how exactly do games use SSDs during gameplay, and does upgrading to a faster SSD actually improve gaming performance? In this article, we'll examine SSD activity in modern games and determine whether higher storage speeds translate into a better gaming experience. Given the surge in SSD prices, will a slower drive deliver comparable gaming performance to high-end alternatives?

Load Times

Load Times

While this article focuses primarily on SSD activity and performance during gameplay, loading times remain one of the most significant advantages of solid-state storage. It's well known that games can take several minutes to load from an HDD, whereas modern SSDs typically reduce load times to just a few seconds.

The gap between SSDs is much smaller. In some titles, PCIe NVMe drives can load games several times faster than SATA SSDs, but in many cases, the difference is relatively small. Naturally, different generations of PCIe SSDs have even smaller differences, with flagship drives often delivering only marginal improvements to loading times despite their substantially higher sequential throughput.

Disk I/O Operations in Games

Disk I/O Operations in Games

Before examining the performance results, it's important to understand the types of disk I/O operations that games typically generate. Two key metrics are block size and queue depth. The block size of an I/O operation defines the size of the chunk of data a game will read at one time, while queue depth is the number of pending I/O requests that a storage device has queued at any one time.

During gameplay, the vast majority of storage activity consists of random read operations as game assets are streamed from storage into memory. Sequential reads are less common, and write activity is generally minimal.

Microsoft provides several storage optimization recommendations for developers in its DirectStorage documentation. Although these guidelines are intended for DirectStorage-enabled titles, many of the underlying principles apply to games in general, even if they do not use DirectStorage. Among Microsoft's recommendations are the use of block sizes of at least 32k and sufficiently high queue depths to maximize storage throughput.

Modern games typically employ 32k or 64k random I/O requests. However, despite Microsoft's recommendations, most games continue to operate at relatively low queue depths.

Performance Comparison

Performance Comparison

Now we will examine SSD activity and compare performance across different SSD generations. The games included in our testing were selected because they either exhibited higher storage bandwidth than usual, had greater peak or average SSD utilization, had support for DirectStorage, or because they are popular titles that many gamers would be interested in seeing analyzed from a storage-performance perspective.

To place as much stress as possible on each SSD, all testing was conducted with 16GB of system memory. Using a smaller memory capacity reduces the amount of game data that can be held in RAM, forcing games to rely more heavily on the SSD than they would in a 64GB configuration. It also better reflects the configurations found in many mainstream gaming PCs. That said, we observed little difference in SSD activity when comparing 16GB and 64GB memory configurations.

When interpreting the SSD activity data below, it is important to understand that 100% utilization does not necessarily mean an SSD is reaching its maximum rated bandwidth. For example, a PCIe 4.0 SSD reporting 100% utilization is unlikely to be delivering its full 7.5 GB/s of sequential read throughput. In gaming workloads, SSDs can become fully utilized while transferring data at significantly lower bandwidths.

This is because the speeds advertised by SSD manufacturers are based on sequential read performance, whereas games primarily rely on random reads. While random-read throughput can approach sequential-read speeds under certain conditions, doing so requires sufficiently large block sizes and high queue depths. Although modern games often use relatively large block sizes, they typically do not use the high queue depths needed to fully saturate an SSD's theoretical bandwidth.

We have also chosen not to test any traditional hard drive disks. There are many documented examples of modern games performing poorly, taking an unbearably long time to load, or having visual issues when installing the game on an HDD. There are some games where this may not be an issue, but we still do not recommend playing modern games on an HDD.

The testing was conducted in gameplay. No built-in benchmarks were used. We found that SSD activity increases with higher settings, so that is where we focused our testing. We set each game to its absolute maximum settings.

We used HWiNFO to measure SSD activity and set it to the lowest polling rate possible to capture activity accurately without incurring overhead.

Test system

• MSI GeForce RTX 5090 Vanguard SOC
• Ryzen 7 9800X3D
• 16GB G.Skill Trident Z5 DDR5 6000 MHz CL30
• Crucial T700 Gen5 SSD
• Sabrent Rocket 4 Plus-G Gen4 SSD
• Adata XPG SX8200 Pro Gen3 SSD
• Samsung 870 EVO SATA III SSD
• Asus ROG Strix B850-F Gaming WiFi
• Corsair Nautilus 360 RS AIO Cooler
• MSI MPG Ai1600TS PCIe 5 PSU
• HAGS enabled
• Windows 11 25H2 (Build 26200.8737)
• Nvidia driver 610.62

Forspoken

Forspoken

Forspoken was the first game to support DirectStorage on PC. DirectStorage is a feature that aims to allow games to make full use of high-speed storage devices while reducing the CPU overhead associated with making reads from disk.

Looking at the table below, you can see that it does exhibit a fairly high peak and average bandwidth. In fact, it’s the highest average bandwidth we have seen in any game we have tested. Average utilization is also quite high, and the game hits 100% SSD utilization frequently.

How does this translate to performance?

While there is no difference in performance between the PCIe NVMe SSDs, you can see that having a PCIe SSD can give you a large boost in the 1% lows compared to a SATA SSD. We see a 39% increase in 1% lows at 4K, a 51% increase at 1440p, and a 22% increase at 1080p.

The higher peak and average bandwidth used by the PCIe drives resulted in higher 1% lows in this game, which makes for a smoother gameplay experience.

Alan Wake 2

Alan Wake 2

Alan Wake 2 makes extensive use of modern rendering technologies, including path tracing, mesh shaders, and RTX Mega Geometry. In a recent interview, Remedy explained that the game operates with a fixed memory budget, continuously streaming the assets required for the current scene from the SSD while evicting assets that are no longer needed. This approach keeps memory usage under control despite the game's highly detailed textures and geometry, making Alan Wake 2 one of the most storage-intensive titles available today.

The SSD activity increases with resolution in this game. We see an average of over 200 MB/s on each drive at 4K, which is rare for a game. We also see a whopping 5.1 GB/s of peak bandwidth on a PCIe 5.0 SSD at 4K.

How does this impact performance?

Unlike Forspoken, we do not see any difference in performance here for any of the drives tested.

Although the faster SSDs reach higher peak bandwidth, these peaks are only brief bursts rather than sustained transfers. In other words, the game is not continuously streaming multiple gigabytes of data every second. Instead, it transfers smaller amounts of data over very short intervals, resulting in high instantaneous bandwidth readings. As a result, the average bandwidth is the more meaningful metric, and it remains relatively similar across all of the SSDs we tested.

The average utilization on the SATA SSD can reach over 80% at 4K, but it is not enough to degrade its performance in comparison to the PCIe drives.

Ratchet and Clank: Rift Apart

Ratchet and Clank: Rift Apart

The disk bandwidth through this sequence is actually lower than the previous two games. This makes sense, as the game uses GPU decompression, which increases the effective bandwidth flowing through the PCIe bus beyond what CPU decompression would. However, this increase in effective bandwidth is not reflected in the data that is being read from disk.

When reading the performance charts below, it is important to note that PCIe SSDs reduce portal transition times by roughly 2-3 seconds compared to SATA SSDs. On SATA drives, Ratchet therefore spends more time within each portal sequence. As a result, average performance and 1% lows can generally be slightly higher on SATA SSDs, since there is less active rendering during these extended transition periods. However, there is one exception.

While the SATA SSD has an advantage at 1440p and 1080p due to the extended time spent inside of each portal, the 1% lows take a hit at 4K. The SATA drive does not handle the first portal transition in the sequence particularly well at 4K, and so the PCIe drives have about a 38% advantage in 1% lows.

This portal sequence is the most storage-intensive part of the game. In normal gameplay, the game uses significantly less bandwidth, and so there is no difference in performance at all between the SATA SSD and the PCIe SSDs.

Spider-Man 2

Spider-Man 2

Spider-Man 2 was the second game to use GPU decompression of assets on PC. This is the third DirectStorage title we have tested so far.

Since the game uses GPU decompression, the effective bandwidth after decompression will be considerably higher than what we see here, but none of the SSDs we tested are being stressed in this game.

Performance is virtually identical on all drives in this game.

Starfield

Starfield

Starfield is an open world game that constantly reads data from disk. It also frequently hits 100% disk utilization as you traverse the world.

In previous games, 100% SSD utilization was accompanied by much higher peak bandwidth. Here, the lower peak bandwidth suggests the game uses a very low queue depth for its read operations. As you move through the open world, it continuously streams data from the SSD in frequent, small transfers, keeping average bandwidth very low.

Once again, we see no improvement in performance going from the SATA SSD to any of the faster PCIe NVME SSDs.

DOOM: The Dark Ages

DOOM: The Dark Ages

DOOM: The Dark Ages was the first game to list an NVMe SSD as a requirement. Not only is it listed as such on the PC system requirements, but Bethesda has a support page that reiterates that installing the game on an NVMe SSD is required.

The developers at idSoftware mentioned that the game uses a proprietary ultra-fast continuous sector streaming tech that virtually eliminates load times between levels.

The drives we tested are hardly being stressed in DOOM: The Dark Ages.

Given the disk utilization figures, it comes as no surprise that all of the tested drives deliver identical gameplay performance.

Although this article focuses on in-game performance rather than loading times, we also found little difference in load times when using a SATA SSD. Loading screens were not noticeably longer, and we observed no visual artifacts or streaming issues while playing on the slower drive.

Based on our testing, it is unclear why Bethesda lists an NVMe SSD as a requirement. A SATA SSD provided an indistinguishable gameplay experience, with no measurable performance drawbacks or observable streaming issues.

Dead Space Remake

Dead Space Remake

Dead Space Remake made headlines when it launched with abhorrent performance and stuttering issues. But how much of that was due to streaming from slower storage? Does a faster SSD help performance?

Although average disk utilization and bandwidth are relatively low, the game frequently reaches 100% utilization. Yet even during these periods, peak bandwidth remains well below what PCIe SSDs can achieve under high-queue-depth random workloads. This indicates the game is issuing read requests at a very low queue depth, leaving much of the SSD's available performance untapped. Additionally, the difference in average bandwidth between the SSDs is minimal.

All drives provide the same performance. There is nothing to be gained from a faster SSD in this title.

Cyberpunk 2077

Cyberpunk 2077

We included Cyberpunk 2077 in our testing because the game features a large and seamless open world, and it is one of the most popular games ever released.

The average bandwidth is by far the lowest we have seen in the games tested so far. This is not entirely surprising. Although the game was released in the same year as the PlayStation 5 and Xbox Series X|S, development began well before those consoles were released. As a cross-generation title, it also needed to run on the previous generation of consoles, which relied on mechanical hard drives. As a result, the game's streaming system was designed to perform well on significantly slower storage devices.

There is no noticeable difference in performance across the drives tested.

The Elder Scrolls IV: Oblivion Remastered

The Elder Scrolls IV: Oblivion Remastered

The Elder Scrolls IV: Oblivion Remastered modernizes the look of the original game with improved visuals. Built on Unreal Engine 5, it utilizes Nanite virtualized geometry, Lumen global illumination, and virtual shadow maps. This is an open-world game, and our first UE5 title is tested in this article.

None of the SSDs we tested reached 100% peak utilization and average utilization is extremely low, too.

As is the case with most of the games we have tested, the SATA SSD holds its own and there is no difference in performance across all of the drives.

007 First Light

007 First Light

007 First Light is a recently released game that uses DirectStorage. In a recent interview, the developers at IO Interactive revealed that the real-time level streaming system in the Glacier engine was redesigned for 007 First Light. The new system uses dynamic loading and unloading to allow for seamless transitions between sequences.

It’s clear that this new streaming system does not place a heavy burden on storage devices.

Given the I/O activity in this game, it is not surprising to see that all SSDs we tested perform the same.

F1 25

F1 25

We included F1 25 because it exhibits higher peak bandwidth and utilization than most games. The game’s high-speed driving requires rapid streaming of trackside assets, as the environment changes quickly and the player covers large distances in very short periods of time.

The game exhibits higher peak and average bandwidth than many of the games we have tested, but it is still not enough to truly stress a SATA SSD.

Once more, all of the drives show identical performance.

What About the Visuals?

What About the Visuals?

Many readers may be wondering if playing on a SATA drive resulted in any visual anomalies, such as slower loading textures, additional pop-in, or other visual artifacts. We did not notice any such issues in the 11 games we tested. There are some games that exhibit pop-in, but that is consistent across all storage devices. All of the drives we used in our testing provided an identical visual experience.

The Verdict

The Verdict

Out of all the games tested, only two showed measurable performance differences between SATA and PCIe SSDs: Forspoken and Ratchet & Clank: Rift Apart.

In Forspoken, moving from a SATA SSD to any PCIe SSD – regardless of generation – resulted in a significant improvement in 1% lows.

Ratchet & Clank: Rift Apart showed more nuanced results. During the portal sequences at 1440p and 1080p, average framerate and 1% lows were slightly higher on the SATA SSD, as slower loading increased the time spent inside the portals. Since the portals contain less on-screen complexity, framerates are naturally higher within them. However, at 4K resolution, PCIe SSDs delivered better 1% lows. Outside of the portal sequence in normal gameplay, there was no measurable performance difference between any of the drives.

Across all other titles, PCIe SSDs provided no meaningful performance advantage, and there were no observable differences between PCIe generations. In the vast majority of games tested, a SATA SSD delivers comparable performance.

MORE: Best SSDs

MORE: Best External SSDs

MORE: Best SSD for the Steam Deck

An Intel patent application published on August 6, spotted by Patentlyze, describes an orbital data center architecture that moves some of the computing used to operate massive satellite constellations off the ground and into space. The architecture proposes a two-tier satellite network in which a small number of more powerful satellites in higher orbits manage large constellations of relatively simple satellites in low-Earth orbit, handling much of the computing and constellation coordination normally performed by data centers and network operations centers on the ground.

The application, US 2026/0230175 A1, is a continuation of an earlier Intel filing that was granted as US 12,542,604 B2 in February.

Intel’s proposed architecture is a different proposition from the orbital AI data centers now being pursued by companies such as SpaceX and Google, which aim to move AI compute itself into low-Earth orbit. SpaceX’s planned AI1 satellite and Google’s Project Suncatcher both envision running large-scale computing workloads in space, with the resulting data beamed back to Earth over high-bandwidth optical links. Intel’s orbital data centers, on the other hand, are designed primarily to serve the satellite network itself, acting as higher-orbit compute and control hubs for the much larger constellations operating below them.

In large LEO constellations such as Starlink, Telesat Lightspeed, and Amazon’s Project Kuiper, thousands of satellites are constantly moving relative to one another and the Earth. While the satellites perform their individual tasks, the network itself still has to determine how traffic is routed, which satellites and links should communicate, how spectrum is allocated, and how the constellation responds to failures, interference, weather, and other changing conditions. Much of that network planning and control processing is traditionally handled by computers on the ground, with routing and operational instructions calculated at terrestrial network operations centers and then transmitted back up to the satellites.

Intel says that this constant dependency on terrestrial infrastructure delays time-sensitive decisions, increases reliance on ground stations, and makes management harder as constellations grow into the thousands of satellites. Intel’s solution is to move part of that control and compute layer into orbit. Its architecture places more powerful satellites — which contain much more compute and storage capability than the individual LEO satellites — in Medium Earth Orbit (MEO), Geosynchronous Earth Orbit (GEO), or highly elliptical orbits, where they can maintain a broader and more persistent view of the LEO constellation below and take over tasks such as routing, mission planning, scheduling and network coordination without continually sending those workloads back to Earth.

The proposed setup does not eliminate the need for ground stations. It just keeps satellite network control processing in space. Intel specifically describes moving mission planning and scheduling operations into orbit. The company also argues that offloading heavier network-management tasks to a smaller number of powerful satellites could allow operators to build simpler, cheaper LEO spacecraft. Under current architectures, individual satellites still have to actively participate in network-control functions, requiring additional onboard compute and communications hardware. There’s currently no indication that Intel is actively building the satellites.

To optimize how software runs on hardware, instruction latency analysis looks at the time it takes for low-level instructions to execute on a processor, either to optimize the architecture or to optimize applications to run on a particular architecture. One hardware researcher, Christopher Domas (@xoreaxeaxeax on GitHub), is taking a different approach with the CPU Deoptimization leaderboard, which looks not to make Assembly instructions run as fast as possible, but as slow as possible to measure the single instruction with the highest latency.

The winner here is fxrstor64, which took 62 seconds, or over 198 billion cycles, to complete. This instruction restores the state of registers used for SIMD calculations to a 512-byte memory location. To achieve the highest (slowest) score, Domas first used their own mmiotic tool to find a high-latency area in the internal PCIe fabric, then forced the CPU to load a 512-byte state from MMIO (Memory-Mapped I/O), essentially processing all of those 512 bytes as slow as possible. That took 74 billion cycles or just over 23 seconds to complete.

Then, they went further "by starving the fabric while the load is in flight." They did so by using a series of 4-byte reads from another high-latency MMIO register, overwhelming the CPU's PCIe root complex and forcing the state restore to queue behind frivolous read ops. The next step is to use the AMX instructions available in Intel's Sapphire Rapids for xrstore64. The state area is increased from 512 bytes to 8KB, which could cause the instruction to hang for more than 1 trillion cycles.

The x86 leaderboard is live on GitHub now, and it looks like Domas has ARM and RISC-V leaderboards planned, as well. There are a few rules for the runs. Domas says any setup is fine, so long as only the execution of a single instruction is scored. Interruptible instructions aren't allowed, nor is scoring emulated instructions run on the handler. All times are normalized based on the CPU's base clock, and the platforms were all run without hardware modifications.

We're dealing with assembly code here, so the ranking is less about the specific instruction and more about what you're doing with that instruction.

Domas primarily used two CPUs for testing: the Intel Core i7-8559U and AMD Ryzen 7 5800H (inside the Trigkey S5). For the rdmsr instruction, however, they used a VIA Eden chip, which was a series of embedded processors from the early 2000s. The rdmsr command is used to read a model-specific register, or MSR. According to Domas, VIA "uses an undocumented register at 0x133 that gives wildly high response time." That command took 202 microseconds or 161,602 cycles to execute.

This is not the developer's first foray into wild experimentation with low-level instructions. An earlier project, called movfuscator, is a C compiler that solely uses the mov (move) command.

An AI model has just created new viruses that never existed in nature. According to a New York Times report, researchers trained a genomic AI model to design complete DNA sequences for viruses, then chemically built the results and watched some come to life. Of 285 AI-generated viral genomes tested, 16 successfully assembled into functioning viruses capable of infecting bacteria and reproducing.

In the study, published in the journal Science, researchers at Stanford University and the Arc Institute trained genomic AI models called Evo on trillions of nucleotides — the individual building blocks of DNA found in every living organism — allowing it to learn statistical patterns in how biological DNA is arranged. Much as a language model learns which combinations of words tend to make sense, Evo learned which combinations of nucleotides tend to produce biologically meaningful sequences.

Scientists have been synthesizing viruses for decades, typically to study how they work and to test antiviral drugs and vaccines. In these projects, researchers typically use the genetic sequence of a virus that already exists to manufacture new ones. Basically, they use existing genomes — genetic sequence data that holds information on how proteins come together to form an organism — replicating and editing this data.

Every organism comprises nuclotides/DNA. The genome provides the instructions for exactly how these building blocks form to create that specific organism. We can copy the data and replicate the organism, aka cloning. However, because trillions of possible genomes exist, it has been statistically impossible for humans to study enough of them to know the formula for how nature puts these building blocks together to form an organism. The blocks were known, but the formula for arranging them in a way that created viable genes wasn't. Until now.

After training the AI model on over 9 trillion nucleotides spanning over 128,000 genetic sequences drawn from millions of animals, plants, microbes, and viruses, it was able to discover patterns and use those patterns to design new genes that could instruct cells to make specific proteins. Using the pattern, the scientists wanted to see if the AI could create a blueprint for a new, simple organism, such as a virus that contains only a few thousand building blocks. For context, humans contain over three billion.

The researchers then trained the Evo AI model on the 11 genes and 5,386 nucleotides of the Phi X-174 virus and about 15,000 of its closest relatives. The virus was chosen because it has been studied for over a hundred years and is known to attack only E. coli bacteria. After studying the virus’s genome, the model proposed 700,000 new potential genomes, which the scientists trimmed down to 285. The scientists synthesized potential organisms using these instructions, out of which 16 resulted in new viable viruses that have never existed in nature. Like their relative, Phi X-174, the viruses were observed attacking E. coli bacteria and reproducing. It's important to note that the new viruses are quite similar to the source. The building blocks are very much the same, with some mutations in how they are arranged.

The potential applications of the study can be far-reaching. The scientists demonstrated one of them via a resistance experiment. They first generated three E. coli strains that had evolved resistance to natural Phi X-174. The researchers then exposed those resistant bacteria to cocktails made from the AI-designed phages. Those phage populations evolved during the experiment and ultimately overcame resistance in all three bacterial strains.

The scientists say the new viruses were completely harmless to humans, noting that replicating the same results in pathogens known to affect humans would be a different ballgame. They were also careful not to train the AI on any data from organisms known to affect humans. However, the study inadvertently shows that highly dangerous applications are a possibility. Experts worry that such studies are way ahead of necessary guardrails and regulations. AI has also been known to fly off the rails autonomously. An OpenAI agent recently went rogue and hacked Hugging Face.

NASA ensured that all laptops destined for use in the International Space Station (ISS) had the same power connector. Engineer George Roush, who has a certain fondness for space tech, recalled details of this laptop modification earlier this week. Standardizing on a single ‘cannon plug’ meant that instead of requiring a multitude of different chargers, ISS visitors could plug into the onboard DC power outlets in a standard fashion. Roush took the photos shared below during a visit to the NASA Stennis visitor center.

Laptops destined to be used on the ISS would have their modem ports removed and replaced by the cannon plug you can see in the engineer’s photos. Of course, modem ports would be useless on the ISS, so that’s why they would be replaced with this modification. In space, every gram, every connector, and every failure mode matters, which makes this power standardization a smart move.

Sources indicate that Lockheed Martin was contracted to make these modifications to the ThinkPads and other machines that ended up being taken to space. Roush also discusses networking/comms on the ISS, which relied on the MIL-STD-1553 system, rather than Ethernet.

Laptops with built-in telephony modem ports certainly date this story. However, we must remember that the ISS was constructed in the late 1990s when modems were still essential for PCs. The first crews docked in 2000.

Original web sources are vanishing

When refreshing his knowledge about this laptops-in-space mod, the space enthusiast and engineer was struck by another realization, which he found troubling. The NASA Stennis visitor center had barely any information about the laptops or the cannon plug modification that was on display.

Turning to some online resources he remembers referring to previously, Roush noticed a lot of them had disappeared. “The old internet is dying, and with it, we risk losing a lot of useful history,” said the engineer. “I fear that the false belief of the 'eternal' internet will cause otherwise recorded history to be lost when the servers shut down.”

Though some resources may still be able to be unearthed via the Wayback Machine, it seems perilous that so much web history is archived by a charity (The Internet Archive). In May, we reported that the internet is getting harder to archive because the AI boom has caused a storage crisis, with both NAND and mechanical drives facing shortages. This means skyrocketing storage bills to support the activities of the Wayback Machine.

HP replaced all the ISS laptops in April 2026

The astronauts in the ISS aren’t still using old modded laptops like the one showcased in the NASA Stennis visitor center. The latest updates came in April this year, with over 100 HP Z Workstations packing a mix of Core Ultra 9 processors and RTX Pro Blackwell graphics sent into orbit.

HP talks about putting these PCs through two years of rigorous testing ahead of deployment. We don’t get much in the way of technical info from its release, but again, it was recognized that a single power standard was important. The computer maker supplies universal AC/DC power adapters with its latest crop of space laptops that are specially designed to function on both the ISS and Earth.

Valve's second-gen Steam Controller has been a major hit with the community thanks to the company's open approach to its hardware. Modders have already used the haptic motors inside to move it across a desk and charge automatically, while others have built their own open-source pucks. Now, a new project from Pixel1011, called "Steam Haptics Player," can turn these motors into a stereo speaker capable of playing music both wired and wirelessly.

For context, the Steam Controller doesn't have a speaker; there are two linear actuators under each trackpad that vibrate to provide haptic feedback. Previous projects have leveraged these components to make the controller vibrate in a certain way. Since vibrations are the basis of sound, Valve itself built a secret easter egg into the controller that randomly makes it do a Wilhelm scream when dropped from certain heights.

Pixel 1011 discovered that the haptic motors inside can be targeted with low-level HID (Human Interface Device) feature commands using HIDAPI. As such, Steam Haptics Player takes an audio file, downsamples it through FFmpeg, splits it into chunks of a PCM stream, and sends those packets directly to the controller, bypassing high-level software like Steam Input. The controller's firmware then feeds this raw data to the motors, at which point they vibrate to produce sound.

Because of the DIY nature of this process, bandwidth is an issue, especially wirelessly, since the puck isn't designed to carry audio data over its radio signal. This means the audio has to be compressed, and there's a limit to the overall quality the motors can transmit — 8-bit 8kHz (μ-law) over the puck and 16-bit 8kHz PCM over a wire.

Still, the 16-bit audio is enough to produce a surprisingly loud and balanced result that can deceive you as if it's coming from a speaker. The video at the start features Pixel1011 playing the song "Want You Gone" from the end credits of Portal 2, and you can hear the quality for yourself. There's a genuine low-end that you would never expect from a jerry-rigged setup like this, even if it's sitting on top of a towel on a flat surface to give it its best chance.

The wireless audio quality is rough in comparison; it crackles and pops and even cuts off in between because of the bandwidth limitations. Pixel1011 is using 8-bit mu-law companding here, which actually offers dynamic range similar to 14-bit audio while keeping the byte footprint of an 8-bit stream. Even through this compression, the puck is struggling, but the fact that this is even possible is still very impressive, and perhaps there's room for improvement down the line.

You can try Steam Haptics Player yourself on your own second-gen Steam Controller with pretty much any audio file in any format. Just download the archive from the GitHub repo, install FFmpeg (for transcoding, if you don't have it already), and run the program via a PowerShell window in the original directory. Detailed steps are on the linked GitHub page, but we don't know how this affects the controller's battery life. Just don't go all out with your Spotify playlist on these haptic motors.

Software performance expert, scientist, and GitHub developer Daniel Lemire compared historical computer memory prices and showed that decades of falling prices were undone in just a matter of months because of the massive demand for HBM. According to his X post, this event is a historical anomaly, and he cannot think of an instance where technology hardware prices reverted to levels from decades before.

“RAM on a per unit basis is about as expensive as it was in 2007,” Lemire said. He also added, “I do not think anyone can predict what will happen, but I am guessing that we either find a way to build AI systems without so much memory, or we find really clever ways to make much more memory much faster.”

We also checked third-party data compiled and maintained by David Shim for the Stanford DAM Project. The latest price for DDR5 memory sits between $13.28 and $11.41 per GB — the last time RAM hit this price was in 2008, when the price for DDR2 hovered by $15 and 11.00 per GB. These are based on nominal USD values, though. If we take 2024 inflation into account, the current amount for DD5 is between $12.74 and $10.94 per GB, which was last achieved in 2011, when DDR3 prices were at $11.85 per GB.

Technological advances, especially in manufacturing and miniaturization, meant that tech became cheaper over time. This has been true ever since the first general-purpose digital computer was created 80 years ago. ENIAC cost the government $400,000 in 1946, meaning it would probably cost around $6,850,297.44 today. However, its computing power equaled roughly 5,000 additions per second — by comparison, one of the cheapest smartphones you can get today, the Moto G Play, is on sale for $99.99. This smartphone is powered by a Snapdragon 680 processor with AI performance of 3.3 TOPS — that is 3.3 trillion operations per second. We can’t really compare the ENIAC to a modern processor, but it at least shows us the scale by which technology has advanced, and prices have dropped.

The massive increase in memory prices in recent times wasn’t caused by technological regression or a bottleneck in supply — instead, it’s the massive demand for HBM driven by the AI race. “The memory output is increasing by around 20% per year. Now, normally, that would be fantastically fast and amazing for any large, mature industry, but ask yourself, ‘Is the demand increasing by 20% a year?” Elon Musk said during the last SpaceX earnings call. “No, the demand is increasing by 200% a year, maybe higher.”

Other experts and industry leaders agree with Musk. The chairman of the SK Group, the holding company that owns memory chip manufacturer SK hynix, said that RAM prices are “abnormally high,” and that the industry must take steps to increase supply and lower prices. Even RAM that uses chips from Chinese manufacturer CXMT, which some people look to as a cheaper alternative, generally tracked the prices of modules that use chips from the big three (Micron, Samsung, and SK hynix).

Aside from computers, other industries are already being affected by the shortage, including graphics cards, smartphones, gaming consoles, and even automobiles. We do not know how long the current memory crisis will last, but the situation is unsustainable and “something has to give.” The scary thing is that nobody knows what would happen if and when that time comes.

Amazon Web Services is cracking down on internal use of EC2 instances among its engineers. In May, the company reportedly met with engineers and told them to reduce CPU waste to ensure AWS has enough CPU capacity to meet customer demand, The Information reports. The message comes as demand for CPUs in the data center has hit a fever pitch, with the traditional eight-to-one or four-to-one ratio of GPUs to CPUs moving closer to parity.

EC2 instances make up a large chunk of the modern internet, and they're used in private deployments, as well. Traditionally, AWS engineers have been able to spin up their own instances for development, leveraging the relatively low CPU utilization required for web infrastructure to use more virtual machines. Now, engineers say they're waiting days to get access when they previously could get access within hours. One engineer told The Information that they've never had to wait this long for an instance, even after several years of working at Amazon.

Amazon deploys several different types of CPUs in EC2 instances, including AMD and Intel options and its relatively new Graviton5 chip. Graviton5 is Amazon's most powerful CPU to date, and it uses an Arm-based architecture along the lines of Nvidia's Vera CPU and Arm's own AGI.

The increased demand for CPU comes on the back of AI agents, a new paradigm in productivity that even companies as large as Amazon are struggling to reckon with. Last month, for instance, a coding agent blew through $1.8 million in token costs at Amazon, surpassing a development budget by 860%.

Much of the AI infrastructure currently in place is designed around inference, a workload that's accelerated by GPUs. With the four-to-one ratio, the CPU served as a way to keep the GPUs fed, and nothing more. However, agentic workloads are much more complex. They often involve tool calls that run on the CPU, as well as more complex orchestration of inference on GPUs. This is what has brought CPUs center stage in the agentic era. Intel, AMD, and others have echoed what we've heard from memory and storage companies over the past several months: the demand is so high for CPUs that most companies will take whatever they can get.

The major players are capitalizing on that demand. AMD just recently unveiled its portfolio of Zen 6 'Venice' CPUs for the data center, marking the first time AMD has launched a new architecture in the data center before the client market in decades. Nvidia has also pivoted its messaging away from accelerators and toward its new Vera CPU, vying to stake its claim in an expanding market of agentic AI infrastructure.

Although Amazon is wrestling with CPU capacity across its internal engineers and external customers, The Information reports that shortages are largely a problem for spot instances. A consultant told the outlet that contracted capacity hasn't experienced any shortages.

Following publication, an Amazon Web Services spokesperson reached out with the following statement:

"Demand for AWS services, including EC2, is incredibly strong and growing. Even with this heavy demand, we continue to satisfy the overwhelming majority of compute needs for both our internal and external customers. We work closely with internal teams to meet their compute needs while ensuring they use EC2 resources as efficiently as possible, such as reclaiming idle instances, right-sizing, and scaling up and down as needs change – just as we’ve always done. These efficiencies help manage capacity for internal and external customers alike, and any suggestions that these long-standing efforts reflect new capacity constraints is simply wrong.

This premise is sensationalized. Frugality is in our DNA since Day 1. We have always encouraged our teams to operate efficiently. We also share best practices for how customers can optimize resources to external customers. This isn’t a new directive, and encouraging efficient use of resources isn’t unique to Amazon."

We pressed Amazon on if engineers had been given deadlines to reduce their compute usage, as The Information originally reported. An AWS spokesperson responded: "This narrative on EC2 is inaccurate. As part of our normal business operations, and backed by our leadership principle of frugality, we are always driving efficiency across our resources to ensure teams are optimizing capacity."

The Salem City Council convened a public meeting last July 27, where it discussed the proposed $5.1 billion Oakline at Mill Creek data center by Verrus. As the people converged on the meeting site, a TikTok user spotted a guillotine towed by a bike outside of the venue, signaling the people’s discontent with the project.

According to NewsData, the consultation lasted for more than six hours as several members of the community voiced their opposition towards it. The developer’s representatives were initially on-site to hear the comments, but had to leave midway through as they were confronted by some protesters.

“During a break in the proceedings, a group of opposition members cornered Verrus representatives and shouted insulting remarks, to the point where it no longer felt safe to stay,” Verrus spokesperson JP Newmann said, according to the publication. “We appreciate the support of local law enforcement in helping de-escalate the situation and escorting Verrus representatives from the site safely.”

One week after the meeting, Salem announced a data center moratorium that would halt all data center developments in the city for one year as it considers zoning laws to accommodate developments such as this. This adds Salem to the long list of jurisdictions that have passed similar moratoriums, including Seattle, Washington, and the State of New York.

Unfortunately, the Verrus project will likely not be affected by this delay as it filed its application three days before the announcement. Despite this, Governor Tina Kotek (D) announced that Oregon will terminate a deal to sell a 32-acre plot of land that forms part of the three plots of land on which the project will sit. While it still leaves Verrus with approximately 43 acres, it means that it would have to go back to the drawing board to change its plans. This is similar to what Nashville did when it used its eminent domain power to deny the land that another developer wanted to use to build a data center near the historic Nashville Zoo.

But because Verrus submitted its application before the moratorium announcement, the city is still processing its proposed data center project. That does not mean that it will get an immediate green light, though. It still has to go through an extensive permitting process, which includes technical studies, permit conditions, and enforceable agreements, among others.

Oregon is one of the first states to enact a law that ensures large consumers, like data centers, pay for the majority of the grid upgrades needed to serve their needs. This allowed Portland General Electric, the state’s largest power provider, to increase the bills of users who used more than 20 MW by 30%, while also affecting a 1.3% cut in residential electricity costs. But despite this protection that stops ratepayers from price shocks, many people are still against data center projects in their towns and cities.

In case you've not subscribed to Tom's Hardware Premium yet, here's what you've been missing out on in this week's selection of articles.

First of all, we've fully redesigned Bench, our browsable database of hardware benchmarks. Using the same data that informs our massive hierarchy pages, Bench 2.0 allows you to granularly search and compare benchmarks between products. While you get an aggregated set of data in our reviews and hierarchy lists, Bench allows you much finer detail in seeing every single benchmark we run, whenever we review a new product.

Bench is exclusively available for Tom's Hardware Premium subscribers.

Earlier this week, we published a roadmap, focusing on the foundries that are making co-packaged optics (CPO) a reality. As the demands of data centers increase, so too do the demands on scale-up and scale-out connectivity. With AI accelerators now demanding extremely fast signalling speeds, CPO puts optical engines next to the processor or ASIC in order to shorten electrical paths, increasing data rates, shortening the electrical path, and consuming less power.

We dive deep into TSMC's COUPE roadmaps, which extend to 2030, and to a blistering lane speed of 400 Gb/s, Intel's CPUs with OCI chiplets, Samsung Foundry's path to building a CPO turnkey, and GlobalFoundries' vendor-agnostic OCI-MSA CPO platform.

• Co-Packaged Optics (CPO) foundry roadmaps — breaking down TSMC, Intel, Samsung, and GlobalFoundries' approach to next-generation scale-up connectivity

That's not all we had on the roadmaps front this week, as we've taken a deeper look at China's chipmaking tool roadmaps, hot on the heels of the country announcing it will build DUV immersion lithography machines capable of producing 7nm chips by 2030. The more difficult equation is building a domestic EUV machine, which demands a robust supply chain of the right tooling and equipment, with over 3,000 workers reported to be working on what's now dubbed China's "Manhattan Project."

The country still relies on ASML's DUV lithography machines for chipmaking, as modern systems are subject to export controls, which is why Huawei, the creator of the Ascend series of AI accelerators, is keenly interested in getting the project off the ground.

• Chinese chipmaking tool roadmaps examined — Beijing's nascent lithography tools target DUV production at five machines a year, and an EUV prototype with no chips

But China has more on its mind than merely producing the machines; it also wants to spin up its DRAM output via CXMT, which is planning to expand to a brand-new fab to boost memory output. This comes following consumer brands and computer makers like Dell, Lenovo, and more. But it's no simple task, as we explore how the MATCH Act could impact CXMT's ability to produce DRAM modules on more advanced nodes.

There are also concerns around expanding wafer capacities and procuring the tools required for such an expansion, which targets 30% of market share by 2030.

• China's CXMT targets 30% DRAM memory market share by 2030 with sixth mega-fab — future plans bottlenecked by access to advanced chipmaking tools

On the memory front, Samsung has also debuted three brand-new memory technologies, zHBM, zNAND-O, and BV-NAND, which each target different markets and are all built upon advanced wafer bonding techniques. zHBM is expected to enable 8x the performance of HBM5, while zNAND-O,. which would allow the company to put four or eight stacks of NAND devices on top of logic dies. Meanwhile, BV-NAND represents Samsung's 10th-generation V-NAND and is the closest of the three announcements to commercialization.

You can read this lengthy technical breakdown of Samsung's announcements for free, even if you're not a Tom's Hardware Premium subscriber.

• Samsung debuts three next-generation memory technologies for AI data centers — zHBM, zNAND-O, and BV-NAND all rely on advanced wafer bonding technologies

Is the AI tokenmaxxing party back on? With the release of the modestly priced Kimi K3 and DeepSeek V4 Flash-0731 over recent weeks, the emergence of competitive, cheap Chinese AI models has now forced companies like OpenAI, which launched GPT 5.6 Luna back in June, to cut prices by a staggering 80% in the wake of the competitive open-weight Chinese models. And OpenAI isn't the only company trying to offer more value to users. Anthropic replaced Opus 4.8 with 5.0, and Google also came out with the affordable Gemini 3.6 Flash and Gemini 3.5 Flash-Lite models, too.

So, where does that leave the AI industry? Western frontier labs must keep pace with the competition on offer overseas, and the intensifying competition is slimming down margins for what we assume to be already subsidized token costs.

• AI companies are now racing to the bottom — crashing token prices and competitive models push companies to cut costs

Within the world of artificial intelligence, other stresses are looming for ChatGPT-maker OpenAI. The company is heading to court with tech titan Apple over claims that OpenAI stole trade secrets from Apple. The claim comes amidst OpenAI's development of dedicated AI hardware. OpenAI says it didn't want Apple's trade secrets, and that the Cupertino company was simply getting things wrong. Apple, however, alleges that 13 of its former employees passed along the aforementioned trade secrets, including documents sharing details of unannounced products and screenshots of sensitive materials.

We explore how the lawsuit is developing in this rather unusual offshoot, which centers around products that have yet to see the light of day.

• Apple is taking OpenAI to court over alleged theft of trade secrets — ChatGPT maker suggests it doesn't want Cupertino's knowledge anyway

That wraps up most of the major pieces of content that we published this week on Tom's Hardware Premium, with the bonus of being able to access our deep dive into Samsung's new announcements without a subscription.

For all of the other articles in this list, and to use our Bench tool, you can sign up to Tom's Hardware Premium today to read more. We'll be back next week with another roundup to keep you up to date.

If you're a content creator or a gamer that needs to have a whopping 32GB of VRAM to power either your software applications or run your favorite games on extreme fidelity settings, then you're looking to own an RTX 5090 graphics card. Unfortunately, these GPUs do not come cheap, with the current lowest price on an available-to-buy 5090 sitting at an eye-watering $4,399.99. However, with this amazing Dell deal on an Alienware Area-51 gaming rig for $4,779.99, you can bag a whole gaming PC that includes the RTX 5090 for just $380 more than purchasing the standalone graphics card.

● Check out this deal at Dell

The Alienware Area-51 gaming PC in question comes packed with the aforementioned Nvidia GeForce RTX 5090 graphics card, but also includes the 20-core Intel Core Ultra 7 265K processor, 32GB of high-speed DDR5 7200 MT/s memory, and a 1TB SSD for the operating system and game storage.

All of the above components are packed into Dell's own Alienware case with a tempered glass side panel to show off the internals of the system. A 240mm liquid AIO cools the Intel Core Ultra 7 265K, and a beefy 1,500W 80 Plus Platinum-certified power supply has all the juice needed to run the RTX 5090. The Area-51 does away with Dell's old proprietary parts ecosystem and lets you add upgrades more easily, so you're able to add more memory or a bigger SSD down the line.

The RTX 5090 is the most powerful consumer GPU on the market right now. It comes with 21,760 CUDA cores, over double the RTX 5080's 10,752 count, along with a massive 32GB of GDDR7 VRAM, using a 512-bit memory bus with memory bandwidth up to 1,792 GB/s. In our benchmarking results from our review of the RTX 5090 we can see that this super powerful GPU dominates the competition in all performance metrics.

The Alienware Area-51 gaming PC in this deal has the ability to play any game at extreme frame rates and power high-refresh-rate gaming monitors, even at 4K resolutions. It's the best GPU to experience ray-traced graphics without severely compromising in-game graphics settings, and with the extreme pricing of the RTX 5090 at the moment, it's one of the best value ways to purchase an RTX 5090-powered gaming PC if you can afford one.

Yes, it should be $2000 cheaper, but thanks to the AI-boom ruining PC component prices for us enthusiasts, this is one of the best deals you will find on an RTX 5090 prebuilt gaming PC right now.

If you're looking for more savings, check out our Best PC Hardware deals for a range of products, or dive deeper into our specialized SSD and Storage Deals, Hard Drive Deals, Gaming Monitor Deals, Graphics Card Deals, Gaming Chair, Best Wi-Fi Routers, Best Motherboard, or CPU Deals pages.

Tom’s Hardware is renowned for the benchmark tests and performance data we collect on every CPU, GPU, SSD, laptop, and more that runs through our labs — hundreds of products every year, thousands of tests, tens of thousands of data points. Why? Because details matter, and we know that Tom’s Hardware readers value that data.

So for the launch of Tom’s Hardware Premium, we built a new tool called Bench, which offered one-stop access to all of that data. But the tool? We confess, it was merely okay. All those tests were hard to sort through as one giant list, the data was fairly static, and it was hard for the user to put a given score in context to their own rig.

Today, we’re pleased to introduce you to Bench 2.0, a ground-up rethink of our benchmarking tool that makes it the most powerful, most intuitive benchmark browser in the business. Bench 2.0 has tons of new features designed to make the data easier to understand, easier to sort through, and far more usable. Best of all, we've spent dozens of hours re-testing our entire slate of CPUs and GPUs, too.

The most obvious change is the introduction of a detailed hierarchy for tests, which were presented as one long (very long) list in Bench 1.0. Instead, you can pick out just gaming tests or just synthetic benchmarks, or zero in on a given app, whether it’s for content creation or encoding or whatever. We’ve also added detailed product pages, with specs, benchmark test results, and pricing information.

And this is cool: Pricing information culled from our proprietary Hawk tool allowed us to create a neat new visualization mode: Bang for the Buck charts, which plot performance against cost, a crucial data point in today’s expensive market. In both bar charts and these new graphics, mouse over the graphics to interact with them. You can click a given data point to set it as a baseline for comparison, to at a glance see how everything rates against, say, a GeForce RTX 4070 Ti.

Speaking of, did you buy an RTX 4070 Ti? Flag it as your own! You can store your products and mark them as actively in use or in storage. Products you own will show up as blue dots or bars in the charts to showcase your performance against the hottest new gear. You can also pick products you’d like to buy, creating a watch list of what’s on your radar. It’s all tucked away in the 'My Rig' section, accessible from the little suitcase icon in the top right corner.

Finally, we’ve built a better way to share some of this data: The 'Share Results' drop-down has options to let you download an image of whatever chart you’re looking at or create an embeddable, fully interactive widget.

Oh, and don’t miss the settings options at the top right of the page as well, just below My Rig: We’ve built light and dark modes, added sound effects (and the ability to turn them off), and given you a peek into our plans for Version 3 of Bench, which will be expanded to include our prebuilt gaming PC test results, and more. Have fun with the Easter Eggs as well. As befits a tool running on Tom’s Hardware, this one does, in fact, run Doom.

Thoughts? Suggestions? Complaints? We’re all ears. Leave a comment and let us know what you think.

Bench 2.0 is live now, exclusively for Tom's Hardware Premium subscribers.

If you’re looking for the absolute best performance in a gaming laptop, then you’ll likely want to consider a desktop replacement. If you’re in the market, the HyperX Omen Max 16 makes a strong case with a 16-inch form factor and a thick chassis that accommodates some beefy hardware.

Our review unit carries a sticker price of $4,699, packs a Core Ultra 9 290HX Plus processor, and includes an RTX 5090 GPU. It also tips the scales at 6.56 pounds, making it a more suitable companion for your desk than for lugging around while traveling.

Despite its imposing dimensions and beefy hardware, the Omen Max 16 doesn’t quite pass muster in performance against its slimmer competitors among the best gaming laptops.

Design of the HyperX Omen Max 16

“Solid, solid as a rock,” is the phrase that comes to mind when describing the Omen Max 16. The laptop feels like it was forged from a solid piece of granite, but in actuality, it’s aluminum. It’s also thick, measuring 1.13 inches at its thickest point (at the rear).

To my eyes, the Omen Max 16 is attractively designed with no outrageous styling gimmicks. The aluminum is coated in a matte black finish, and design flourishes (like the O16 printed vertically along the deck and the HyperX logo on the lid) are subdued and don't draw significant attention. There’s also a laser-etched grille that runs nearly the width of the chassis that sits above the keyboard.

The only thing that really clues you into the gaming nature of the laptop (other than its sheer size) is the per-key RGB lighting for the keyboard, and a single LED strip on the bottom front edge of the chassis that you can configure with the Omen software.

When it comes to external ports, the Omen Max 16 has a USB-A (10 Gbps) port, two Thunderbolt 4 ports, and a 3.5 mm audio jack on the left. On the right side of the chassis, you'll find another USB-A port. The back panel of the laptop is home to an HDMI 2.1 port, a 2.5 GbE port, and a proprietary power port (a 460W GaN power adapter is included).

The Omen Max 16 measures 14.04 x 10.59 x 1.03 inches and weighs 6.56 pounds. For comparison's sake, the Razer Blade 16 is 13.98 x 9.86 x 0.69 inches and weighs 4.71 pounds. The MSI Raider 16 HX comes in at 14.29 x 10.62 x 1.14 inches and weighs 5.73 pounds. The Alienware 16 Area-51 weighs 7.49 pounds and measures 14.37 x 11.41 x 1.12 inches.

HyperX Omen Max 16 Specifications

Gaming and Graphics on the HyperX Omen Max 16

The Omen Max 16 is equipped with an Intel Core Ultra 9 290HX Plus processor, 32GB of RAM, and an RTX 5090 GPU (175W TGP). Given these specs and a sizable chassis to aid cooling (300W total system power), I was expecting strong numbers from the Omen 16 Max. Instead, the laptop was relegated to the back of the pack, except for one game benchmark: Far Cry 6.

My game of choice these days is Battlefield 6, running at native resolution in Overkill detail mode, with DLSS Quality and frame generation enabled. Using these settings, I was able to average between 100 and 110 frames per second.

For the competitive set, we have the Blade 16 (Core Ultra 9 386H at 25W, RTX 5090 at 175W TGP), MSI Raider 16 Max HX (Core Ultra 9 290HX Plus at 55W, RTX 5090 at 175W TGP), and the Alienware 16 Area-51 (Core Ultra 9 290HX Plus at 55W, RTX 5080 at 175W TGP). It should be noted that the Omen Max 16 uses slower DDR5-6400, whereas systems like the Blade 16 use faster LPDDR5X-9600.

We started our benchmarking spree with Shadow of the Tomb Raider (Highest setting), and here the Omen Max 16 achieved 171 FPS at 1080p, putting it in last place behind the Razer Blade 16 (182 FPS). At 1600p, it landed in a comfortable third place (114 FPS) behind the Alienware 16 Area-51.

Cyberpunk 2077 (Ray Tracing Ultra settings) wasn't a strong showing for the Omen Max 16, either. It brought up last place at 1080p (59 FPS) and 1600p (37 FPS).

Far Cry 6 (Ultra settings) was the lone bright spot for the Omen Max 16, as it garnered 144 FPS at 1080p and 125 FPS at 1600p, easily putting it in first place. Both of these results far outpaced the rest of the field, with the next closest competitor being the MSI Raider 16 Max HX at 119 FPS and 114 FPS, respectively.

Moving to the Red Dead Redemption 2 (Medium settings) benchmark, the Omen Max 16 again wound up in last place, hitting 93 FPS at 1080p and 65 FPS at 1600p. However, in this instance, the margin between the Omen Max 16 and The Blade 16 at 1080p was just 1 FPS.

The Omen Max 16 again had trouble during the Borderlands 3 (Badass settings) benchmark, falling well off the pace set by its three competitors. The laptop hit 146 FPS at 1080p, nearly 20 FPS off the pace of the third-place Alienware 16 Area-51. At 1600p, it mustered just 113 FPS.

Metro Exodus is our go-to benchmark for laptop stress testing. Our review unit averaged 141.62 FPS at 1080p across 15 loops using the RTX benchmark preset. The CPU performance cores averaged 4.91 GHz, and the efficiency cores averaged 3.95 GHz. The RTX 5090 GPU ran at 1.81 GHz.

Productivity Performance on HyperX Omen Max 16

With its Core Ultra 9 290HX Plus processor, 32GB of RAM, and 2TB PCIe 5.0 SSD, the Omen Max 16 fared better in the productivity tests than it did with gaming.

On Geekbench 6, the Omen Max 16 with the Core Ultra 9 290HX was able to muscle past the Blade 16 with its Core Ultra 9 386H on both the single-core (3,182) and multi-core (17,989) benchmarks. However, the Alienware 16 Area-51 and Raider 16 Max HX, which use the same Core Ultra 9 290HX processor, achieved much higher multi-core numbers (each surpassed 20,000).

The PCIe 5.0 SSD used in the Omen Max 16 performed quite well in our 25GB file transfer test, coming in second place at 2,439.22 MBps. The Alienware 16 Area-51 was a few ticks ahead at 2,738.90 MBps.

The soft performance of the Omen Max 16 continued with our Handbrake test, which involves transcoding a 4K video file to 1080p. Here the laptop performed the task in 2 minutes and 59 seconds. However, at least one other laptop with the same Core Ultra 9 290HX did the deed in less than two minutes.

Display on the HyperX Omen Max 16

The Omen Max 16 uses a 16-inch 2560 x 1600 OLED panel with a 240 Hz refresh rate. As is the case with most OLED displays that we test, this one had a glossy finish.

Instrumented testing on the OLED panel showed it excelling against the competition, covering 131.2 percent of the DCI-P3 color space and 185.2 percent of sRGB. Even better, the Omen Max 16 nudged past the Raider 16 Max HX to also take the crown for panel brightness (464.6 nits versus 455.6 nits).

OLED panels rarely disappoint, and I was more than impressed with the overall quality and brightness. The brightness was high enough that I was able to game while on my back porch during a partly cloudy day with minimal interference from reflections on the glossy panel.

I also viewed the first Avengers: Doomsday trailer using the display, and besides getting a tiny bit emotional while watching it, was delighted with the color reproduction, particularly in the short battle between Gambit and Shang-Chi, with flashes of purple playing cards and orange rings.

Keyboard and Touchpad on the HyperX Omen Max 16

The Omen Max 16 features a full-size RGB backlit keyboard (with per-key lighting) and NKRO anti-ghosting technology. RGB effects are customizable using HP's Omen app or directly from within Windows 11.

The keys felt quite springy when typing, and I found it incredibly comfortable to rest my fingers and click away. And while some keyboards audibly make their presence known as you type at a furious pace, the key actuation on the Omen Max 16 was whisper-quiet.

Turning my attention to the Keyhero.com typing test, I plugged away at 95 words per minute with 97% accuracy, which is definitely on the high side for me (I’m usually in the mid-80s realm).

Given the Omen Max 16 price tag of over $4,500, I was disappointed that HP didn’t go with a haptic touchpad. Instead, we get a top-hinged trackpad that is just OK. It gets the job done, but for this much money, I’d prefer to be able to click anywhere on the touchpad surface without hitting a dead zone. At least while gaming, you probably use a mouse or controller.

Audio on the HyperX Omen Max 16

Audio quality is another area where I think HP missed the mark. To beat a dead horse, but this is a laptop that retails for $4,699, and it only comes with stereo speakers. Would it have killed HP to add another pair of speakers in this tank of a chassis?

While the speakers had good overall volume levels, I felt the sound lacked some depth compared to similarly priced laptops with four (Raider 16 Max HX) or even six speakers (Blade 16). While I was more than happy listening to Keane’s “Fly to Me” from the album Hopes and Fears, I was disappointed in the sound while playing Battlefield 6.

When it came to gunfire, vehicle sounds, and background explosions while gaming, I wasn't impressed. I prefer throwing on a pair of the best gaming headsets, but there’s still no excuse to drop the ball at this price point.

Upgradeability of the HyperX Omen Max 16

Six screws need to be extracted to remove the bottom aluminum panel. While that was easy enough, it’s only half the battle. All of the serviceable components (save for the readily accessible battery) are underneath an extensive metal shield. That shield is held in place by roughly a dozen screws, making upgrades a bit tedious.

Once the shield is removed, you’ll find access to the M.2 slot for the Wi-Fi/Bluetooth module, two SO-DIMM slots (both of which were filled on our review unit), and an additional M.2 slot for a second SSD. Yet another metal shield covers the first M.2 SSD slot, although I didn’t go through the process of removing another dozen screws to uncover it.

Battery Life on the HyperX Omen Max 16

Despite its considerable girth, the Omen Max 16 is equipped with just an 83 WHr battery. For comparison, the thinner and lighter Blade 16 manages to squeeze in a 90 WHr battery.

The Omen Max 16 lasted 5 hours and 7 minutes on our battery test, which consists of web browsing, video streaming over Wi-Fi, and OpenGL testing with the screen brightness set at 150 nits. This was over three hours shorter than the Raider 16 Max HX, while the Blade 16 lasted over twice as long on a charge (12:46).

Heat on the HyperX Omen Max 16

During the Metro Exodus stress test, we measured the Omen Max 16's keyboard at 106 degrees Fahrenheit (F) between the G and H keys, while the touchpad was 84.9 F.

The hottest part of the laptop was near the rear center of the bottom of the chassis, which measured 114 F.

While running the Metro Exodus stress test, the CPU measured 87 degrees Celsius (C). Meanwhile, the RTX 5090 was cooler at 75 C.

Webcam on the HyperX Omen Max 16

Another area where it feels like HP skimped is with the webcam. Rather than going with a higher resolution unit, the Omen Max 16 features a bog-standard 1080p unit that looks about as good as one you’d find in a $1,000 entry-level gaming laptop.

The camera didn’t do particularly well with a nearby window where light was shining through, and there were a lot of noise artifacts in less-than-ideal light conditions. In addition, the camera didn’t really pick up fine details well, which is disappointing considering the price.

Software and Warranty on the HyperX Omen Max 16

Like many HP laptops, the HyperX Omen Max 16 comes loaded with additional software bloat that you can mostly ignore or simply delete. Scrolling down the list of “H” apps in the Start Menu, HP basically set up camp with no less than eight apps/shortcuts—everything from HP Documentation to HP Privacy Settings to HP TV+. In addition, you’ll find the usual third-party add-ons like McAfee and promotional links for Dropbox, Adobe, and Booking.com.

The only really useful preinstalled app is the Omen Gaming Hub. It gives you control over power modes, lets you see CPU/GPU/RAM utilization, and provides plenty of configuration options for the onboard RGB lighting.

The Omen Max 16 comes with a one-year manufacturer warranty.

HyperX Omen Max 16 Configurations

Our Omen Max 16 came configured with an Intel Core Ultra 9 290HX Plus processor, 32GB of DDR5-6400 memory, a 2TB PCIe 5.0 SSD, RTX 5090 GPU, and a 16-inch 1600p 240 Hz OLED display. The MSRP on our review unit was a staggering $4,699.99.

There are also several less expensive options, with the cheapest being a $2,699 SKU with a Core Ultra 7 270HX processor, 32GB of RAM, a 1TB SSD, and an RTX 5080. For $3,099, you can swap the Core Ultra 7 270HX for an AMD Ryzen AI 9 HX 475 and double the SSD capacity to 2TB.

Bottom Line

The HyperX Omen Max 16 represents a swing and a miss. When I first saw the enormity of the chassis, I was looking forward to some impressive gaming performance given the 55W processor and 175W RTX 5090 onboard. Instead, what I was left with was a $4,699 gaming laptop that lingered in last place for most of the gaming benchmarks, and often by a large margin. Even the Razer Blade 16 with its 25W processor managed to get the upper hand in most benchmarks.

But the gaming wasn’t the only disappointment. Given the price tag, I would have expected a better-equipped machine. Where’s the haptic trackpad, and why do we only get two speakers? And on a minor annoyance angle, while I appreciate the two M.2 slots for storage and replaceable memory, why do I need to remove another dozen screws just to access those components?

In the end, the Omen Max 16 could have been so much more. Quite frankly, it would make sense to spend the extra $200 and grab a Razer Blade 16. Not only is it thinner and lighter (by two pounds), but it offers double the battery life.