News Posts matching #GPU

Return to Keyword Browsing

Thermal Grizzly Announces New TG x CENS LN2 Container

Every record is an important milestone in pushing the boundaries - until someone pushes them further. Ready to push your GPU to its limit? The TG x CENS LN2 Container was developed with experienced extreme overclocker Clemens "CENS" Nüssler to help overclockers push GPU performance beyond conventional limits using liquid nitrogen.

The special design of the inner structure was optimized to allow for both maximum heat dissipation and precise temperature control down to -196 °C. The modular mounting system for selected high-end AMD and NVIDIA GPUs provides optimized contact pressure and efficient heat transfer. The TG x CENS LN2 Container is built for your next extreme overclocking session.

NVIDIA GeForce RTX 5080 Reaches Nearly $2,000 Across U.S. Retailers

NVIDIA's GeForce RTX 5080 has seen a significant increase in its listed price over the past few weeks, with this high-end GPU now costing nearly $2,000. Due to price hikes throughout the supply chain and the GPU's 16 GB of GDDR7 memory, which is currently in short supply, the listing price has doubled its original MSRP. NVIDIA initially launched the GeForce RTX 5080 in early 2025 with an MSRP of $999. However, add-in card partners introduced custom solutions such as new PCBs, enhanced cooling, and other features that gradually increased the price. The pricing issue escalated when the memory supply became constrained, making the GDDR7 memory modules used in this GPU much more expensive.

Across U.S.-based retailers like Best Buy, Micro Center, Amazon, Newegg, B&H, and others, the listing price now starts at nearly $2,000. On B&H, older listings show better pricing but are marked as "Temporary out of stock." Only the in-stock listings are priced at $2,000 or more, indicating that new pricing adjustments have been made across all retailers. In just two months, the listings have jumped from $1,200 to $2,000, which is a dramatic increase in pricing.
Below are screenshots collected across U.S.-based retailers.

TSMC Could Charge Up to 20% More for CoWoS Packaging Amid AI Crunch

TSMC is reportedly considering raising prices for its CoWoS (Chip on Wafer on Substrate) advanced packaging by as much as 20%, according to supply chain sources cited by Taiwan's Commercial Times. Foundries and outsourced packaging firms are apparently weighing another round of 10% to 20% increases, as CoWoS demand has spread beyond AI GPUs to server CPUs and custom ASICs, leaving a growing backlog of unfilled orders. TSMC hasn't confirmed anything yet.

Testing is turning into a bottleneck of its own. Nomura estimates, via Seoul Economic Daily, that NVIDIA's Blackwell takes roughly four times longer than Hopper in final testing, while Rubin could need seven times as long. Packages keep growing too, with TSMC already mass producing CoWoS-L at 5.5 times the reticle limit. Tool lead times have reportedly stretched from three to six months to as long as eight months. TSMC, which usually negotiates for discounts or extra features, apparently placed this year's equipment orders at existing prices instead, which says a lot about how tight things are.

ASUS Announces TUF Gaming Radeon RX 9070 XT Onimusha Special Edition

ASUS today announced the TUF Gaming Radeon RX 9070 XT Onimusha Special Edition graphics card, arriving alongside Onimusha: Way of the Sword, which releases on September 4, 2026. More than a performance powerhouse, this card is a collectible piece of artistry—bundled with the game, slashing onto the scene, and ready for action.

A masterpiece of dark artistry
The TUF Gaming Radeon RX 9070 XT Onimusha Special Edition pays visual tribute to the world of Onimusha: Way of the Sword. A dark fantasy set in Edo-period Kyoto, the game invites players to explore the historic Japanese capital on the edge of a blade, battling underworld monstrosities with shadow and steel.

NVIDIA Allegedly Stops Accepting RTX 5090 Orders

Although NVIDIA has officially denied the discontinuation of the GeForce RTX 5090, more rumors are emerging that substantiate claims about the flagship gaming GPU being discontinued. This time, NVIDIA board partners are reporting that NVIDIA has officially stopped accepting orders for the 5090, according to a recent report by Der8auer on YouTube, and it appears as though it's not just a temporary freeze on GB202 supply, as has been the case prior, since pre-orders are also not available to buyers.
Unnamed AIB partner via Der8auer on YouTubeNo, NVIDIA did not inform us that the RTX 5090 is EOL. But currently we cannot purchase GB202 together with 32 GB GDDR7 as a complete set, and we also don't know if ordering will be possible in again in future. Similar supplier shortages already happened several times before. But in the past NVIDIA always allowed us to place backorders, which is currently not possible anymore.

New DLSS 5 Mod Borrows a VR Trick for Better Ultrawide Performance

A new DLSS 5 mod borrows a trick from VR headsets to make Neural Rendering less punishing on ultrawide monitors. FoveatedDLSS5, a ReShade add-on built by maximedgoulet, keeps neural rendering at full resolution in the middle of the screen and steps it down toward the edges, where peripheral vision picks up less detail. In a Reddit post, the modder says The Witcher 3 went from 49 FPS to about 80 FPS on an RTX 4090 at 7680x2160 with 2x multipass, a gain of over 60%.

With default settings on a 7680x2160 panel, Neural Rendering ends up processing about 71% of the original pixels. Smaller screens gain less, however. At 4K, the defaults leave 93% of the pixels, and at 3440x1440 the mod turns itself off, so 16:9 and 21:9 users should shrink the center zone. It's also a pre-release tested in one game, and it only works when neural rendering runs after upscaling, which rules out pairing it with the Neural Upstream trick. Videocardz points out a similar project, CheekyFoveatedDLSS, that has been out since early September and adds eye-tracked foveation for VR headsets.

PlayStation Handheld May Be Only Portable Console With RDNA 5

It has long been rumored that Sony is working on a PlayStation 6 handheld with a powerful AMD APU codenamed Canis, which has been alleged to feature a combination of Zen 6 CPU cores—potentially with all-new Zen 6 LP cores—and RDNA 5 tech for the iGPU. Now, according to Richard Leadbetter on episode 287 of the Digital Foundry Direct Patreon-exclusive podcast, Sony may be the only gaming handheld to feature this RDNA 5 iGPU. However, it may not be an exclusivity agreement in the traditional sense of the word, since it appears to be as a result of AMD's requirements to build custom silicon for hardware OEMs.

Rumors, albeit disputed rumors, have mentioned that part of the reason Xbox cancelled its own in-house handheld console is because AMD wanted a minimum commitment of 10 million unit sales before it would engineer a custom APU for a console. This may also be why Valve's Steam Deck 2 will seemingly use an off-the-shelf component instead of another custom silicon design. This leaves questions about what CPU and GPU combo will be in the upcoming Steam Deck 2 and Xbox's previously leaked handheld, which now seems like it will be another third-party device, like the ROG Xbox Ally series.

NVIDIA Cuts HBM4 Speed and NVLink Bandwidth Targets for "Vera Rubin"

NVIDIA has officially implemented changes across its "Vera Rubin" product line, reducing target specifications for memory and interconnect fabric. According to the updated product specification sheet, the first change involves HBM4 memory. Previously, NVIDIA used HBM4 with a speed of 11.7 Gbps per pin, which has now been reduced to 9.6 Gbps per pin. In a "Vera Rubin" NVL72 rack-scale solution, NVIDIA now offers a total memory bandwidth of 1,400 TB/s, down from the previously advertised 1,580 TB/s. While this issue does not stem directly from NVIDIA's suppliers, it indicates that the company's implementation of memory bus width and overall electrical signaling is insufficient to support higher speeds.

However, the changes are not limited to memory. NVIDIA has also reduced the previously advertised NVLink bandwidth per GPU, decreasing it from 1.8 TB/s to 1.5 TB/s. This results in a reduction from 3.6 TB/s to 3 TB/s in bidirectional NVLink GPU bandwidth. The underlying issue is related to signal integrity, as NVIDIA reportedly lowered the NVLink Nyquist frequency target from 56 GHz to 47 GHz. Lowering the Nyquist frequency reduces the electrical demands on the SerDes, cables, and materials, which can improve manufacturing yields and overall feasibility, even though it requires accepting a small reduction in peak interconnect bandwidth. Thus, NVIDIA chose to sacrifice some bandwidth to enhance system stability. Below is the new system specification in light mode, and old system specification in dark mode.

NVIDIA RTXPT 1.9 Path Tracer Gets Around 20% Faster, Adds Windows on Arm Support

NVIDIA has released version 1.9.0 of RTX Path Tracing (RTXPT), its open-source reference path tracer on GitHub, claiming roughly 20% better performance than the previous 1.8.1 release, with comparable or better image quality. RTXPT is a code sample for developers, meant as a starting point for building a path tracer or as a reference for NVIDIA's other ray tracing SDKs. Several changes add up to that uplift. The path tracer can now optionally trace only three quarters, half, or a quarter of the pixels each frame, and the built-in performance presets use this by default, along with lower bounce limits in the faster modes.

On DirectX 12, async compute lets the previous frame's DLSS Ray Reconstruction and post-processing overlap with the next frame's path tracing, and TLAS instance generation has moved to the GPU. Lighting got some attention too. NEE-AT light sampling now handles up to 2 million lights, up from 512K, and the SHaRC radiance cache is on by default in real-time mode, apparently feeding more indirect lighting into DLSS Ray Reconstruction at about the same cost. Version 1.9 also adds Windows on ARM64 support. The release notes don't say why, though it's probably no coincidence, given NVIDIA's own Arm-based RTX Spark chip for Windows PCs.

New Linux Tool Brings Resizable BAR to Intel Arc GPUs on Old Motherboards

Intel Arc cards have always leaned heavily on Resizable BAR, to the point where Intel once told Alchemist buyers without it to stick with other vendors, and TechPowerUp's B580 review listed it as required for good performance. As spotted by Videocardz, developer Rod Burlamaqui has now released arc-rebar, a Linux tool that turns it on for Arc GPUs on motherboards that never got a ReBAR option. It runs from the initramfs at every boot, before the GPU driver loads, and doesn't touch the BIOS, kernel, or Intel's drivers. That sets it apart from earlier workarounds like ReBarUEFI, which has to be added to the motherboard's firmware.

The fix targets the PCIe switch built into Arc cards, whose small upstream BAR blocks Linux from enlarging the GPU's memory window. arc-rebar writes the card's ReBAR register, then removes and rescans the root port. Intel engineers have proposed a kernel-level fix for this, though it hasn't been merged yet. On a Supermicro X9DRH-7TF with dual Xeon E5-2670 v2 CPUs and a 2015 BIOS, a B580 went from 256 MiB of CPU-visible VRAM to a 16 GiB window covering all 12 GB. Measured VRAM bandwidth jumped from 15 GB/s to 392 GB/s, and a 4K Vulkan blur that used to stall ran at 39 to 50 FPS.

NVIDIA Eyes Software Revenue With a $2,000 per GPU Yearly Figure From Huang

NVIDIA CEO Jensen Huang has put a rough price on the software side of the company's AI business. During a Q&A session at GTC Berlin last week, Huang said software licensing will become a growing part of NVIDIA's business, and that it costs about $2,000 per GPU per year to run all the NVIDIA software that powers agents in AI factories. The Register was first to report it, and Huang seems to have offered it as an example rather than a formal price.

The number is a bit confusing, since NVIDIA already sells its AI Enterprise suite on a per-GPU basis. The company's own licensing guide lists it at $4,500 per GPU for a one-year subscription with support, or $18,000 per GPU for five years. NVIDIA hasn't said whether Huang's $2,000 figure covers the same software, a narrower agent-focused bundle, or a future licensing model, so it probably shouldn't be read as a price cut. There's also no word on timing.

GeForce Unlocked: Modders Boost RTX 20, RTX 30, and RTX 40 Compute Performance by Up to 2x

NVIDIA's GeForce RTX 30-series and RTX 40-series graphics cards are based on the same dies as their professional-grade GPUs. This means the compute potential of consumer-oriented RTX GPUs is restricted by NVIDIA's firmware stack, but modders have managed to unlock the instruction issuing rate equivalent to their professional GPU counterparts. Starting with the GeForce RTX 30-series "Ampere," a researcher developed an unlock that effectively doubles the instruction issuing speed of FP/BF16 FMA into FP32 operations on GA102, GA103, GA104, and GA106 cards. As a result, cards like the GeForce RTX 3090 Ti experience up to 80% faster matrix multiplication in BF16, which is crucial for running nearly all neural networks. The unlock supports GA102, GA103, GA104, and GA106-based GeForce RTX graphics cards.

Next, we have the GeForce RTX 40-series "Ada Lovelace" with another project that has demonstrated a massive 2× speedup in FP16 and BF16 tensor throughput with FP32 accumulation. When the GeForce RTX 4090 was tested with FP16 tensor throughput and matrix accumulation, the GPU achieved 165.6 TeraFLOPS, while the unlock pushed that number to 331.6 TeraFLOPS. This is intriguing as these computational performance numbers are typically reserved for the higher-end professional segment like the NVIDIA RTX 6000 "Ada Lovelace." It seems NVIDIA artificially limited the AD102 silicon's performance in compute-sensitive matrix multiplication and accumulation, affecting the performance of running large language models. For this unlock, AD102, AD103, and AD106 are officially unlocked.

YouTube Sets Early 2027 Target for AV2 Streams on the Web

The Alliance for Open Media (AOMedia) shared an update on AV2 this week, and the clearest date in it comes from YouTube. Des Keane, Engineering Director at YouTube, confirmed the platform will begin streaming in AV2, which is the royalty-free successor to AV1 that AOMedia says offers roughly 30% better compression. In practice, that means either sharper video at the same bitrate or the same quality using less bandwidth, and the codec also adds support for multi-stream and multi-view video, higher frame rates, and improved HDR. Keane put it plainly: "YouTube AV2 streams will be available for Web as soon as early 2027."

Chrome is the other half. Jim Bankoski, VP of Engineering for Google Chrome, said AV2 playback has been demonstrated in the browser, though Google hasn't said which Chrome version will get it or whether it will be on by default. That demo dates back to CES 2026, where Google streamed AV2 videos from YouTube to a gaming laptop at 1080p and 24 FPS using a custom Chrome build with the reference decoder. Netflix and Prime Video are also exploring the format, but neither has shared a timeline. Meta wants to repeat its AV1 rollout with AV2, and Intel is working on SVT-AV2.

AMD Lists Three Zen 6 "Medusa" GPU IDs in HIPRT 3.2

AMD has published HIPRT 3.2, the latest version of its ray tracing library for HIP, and the release notes name the integrated graphics of its upcoming Zen 6 "Medusa" processors outright. Under new features, AMD lists Medusa support for three GPU architecture IDs: gfx1170, gfx1171, and gfx115E. HIPRT is mostly used by rendering apps like Blender, so this says nothing about gaming performance. Still, the link between these IDs and Medusa has so far rested on compiler patches and leaks, and now the Medusa name sits right in AMD's own changelog.

The gfx117x IDs belong to RDNA 4m, which first showed up in LLVM back in February. AMD added GFX1171 and GFX1172 targets in March, merged GFX1171 into Mesa 26.3 in August, and pushed RDNA 4m firmware later that month. Despite the name, RDNA 4m sits in the GFX11 family with RDNA 3 and 3.5, plus a few RDNA 4 style additions like FP8/BF8 conversion and new matrix instructions. The gfx115E entry is the more curious one. GFX115x covers the RDNA 3.5 graphics in Strix Point and Strix Halo, so it looks like at least one Medusa chip could stick with RDNA 3.5. The same release fixes a bug that truncated architecture numbers containing a letter, gfx115E included.

ASUS Rolls Out FSR 4.1 Preview for ROG Xbox Ally Handhelds

ASUS has started rolling out a preview GPU driver that brings AMD FSR 4.1 upscaling to the ROG Xbox Ally X and ROG Xbox Ally X20. Owners can grab it through Armoury Crate SE under Update Center, then Preview Updates. This follows AMD's announcement earlier this week that FSR 4.1 is expanding to Radeon RX 6000 series GPUs, APUs, and Xbox. According to ASUS, the handheld version runs on a lighter model that is less than half the size of the original INT8 model, built with the Ryzen AI Z2 Extreme in mind. FSR 4.1 works in over 400 games that support FSR 3.1 or newer. Handheld owners have waited a while for this, as AMD previously had no definitive decision on RDNA 3.5 APUs before confirming FSR 4 for APUs by the end of 2026.

Once FSR is enabled in a game's settings, users also have to add the FSR tile in Command Center and switch it on. ASUS warns that RSR, the system-level upscaler, should not be combined with it. Neither AMD nor ASUS has shared any performance numbers yet, and ASUS notes that preview updates may still have bugs. As for the regular ROG Xbox Ally, ASUS says it will also get support through a GPU driver update coming soon, though there's no date yet. The Xbox Series X|S consoles are next in line and will reportedly get FSR 4.1 in the coming weeks, while FSR Diamond has already been teased for Project Helix.

Two Pallets of RTX 5090s Vanish From São Paulo Airport, One Card Already Listed Online

Three airport workers have reportedly been identified as the people behind the disappearance of two pallets of ZOTAC GeForce RTX 5090 cards from Guarulhos International Airport in São Paulo, Brazil, and at least one of those cards has already popped up for sale online. The shipment belonged to Oderço, a distributor based in Maringá, which imported the cards from the US and puts their cost at around R$2.5 million (roughly $470,000). According to a police report filed by the airport operator, two outsourced forklift operators moved the pallets out of the cargo terminal's warehouse on September 21, which security cameras recorded, while a third employee, a checker, released the goods in the electronic system so they could get past the airport's controls. The cards were then loaded onto a truck and haven't been seen since.

Apparently, one of them later turned up on OLX for R$32,000 (about $6,000), and Brazil's Jornal da Band reported that its serial number matched one of the imported cards. The listing has since been pulled, and it's not clear who posted it or whether that person was involved. There's no word on arrests either. Oderço's president, Oderço Matos, said two pallets don't just vanish on their own inside an international airport. GPU shipments have gone missing before, including containers of MSI RTX 3090 cards and a truck full of EVGA cards. However, this haul is worth a lot more on the street right now, with some RTX 5090 listings going for well past $8,000 as the card remains in heavy demand for AI workloads.

GALAX GeForce RTX 5080 HOF Extreme Comes with Dual 12V-2x6 Power Connectors

GALAX has unveiled the specifications for its GeForce RTX 5080 Hall Of Fame Extreme edition graphics card, the only RTX 5080 model featuring two 12V-2x6 power connectors. Interestingly, the card has a TGP of 400 W, and the official GALAX announcement states that the two power connectors are not required for simultaneous use. Instead, they are offered as alternatives to suit different PC builds and cable management preferences, so even if both connections are used, only one will be active. One standard connector is located on the top of the GPU, in the usual spot. The second 12V-2x6 power cable is positioned at the bottom of the card, concealed behind a GPU shroud near the PCIe x16 connector. This serves as an alternative power supply, allowing system builders to hide visible cables for a cleaner look.

For the GPU itself, GALAX has implemented a 20-phase power delivery system for the GPU and a 4-phase power delivery design for the GDDR7 memory, along with a digital PWM controller, eFuse, and HOF-branded inductors. All of this is mounted on a customized white PCB with gold plating on key connections. In OC Mode, which raises the TGP to 400 W, the GPU clock speed reaches 2,805 MHz. In regular Gaming Mode, the TGP is reduced to NVIDIA's reference 360 W, with clock speeds at 2,655 MHz. Since the 12V-2x6 connector is sufficient for these TGP designs, a second power source is unnecessary. GDDR7 memory is running at 30 Gbps on a 256-bit memory bus for a total of 960 GB/s memory bandwidth.

Server CPU Shipments Projected to Grow Over 40% YoY in 2027, with Arm-Based Products Continuing to Gain Market Share

TrendForce's latest AI server industry research indicates that demand for server CPUs has grown steadily since the second half of 2025 as AI server providers actively introduce AI agent applications. Server CPU shipments are projected to increase 31% YoY in 2026.

In 2027, continued growth in AI inference demand and rising adoption of agentic AI services are expected to drive server CPU shipments up by more than 40% YoY, potentially reshaping a market landscape traditionally dominated by x86 processors.

Super Flower's 12V-2x6 OCP Dongle Watches Every Pin, Blacks Out Your Screen on Overcurrent

Super Flower has launched the VGA 12+4PIN OCP Dongle, a small inline box that sits between the PSU cable and a graphics card's 12V-2x6 connector and keeps an eye on the current running through each 12 V line. It was first shown at Computex 2026, and can now be seen on Super Flower's website, priced at NT$1,190 (roughly $38) in Taiwan. However, you can't buy it on its own just yet. For now, it's only offered as part of a limited promotional bundle with Leadex VIII power supplies, and there's no word on standalone sales or a wider release.

As for the protection part, each alert needs to hold for 10 seconds before it fires. Between 9.5 A and 12 A on any line, or if two lines drift more than 6 A apart, the LED turns orange and the buzzer starts beeping. Past 12 A, it flashes red with a continuous tone for 180 seconds, stays solid red for another 15, and then triggers "VGA Black Screen Protection." That mode stays active even if the current settles. Super Flower doesn't say how the black screen works or whether power is cut, and simply advises shutting down and checking the cables. That part can be a bit unreliable, but it's safe to assume that most users will suspect something is wrong if the screen blacks out for a significant bit of time.

Leak: Intel Core Ultra 9 4995HX Leads Nova Lake Laptop Lineup

Intel's Core Ultra 9 4995HX looks set to be the fastest Nova Lake chip for laptops. Leaker Jaykihn on X has confirmed that the 4995HX uses an 8+16+4+2 layout, meaning 8 P-cores, 16 E-cores, 4 LPE cores, and 2 Xe3 GPU cores. That adds up to 28 CPU cores, four more than Intel's current Arrow Lake-HX chips, with the extra cores coming from the LPE cluster Arrow Lake-HX lacked. The overall layout matches the 8P+16E+4LP ceiling earlier reports pointed to, and the name itself fits the Core Ultra 4000 branding said to be extending to HX parts. Intel hasn't confirmed any of this, and clocks, cache, and power figures are still unknown.

The same post lists the Core Ultra 9 4970H as 4+8+4+12, so 16 CPU cores and a 12-core Xe3P iGPU. That is the same core count as Panther Lake's flagship Core Ultra X9 388H, so any gains here should come from the new Coyote Cove P-cores, Arctic Wolf E-cores, and Xe3P graphics rather than extra cores. The tiny iGPU on the 4995HX is less of a concern, since HX chips almost always end up paired with a discrete GPU anyway. Jaykihn also put the final 8+16 HX compute tile, built on TSMC N2P, at 14.8 x 6.6 mm, or roughly 98 mm², shortly after die shots of both chips surfaced. Current reports point to the first Nova Lake-H and HX laptops showing up around CES 2027, though that timing should be treated as tentative for now.

NVIDIA Has January Discontinuation Dates Set for RTX 5090 and RTX 5080 16 GB, Claims Leaker

Earlier this month, leakers claimed NVIDIA had stopped allocating GB202 silicon to the GeForce RTX 5090, effectively discontinuing the Blackwell flagship. NVIDIA denied it a few days later, with SVP Jeff Fisher saying the RTX 5090 hasn't been discontinued and will keep shipping. Now, Moore's Law Is Dead (MLID) says the plans are still in place, and the RTX 5080 is part of them too. In the latest Broken Silicon episode, MLID read out a message from what he calls one of his oldest sources, apparently a major retailer, saying NVIDIA has set discontinuation dates for both cards in the first half of January 2027.

According to that source, the last RTX 5090 shipments arrived in September, while the final RTX 5080 batch should land by the end of October and sell through Thanksgiving. We recently reported that Micro Center has already marked its RTX 5090 listings as "Not Carried," which could be further evidence of the RTX 5090 leaving shelves. The retailer from MLID's video also says its system already has a placeholder SKU for an RTX 5080 24 GB, which aligns with BenchLife's earlier report of a Q1 2027 launch.

Rumor: NVIDIA RTX 6060 May Be Years Away, and Possibly Slower Than the PS6

Anyone holding out for an affordable next-gen NVIDIA card might have a long wait. Hardware leaker Kepler_L2 claims on NeoGAF that the GeForce RTX 6060 will "launch much later," adding that, based on recent trends, he doubts it will even match the PlayStation 6. He was replying to a forum member who argued the PS6 would compete with the RTX 6060, in the same thread where he claimed the console's RAM and storage alone cost over $680. Kepler didn't give an actual date, and the 2029 figure going around seems to be an extrapolation. Both Kepler and kopite7kimi have placed NVIDIA's Rubin-based RTX 60 series in 2028, and the x60 cards usually trail the flagship by a few months. The RTX 5060 arrived under four months after the RTX 5090, while the RTX 4060 took over eight months to follow the RTX 4090.

With a flagship launch around mid-2028 or later, that could push the RTX 6060 into 2029. That said, the "much later" comment seems to be relative to the PS6, which is reportedly targeting late 2027, so the exact timing is still anyone's guess. Kepler didn't say why, but the ongoing memory shortage looks like the obvious suspect. NVIDIA reportedly postponed its RTX 50 SUPER refresh earlier this year to prioritize AI products that use the same GDDR7 memory, and the company recently admitted that RTX 5090 supply is tight. Of course, NVIDIA hasn't announced the RTX 60 series officially yet, so treat this as a rumor. Anyone looking to buy a new NVIDIA GPU should probably expect the current RTX 50 lineup to stick around for quite a while.

NVIDIA's Faster DLSS 5 Model Gets Extracted From NBA 2K27, Modders Report Up to 90% Gains

NVIDIA shipped its faster DLSS 5 Neural Rendering model with NBA 2K27 on October 16, and it didn't take modders long to pull it out and test it elsewhere. Renan Maniero on X extracted the new 310.9.16.0 build of the DLL. That lets people inject it into games that don't officially support DLSS 5 through tools like RenoDX and OptiScaler. The new DLL also keeps the older model around as Preset A, with the new one sitting under Preset B, so comparing the two is as simple as switching presets. Gaming Optimized on X did exactly that, and found the new model running up to 90% faster than the old one in The Witcher 3. The channel points out that the gain varies from game to game, so 90% looks like the best case rather than the norm.

That roughly aligns with NVIDIA's own claims. The company says the October update shrinks the model by 24% to roughly 118 million parameters, cuts VRAM use at 4K by 20%, and runs the neural rendering workload up to 2x faster than the September version. NVIDIA also quoted up to 40% higher frame rates in NBA 2K27 specifically. Officially, DLSS 5 is still limited to RTX 50-series GPUs and GeForce NOW. NVIDIA plans to bring it to the RTX 40-series, though there's no date yet, and there's no word on RTX 30 or RTX 20 support. Modders, apparently, aren't waiting, since the OptiScaler README notes a modified runtime that gets it running on RTX 20-series cards as well.

Uncharted Trilogy Reaches PC Through Unofficial Ports

The first three Uncharted games have made their way to PC, though Sony had nothing to do with it. Developer SuperMedo, who is also behind an unofficial Bloodborne port, has released Uncharted for Windows. It's an unofficial port of the PS4's Uncharted: The Nathan Drake Collection, covering Drake's Fortune, Among Thieves, and Drake's Deception. Its first public build went up on GitHub on October 15. Sony did bring Uncharted 4 and The Lost Legacy to PC back in 2022, but the original trilogy never followed. The port runs the PS4 version's own x86-64 game code directly on the PC and replaces the console's system functions. It then translates the graphics work to Vulkan through a renderer based on the shadPS4 emulator. Players need their own decrypted dump of the collection, since no game files are included.

All three games get 60 and 120 FPS modes, with anything higher marked as experimental. Anti-aliasing options include TAA, FSR 3.1, FSR 4, and DLAA. Frame generation comes as FSR 3 on any GPU, or as DLSS Frame Generation at 2x on RTX 40 cards and up to 4x on RTX 50. It's not all smooth sailing, of course. Per the GitHub page, one user finished Drake's Fortune on an RTX 4060 laptop at 1080p with DLAA, mostly seeing 100 to 120 FPS, though a crash in chapter 11 corrupted their save. A separate project from BilalReloaded takes a similar approach, with render resolutions from 720p up to 4K. With Sony reportedly scaling back its single-player PC releases, community projects like this one and the recent native Driveclub port are piling up, while work on PS5 emulators continues to be impressive.

Rumored Steam Deck 2 APU Appears in Leaked Benchmarks with 50% More CPU Cores

A new AMD Gainsborough APU, which has been identified as a potential candidate for the Valve Steam Deck 2, has apparently been leaked onOpenBenchmarking. The new engineering sample shows a total of six cores and twelve threads on the CPU, an increase of 50% over the current Steam Deck APU, and a base clock of 2.4 GHz.

Another notable change from the original Steam Deck is that the Gainsborough APU in question has 24 GB of LPDDR5 RAM, instead of 16. That memory was also running at a whopping 9600 MT/s, as opposed to the Steam Deck's 6400 MT/s. This isn't necessarily to say that the Steam Deck 2 will have 24 GB of memory, but it's an indication that AMD, Valve, or some other OEM is testing that configuration. In the benchmark results, the APU reaches 3,508 points in Furmark. While it's not directly comparable, since the Gainsburough APU was running at 1024 × 768 the Valve Steam Deck achieved 3,323 in Furmark at 1280 × 800. Those two results are close, especially given the higher resolution in the Steam Deck testing, but if this is on early drivers, it's a good sign for GPU performance on the Steam Deck 2.
Return to Keyword Browsing
Nov 22nd, 2026 16:32 CST change timezone

New Forum Posts

Popular Reviews

TPU on YouTube

Controversial News Posts