News Posts matching #GPU

Return to Keyword Browsing

Apple Unleashes M5, the Next Big Leap in AI Performance for Apple Silicon

Apple today announced M5, delivering the next big leap in AI performance and advances to nearly every aspect of the chip. Built using third-generation 3-nanometer technology, M5 introduces a next-generation 10-core GPU architecture with a Neural Accelerator in each core, enabling GPU-based AI workloads to run dramatically faster, with over 4x the peak GPU compute performance compared to M4. The GPU also offers enhanced graphics capabilities and third-generation ray tracing that combined deliver a graphics performance that is up to 45 percent higher than M4. M5 features the world's fastest performance core, with up to a 10-core CPU made up of six efficiency cores and up to four performance cores. Together, they deliver up to 15 percent faster multithreaded performance over M4. M5 also features an improved 16-core Neural Engine, a powerful media engine, and a nearly 30 percent increase in unified memory bandwidth to 153 GB/s. M5 brings its industry-leading power-efficient performance to the new 14-inch MacBook Pro, iPad Pro, and Apple Vision Pro, allowing each device to excel in its own way. All are available for pre-order today.

"M5 ushers in the next big leap in AI performance for Apple silicon," said Johny Srouji, Apple's senior vice president of Hardware Technologies. "With the introduction of Neural Accelerators in the GPU, M5 delivers a huge boost to AI workloads. Combined with a big increase in graphics performance, the world's fastest CPU core, a faster Neural Engine, and even higher unified memory bandwidth, M5 brings far more performance and capabilities to MacBook Pro, iPad Pro, and Apple Vision Pro."

Apple Introduces the Powerful New iPad Pro with the M5 Chip

Apple today introduced the new iPad Pro featuring the incredibly powerful M5 chip. M5 unlocks the most advanced iPad experience ever, packing an incredible amount of power and AI performance into the ultraportable design of iPad Pro. Featuring a next-generation GPU with a Neural Accelerator in each core, M5 delivers a big boost in performance for iPad Pro users, whether they're working on cutting-edge projects or tapping into AI for productivity. The new iPad Pro delivers up to 3.5x the AI performance than iPad Pro with M41 and up to 5.6x faster than iPad Pro with M1. N1, the new Apple-designed wireless networking chip, enables the latest generation of wireless technologies with support for Wi-Fi 7 on iPad Pro. The C1X modem comes to cellular models of iPad Pro, delivering up to 50 percent faster cellular data performance than its predecessor with even greater efficiency, allowing users to do more on the go. Available in space black and silver, iPad Pro comes in 11-inch and 13-inch sizes, and features the Ultra Retina XDR display for an unparalleled viewing experience. The game-changing features of iPadOS 26 supercharge iPad Pro and help users handle demanding creative and professional tasks with ease. With staggering performance gains and breakthrough improvements over M1 models, there has never been a better time to upgrade. The new iPad Pro is available to pre-order starting today, and will be available in stores beginning Wednesday, October 22.

"Powered by the next generation of Apple silicon, the new iPad Pro delivers our most advanced and versatile iPad experience yet," said John Ternus, Apple's senior vice president of Hardware Engineering. "iPad Pro with M5 unlocks endless possibilities for creativity and productivity—with a huge leap in AI performance and a big boost in graphics, superfast wireless connectivity, and game-changing iPadOS 26 features, it pushes the boundaries of what iPad can do yet again."

NVIDIA Taps Samsung Foundry for Custom Silicon Manufacturing

NVIDIA announced at the OCP Global Summit that Samsung Foundry is joining its NVLink Fusion ecosystem, bringing Samsung's design and manufacturing muscle into the fold for custom CPUs and XPUs. The partnership positions Samsung to offer end-to-end support. Everything from silicon design and verification through integration and tape-out for chips that can connect directly into NVLink-enabled MGX and OCP rack architectures. For hyperscalers and AI data centers chasing greater performance and tighter CPU-GPU coupling, the development opens another route to custom silicon that speaks NVIDIA's high-bandwidth fabric language. Technically, NVLink Fusion functions as an IP and chiplet framework that enables non-NVIDIA processors and accelerators to become first-class participants in the company's data center fabric.

Earlier partners and chipset makers can now interoperate inside the same MGX and OCP designs, and Samsung adds both fabrication capacity and design services. NVIDIA's own engineering work on the Vera Rubin OCP rack and MGX integration was cited as the backbone that makes these cross-vendor connections possible, and industry players such as Intel, Fujitsu, and Qualcomm were highlighted as part of the larger custom-silicon network. Bringing Samsung Foundry into the NVLink Fusion circle should help shorten development cycles, reduce single-source risk, and give companies more choice when building CPUs and XPUs tuned for large language models and agentic era AI. For integrators, that means faster prototyping of workload-specific hardware and simpler paths to volume production.

HP Expands Its Riot Games Lineup With New OMEN 16 League of Legends and OMEN 25 Gaming Monitor

Today at the 2025 League of Legends World Championship, HP Inc. introduced two new products developed with Riot Games: the OMEN 16 League of Legends Limited Edition Laptop and the OMEN 25 Gaming Monitor. These devices bring fan-favorite design elements and tournament-grade performance to players around the world, on stage and off. This new lineup marks the latest milestone in HP's global collaboration with Riot Games, first announced in 2024. Building on last month's debut of the OMEN 35L VALORANT Limited Edition gaming desktop, these additions further expand OMEN's portfolio of exclusive collaborations with Riot Games' titles.

"From all-out competition to all-in fandom, players want gear that reflects how they play," said Josephine Tan, Senior Vice President and Division President of Personal Systems Gaming Solutions, HP Inc. "This launch and continued collaboration with Riot Games celebrates both, with performance and design that resonates with the gaming community."

Intel "Panther Lake" Naming Scheme Confirmed with Core Ultra X9/X7/X5 SKUs

Last week, Intel finally revealed details about the much-anticipated "Panther Lake" processors. Our in-depth analysis covers all the specifics. However, one missing piece was the official naming scheme that Intel plans to use. Previous rumors suggested that the new naming scheme would include the X moniker to denote the top SKUs. Today, this has been confirmed by the developers of HWMonitor, a tool used for monitoring key system health sensors such as voltages, temperatures, power, currents, fan speeds, utilization, and clock speeds. They verified that the Panther Lake Core Ultra family will feature X9, X7, and X5 SKUs. These processors will include a larger Xe GPU tile with up to 12 Xe3 cores, the fastest memory subsystem, and a CPU structure consisting of four P-cores, eight E-cores, and four LPE-cores.
Below, you can check out the Core Ultra X9/X7/X5 SKUs, now confirmed by the HWMonitor update:

ASUS Unveils AI Factory and Next-Gen Servers with NVIDIA HGX B300 systems at OCP 2025

ASUS today announced its participation in the 2025 OCP Global Summit, being held from October 13-16 at the San Jose Convention Center, booth #C15. At the event, ASUS unveiled its XA NB3I-E12 series AI servers, based on NVIDIA HGX B300 system integrated with NVIDIA ConnectX-8 InfiniBand SuperNICs, 5 PCIe expansion slots, 32 DIMM and 10 NVMe. Designed for enterprises and cloud service providers (CSPs) managing intensive AI workloads, these servers deliver outstanding performance and stability, unlocking the full potential of AI.

Starting this September, ASUS AI POD built on NVIDIA GB300 NVL72 and XA NB3I-E12 servers based on NVIDIA HGX B300 have begun shipping, giving enterprises and cloud service providers early access to cutting-edge AI performance and reliability.

NVIDIA Spectrum-X Ethernet Switches Speed Up Networks for Meta and Oracle

NVIDIA today announced that Meta and Oracle will boost their AI data center networks with NVIDIA Spectrum-X Ethernet networking switches. Meta and Oracle are standardizing on Spectrum-X Ethernet switches as an open, accelerated networking architecture that speeds deployment at scale, unlocks exponential gains in AI training efficiency and shortens time to insights.

"Trillion-parameter models are transforming data centers into giga-scale AI factories, and industry leaders like Meta and Oracle are standardizing on Spectrum-X Ethernet to drive this industrial revolution," said Jensen Huang, founder and CEO of NVIDIA. "Spectrum-X is not just faster Ethernet—it's the nervous system of the AI factory, enabling hyperscalers to connect millions of GPUs into a single giant computer to train the largest models ever built."

Intel 18A Gets Head Start on TSMC N2 with Fab 52 Ramp

Intel announced this week that its Panther Lake processors have entered volume production at Fab 52 using the company's 18A node, which is touted as the first 2 nm-class process to achieve high-volume manufacturing. This milestone gives Intel a slight timing advantage over rival foundries like TSMC, with its N2 node, and Samsung with its SF2 node. However, being first does not guarantee a lasting lead. Customers, partners, and investors will focus on Intel's ability to maintain parametric yields, control manufacturing costs per unit, and expand 18A beyond limited compute tiles. While Intel has shared improvements in defect density and early functional test results, comprehensive parametric yield metrics that demonstrate consistent achievement of power, performance, and frequency targets remain undisclosed. Initial shipments and early ramp figures will determine whether this production start is symbolic or substantive for Intel's manufacturing revival.

At the transistor level, 18A node introduces two major innovations: RibbonFET gate-all-around transistors and PowerVia backside power delivery. RibbonFET replaces vertical FinFET fins with thin horizontal silicon ribbons wrapped by gate material, enhancing electrostatic control, reducing leakage, and allowing shorter gate lengths. Intel reports approximately 5-10% shorter gate lengths when transitioning from FinFET to RibbonFET on 18A and claims over 20% reduction in per-transistor power. PowerVia relocates the power network to the wafer's backside, freeing up front-side metal for signal routing, shortening power paths, and reducing voltage drop at high switching rates. Together, these changes lower switching energy and provide designers with more flexibility to increase frequencies or reduce power consumption in notebooks and other mobile designs.

Intel Phases Out "Ponte Vecchio" and "Arctic Sound" Data Center GPUs

Intel has apparently started a slow deprecation phase of its "Ponte Vecchio" Data Center GPU Max and "Arctic Sound" Data Center GPU Flex series. According to the changelog in the Intel XPU Manager—a free and open-source tool for monitoring and managing Intel data center GPUs—version 1.3.3 deprecates Data Center GPU Flex Series and Data Center GPU Max Series. Owners of these GPUs are now advised to stick with the older version 1.2.42 for maximum feature support, and not update to the latest release. There are several reasons why this is happening, and a part of the suspicion is that Intel is preparing for "Jaguar Shores" AI accelerator transition, even as its has been only two years since its Data Center GPU Flex/Max series have launched.

Built on the Xe-HPC microarchitecture using Intel's 10 process, Ponte Vecchio is a massive chip, measuring about 1,280 mm² and containing roughly 100 billion transistors. It features 16,384 shading units, 1,024 tensor cores, and has a peak TDP of up to 600 W per card. The Max 1100, 1350, and 1550 SKUs, which offer 48 to 128 GB of memory, once made the design appealing for specialized HPC and AI applications. However, its large scale results in high costs, significant power consumption, and integration challenges. Intel's "Arctic Sound," also based on the Generation 12.5 GPU microarchitecture and manufactured on the Intel 10 node, is a much smaller chip, measuring around 190 mm² with about 8 billion transistors, 8,192 shading units, and 128 ROPs, and has a TDP close to 500 W. The Arctic Sound 1T, 2T, and M SKUs are compact 16 GB designs aimed at midrange data center tasks, with the GPU first appearing in 2021.

Giga Computing Showcases AI-Centric Server Solutions at GITEX Global 2025

Giga Computing, a subsidiary of GIGABYTE and a global leader in AI servers and advanced cooling technologies, announced its participation in GITEX Global 2025, featuring nVent and Iceotope at GIGABYTE booth H10-18. As a trusted partner in the AIDC Infra ecosystem, Giga Computing designs and delivers high-performance AI servers. nVent is a leader in high density liquid cooling, rear-door heat exchangers (RDHx), IT server racks and cabinets, manifolds and iPDUs. Iceotope liquid cooling solutions offer enterprises and cloud service providers a unique combination of peak thermal efficiency and deployment flexibility. Working together, they underscore a collaborative ecosystem supported by a joint commitment to energy-efficient and sustainable AI infrastructure.

Driving Growth and Innovation in the Middle East
Participation in GITEX reflects Giga Computing's commitment to building a strong presence in the Middle East market, a region experiencing a rapid digital transformation and AI adoption. As one of the world's top-tier AI server suppliers, Giga Computing is bringing its complete AI data center solutions to the Middle East, providing enterprises, hyperscalers, and emerging AI innovators with scalable infrastructure designed for cutting-edge workloads.

NVIDIA to Fund Musk's xAI Chips in $20 Billion Investment Round

Elon Musk's AI startup xAI is raising more money than first planned, bringing in backers including NVIDIA to lift its ongoing funding round to $20 billion, according to people familiar with the matter, Bloomberg reports. The financing includes equity and debt in a special purpose vehicle (SPV) that will buy NVIDIA processors and rent them to xAI for use in its Colossus 2 project, the people said. That's the name of its largest data center site, located in Memphis. NVIDIA is putting in as much as $2 billion in the equity portion of the deal, a strategy by the chipmaker that helps speed up its customers' AI investments. However, NVIDIA representatives declined to comment while an xAI spokesperson didn't respond to a request for comment from Bloomberg. Musk posted on X in September that the company was "not raising any capital right now."

Musk's xAI financing would split between about $7.5 billion of equity and as much as $12.5 billion of debt in the SPV, the people said. The vehicle will be used to buy NVIDIA processors, and Musk's AI startup would then rent the chips out for five years, letting Wall Street financiers get their money back. The unique deal structure, backed by the GPUs rather than the company, could give tech firms a way to cut debt exposure. Earlier this week, OpenAI announced a deal to use AMD chips over multiple years, in addition to the announcement from late September that it will install at least 10 gigawatts of NVIDIA systems for its next-generation AI infrastructure. Meta has signed several multibillion-dollar deals in recent months, including a $29 billion financing package for data centers while Oracle also raised a $38 billion debt package for its infrastructure.

AMD Prepares Radeon PRO W7900D GPU for China

AMD appears to have slipped a new professional GPU SKU into a recent driver changelog under the name Radeon PRO W7900D. It is a subtle entry, but it could be AMD's way of slowly introducing a new China-specific workstation GPU. The new suffix suggests a revision of the Navi 31-based W7900 family, making it the third listed W7900 variant after the original W7900 and the later W7900 Dual Slot model. The driver reference went live more than ten days ago. However, AMD has not published a product page, and the W7900D does not appear as a selectable option on the driver download site, nor have any official specifications been provided.

What we do know about the Radeon PRO W7900 series could frame our expectations for this unlisted SKU. Both the W7900 and the W7900 Dual Slot use the Navi 31 die, offer 6,144 SPs, and include 48 GB of GDDR6 ECC memory. The Dual Slot variant arrived with a slimmer cooler and a lower launch price compared with the original W7900. Publicly available figures for those cards list peak FP32 compute around 61 TFLOPS, boost clocks near 2.5 GHz, a 384-bit memory bus running roughly 18 Gbps speeds, and board power in the region of 295 W. The W7900D name has prompted speculation that AMD could be preparing a region-specific variant similar to NVIDIA's China-focused D series, but checks of regional AMD sites have not yielded a confirming product page. We are yet to see how AMD positions this lone SKU, considering it is on the older RDNA 3 uArch, and we are slowly moving into RDNA 5 / UDNA rumors.

OpenAI will Buy AMD GPUs, AMD Offers $30+ Billion in Stock to Sweeten the Deal

AMD (NASDAQ: AMD) and OpenAI today announced a 6 gigawatt agreement to power OpenAI's next-generation AI infrastructure across multiple generations of AMD Instinct GPUs. The first 1 gigawatt deployment of AMD Instinct MI450 GPUs is set to begin in the second half of 2026. AMD's strong leadership in high-performance computing systems and OpenAI's pioneering research and advancements in generative AI places the two companies at the forefront of this important and pivotal time for AI.

Under this definitive agreement, OpenAI will work with AMD as a core strategic compute partner to drive large-scale deployments of AMD technology starting with the AMD Instinct MI450 series and rack-scale AI solutions and extending to future generations. By sharing technical expertise to optimize their product roadmaps, AMD and OpenAI are deepening their multi-generational hardware and software collaboration that began with the MI300X and continued with the MI350X series. This partnership creates a true win-win for both companies, enabling very large-scale AI deployments and advancing the entire ecosystem.

Intel's "Panther Lake" Microarchitecture Deep Dive Set for October, Full Launch in 2026

The story of Intel's first 18A product, "Panther Lake," remains shrouded in mystery, at least for a few more days. According to earlier reports, Intel was expected to start shipping the first PTL-U and PTL-H SKUs by the end of 2025, with additional SKUs to follow in 2026. However, that situation could be subject to change, according to Golden Pig Upgrade. The rumored October 9 launch date is reserved only for the microarchitecture deep dive, and the launch and final specifications of specific models will occur at CES 2026. Being Intel's first big bet with the 18A node, Panther Lake is a product of years of Intel's manufacturing innovation, which needs to demonstrate that the investment in advanced manufacturing was worthwhile. Reportedly, Intel started limited shipments of its 18A nodes to U.S. customers in Q3, with these wafers already in production and initial output of its own CPUs expected in Q4.

Given that Panther Lake is the first 18A node product, we could expect that to be a large part of those shipped wafers, most likely going to OEMs for testing before integration. As a reminder, the low-power PTL-U models are designed for a 15 W TDP and are expected to come in 6-core and 8-core versions. Some SKUs will feature four high-performance P-cores paired with four LPE-cores, while others will have only two LPE-cores complementing four P-cores. Both families will utilize Xe3-based integrated graphics, with entry models featuring four GPU cores. The more powerful PTL-H line will scale up to around 16 CPU cores, comprising four P-Cores, eight E-cores, and four LPE-cores. Some H-series parts might include up to 12 GPU Xe3 cores for integrated graphics, but the final configurations will be revealed when Intel officially launches the Panther Lake product family.

Thermal Grizzly Introduces New Minus Pad extreme 2 and Minus Pad High Compression Thermal Pads

Thermal Grizzly introduces the successor to the minus pad extreme: Minus Pad extreme 2. Moving forward, Minus Pad extreme 2 will lead the TG thermal pad portfolio in terms of thermal conductivity. The most significant difference compared to the minus pad extreme is a new material composition. This results in an optimized price-performance ratio while maintaining the outstanding thermal conductivity at a level comparable to that of the minus pad extreme.

The Minus Pad High Compression thermal pads feature exceptionally high compressibility. This makes them particularly versatile in applications where a single pad needs to bridge varying height differences simultaneously. This is especially useful when modifying graphics cards with a GPU water block or when replacing thermal pads on GPU coolers, as a Minus Pad High Compression (HC Pad) can be used across different components such as VRAM, voltage regulators, and other contact surfaces.

Intel Details Core Options for "Nova Lake" Desktop and "Diamond Rapids" Xeon 7 Processors

Intel confirmed today which core microarchitectures will power its next-generation client and server processors, doubling down on previous rumors that "Coyote Cove" P-cores will power the "Nova Lake" family and "Panther Cove" P-cores will power the "Diamond Rapids" line. The details are outlined in Intel's latest ISA reference update and provide the clearest public signal yet about the company's roadmap through 2026. Nova Lake and Diamond Rapids are both slated for release next year, and the confirmation removes much of the guesswork about which P-core and E-core designs each family will use. On the client side, Nova Lake will pair Coyote Cove P-cores with "Arctic Wolf" E-cores and introduce an updated integrated graphics tile based on the Xe3 lineage.

Desktop Nova Lake-S parts are expected to scale to very high core counts, up to 52 cores, while HX mobile variants will focus on iGPU in power-optimized packages. Intel is also planning a platform update with an LGA 1954 socket, and the company suggests that it will deliver meaningful improvements in IPC and performance per watt compared to the previous Cougar Cove and Darkmont designs. For servers, Diamond Rapids will use Panther Cove P-cores, push core density aggressively, and is expected to arrive in the second half of 2026. These initial Panther Cove server chips will not include simultaneous multithreading, a decision Intel appears set to correct with the follow-up "Coral Rapids" family, which will reintroduce Hyper-Threading for datacenter workloads. There is also discussion of a Panther Cove-X variant designed for workstation users and a Wildcat Lake entry tier that will replace Twin Lake, utilizing Cougar Cove P-cores and Darkmont E-cores.

Aqua Computer Introduces AMPINEL, Active Power Management for Graphics Cards

Aqua Computer presents AMPINEL, a monitoring device for 12V-2x6 power supply connections that combines active current balancing, monitoring, and emergency shutdown for the first time. In the event of uneven load distribution, AMPINEL intervenes without delay, balancing the currents and thus preventing thermal overload of the connectors and cables. AMPINEL monitors all currents, voltages, and temperatures at the connector. The flexible alarm management with emergency shutdown ensures reliable protection of the hardware.

Uniform power distribution - reliable overheating protection
When up to 600 watts flow through the 12V-2x6 connection, insufficient contact between connectors can result in an uneven distribution of power. Under high loads, this inevitably leads to critical temperatures and can cause damage to hardware. AMPINEL eliminates this risk by actively balancing the current flow across all six lines. As soon as the current in a line exceeds 7.5 A, AMPINEL intervenes. The hybrid control structure, consisting of digital and analog control, distributes the load evenly and without delay. The integrated microcontroller monitors all operating conditions in real time and continuously adjusts the control.

Apple M5 SoC Brings 35% GPU Gain and 10-15% CPU Boost vs. M4

Apple's M-series of SoCs continues with generational performance improvements, and the upcoming M5 SoC will be no exception. A Russian YouTuber Wylsacom obtained an unreleased iPad Pro based on the M5 SoC and published benchmark data for it. As a reminder, the same source correctly previewed last year's M4 MacBook Pro. The early numbers suggest that Apple's next-generation tablet silicon focuses more on graphics than on a raw CPU jump. A single-core CPU IPC performance improves by roughly 12% relative to frequency, and multi-core performance improves by approximately 15% over the standard M4 in iPad Pro. Meanwhile, GPU performance shows significantly larger gains in some tests, reaching as much as 35% in graphics workloads. That kind of uplift won't transform every task, but it could deliver noticeably smoother frame rates in games and more sophisticated visual effects in creative apps.

Given Apple's habit of carrying phone-class architecture into its M-series chips, the M5's strengths align with improvements first seen in the family's A-class parts, where advances in neural and graphics subsystems have increasingly been used to augment traditional rendering pipelines. Digging into the sample that surfaced, the leaked iPad Pro is listed as a 256 GB model with 12 GB of RAM, which is a step up from last year's comparable 8 GB configuration. It is reasonable to expect Apple to reserve 16 GB or higher for larger storage tiers as it has in past launches. If the M5 really pairs modest CPU boosts with a substantial GPU jump, the benefits would extend beyond the iPad form factor, with thinner MacBook Air and Pro models that receive the chip also seeing improved CPU and GPU data scores, thanks to bigger cooling capacity. Especially with active cooling, Apple could push these chips way past this point, and again claim the performance crown.

MSI Pushes Afterburner 4.6.6 Stable Update After Years of Beta

MSI has quietly pushed a long-awaited stable update for its Afterburner utility, turning years of beta evolution into an official v4.6.6 stable release that many in the PC community have been waiting for. The new build bundles a refreshed RivaTuner Statistics Server and folds a large number of previously experimental features into the supported release, bringing broader GPU compatibility, more flexible fan and power controls, and several important security fixes. Users will notice official handling for the newest GeForce RTX 50 series desktop cards, support for more complex multi-fan layouts, and added hooks for advanced voltage management on boards that expose those rails. Because some of the deeper hardware controls depend on how a card's firmware and driver behave, a subset of functionality will be most useful on partner designs engineered for overclocking. Additionally, unofficial RDNA 4 support is added, and since MSI doesn't offer RDNA 4 cards, it will remain unofficial.

Beyond device compatibility, the update modernizes the toolchain and usability, as the app now targets the current Visual C++ runtime, and older systems, such as Windows XP, are no longer supported. The installer approach minimizes bloat by fetching required runtime components when needed. The release also introduces UI polish, featuring new skins, localization improvements, and a simplified method for hiding integrated GPUs from the interface, along with an expanded voltage and frequency editor and increased on-screen telemetry capacity. The bundled statistics server gains many behind-the-scenes enhancements and security patches that improve stability and reduce risk.

Open Source Adreno 800 GPU Drivers Start Trickling Into Linux Driver Stack

Qualcomm recently announced the Snapdragon X2 Elite and Elite Extreme, which appear to be putting up some respectable CPU and GPU benchmarks, but there hasn't been much word on Linux compatibility until now. A recent Linux driver patch sees the start of adding open source driver support for the Qualcomm Snapdragon 800 series GPUs in the X2 Elite SoC to the open-source Qualcomm MSM Direct Rendering Management driver. The patch submission specifically calls out changes to the driver that were necessary because of the new slice-based architecture, as well as changes that allow for higher current draw and necessary thermal mitigations.

The development team also mentioned that Mesa and Vulkan support for the Adreno 840 and other 800 series Arm GPUs is currently work-in-progress and "will be posted in the near future". For reference, when it comes to the previous Snapdragon X Elite launch, there were significant driver updates as recently as July 2025, so even if the Qualcomm team is working on Linux driver support, it might be a while before Snapdragon X2 Elite and X2 Elite Extreme laptops are actually at feature-parity with x86 systems on Linux.

Snapdragon X2 Elite Extreme Benchmarks Highlight Impressive CPU Performance but Underwhelming GPU

Qualcomm recently unveiled the Snapdragon X2 Elite Extreme SoC as its latest foray into the world of Windows on Arm hardware. It claimed that the new processor would deliver up to 75% higher CPU performance and a 2.3 × improvement in performance per watt. Now, following the post-launch hype, independent benchmarks are starting to appear and show what Qualcomm didn't on-stage. The most recent set of benchmarks comes from ComputerBase, the German-language outlet known for its in-depth hardware testing, and it shows that Qualcomm's claims of CPU performance were surprisingly impressive. Although it also becomes clear why Qualcomm stuck to performance per watt instead of focusing too much on raw GPU performance.

According to the ComputerBase review, the Qualcomm Snapdragon X2 Elite Extreme was able to outperform Apple's best and brightest, the Apple M4 Pro, in both Cinebench 2024 and Geekbench 6.3 multicore benchmarks, scoring 1,964 and 23,693 points, respectively. Results are a little less consistent when it comes to single-core scores, with the Apple M4 Pro pulling ahead of the Snapdragon SoC by 7% in the Cinebench 2024 single-core test but falling behind the X2 Elite Extreme by 7% in Geekbench 6.5. NPU performance was also impressive, with the X2 Elite Extreme's 80 TOP NPU coming out undefeated, at 128% faster than the Intel Core Ultra 9 288V in the ASUS Zenbook S 14. Unfortunately, that seems to be where the good news for Qualcomm ends—mostly. When it comes to GPU tests, the Snapdragon X2 Elite Extreme remains well behind the Apple M4 Pro in both the 3DMark Steel Nomad Light Unlimited and 3DMark Solar Bay Unlimited tests, scoring 5,687 points and 22,946 points, respectively. While that puts the new Arm silicon a solid 37-40% ahead of the nearest competing iGPU, the Apple M4 10-Core, it is still 37-46% behind the Apple M4 Pro in GPU benchmarks. It remains to be seen how this translates to real-world gaming performance and actual CPU workloads, given the translation layer that Arm SoCs currently still rely on for many non-native Windows apps and games, but it's an impressive showing nonetheless.

AMD Driver Leaks Fluid Motion Frames 3 Ahead of Official Announcement

Just last week, AMD announced that it has enabled PyTorch on Radeon RX 7000 and 9000 GPUs, allowing users to more easily run local LLMs on Windows. Now, however, it seems as though AMD may also introduce AFMF 3 (AMD Fluid Motion Frames 3) in an upcoming driver update. According to recent posts on the Guru3D forums by The Creator and ecffg2010, there is mention of AFMF 3 in the same 25.20.1 Windows Preview testing version of the AMD graphics driver. There is, at the time of writing, no mention of AFMF 3 on the official beta driver update notes or in the AMD Software GUI, but exporting the game profiles with the AMD GPU Profile Manager revealed the inclusion of the new frame generation version.

AMD first introduced AFMF 2 in July 2024, with version 2.1 launching in March 2025, so it is somewhat expected that AMD will soon launch AFMF 3, but it's unclear when the option will be exposed to the end user. So far, AFMF 2.1 has been available for AMD GPUs going all the way back to the Radeon RX 6000 series. PyTorch support, however, is currently only available for RX 7000 and 9000 series GPUs, so it remains to be seen if AMD will abandon the RX 6000 GPUs when it comes to AFMF 3. There is speculation that the driver will drop support for the RX 7000 series altogether, but several users on the forum note that they have installed the driver on their 7000 series GPUs without issue.

AMD Enables PyTorch on Radeon RX 7000/9000 GPUs with Windows and Linux Preview

AMD has rolled out a public preview of ROCm 6.4.4 that finally lets PyTorch run natively on Windows and Linux for a broad swath of its consumer silicon. The update brings official framework support to Radeon RX 9000 (RDNA 4) and RX 7000 (RDNA 3) GPUs, as well as to select Ryzen AI 300 "Strix" and Ryzen AI MAX "Strix Halo" APUs, fulfilling a promise AMD made at Computex 2025 to make ROCm more cross-platform and developer-friendly. "At Computex this year, I shared our commitment to making ROCm a true cross-platform, developer-first stack. I said we'd bring ROCm to Radeon on Windows and Linux in the second half of 2025. I'm proud to say that today, we're delivering on that promise," says Andrej Zdravkovic, Senior Vice President and Chief Software Officer.

For Windows users, this is a genuine quality-of-life change: no more dual-booting, convoluted workarounds, or constant reliance on cloud instances to test and run AI models locally. AMD describes the release as a foundation rather than a finished product. The preview provides native PyTorch wheels, allowing developers to prototype, benchmark, and provide feedback while AMD iterates on performance and feature coverage. The company positions this step as part of a broader effort to extend ROCm beyond the data center and into the everyday machines of creators and developers. At the same time, enterprise customers continue to get higher-scale optimizations through the ROCm 7.0 family for Instinct and EPYC platforms. If community testing is successful, users can expect more frequent updates and enhanced Windows support in the next release cycle.

Alphacool Unveils New ES GeForce RTX 5090 Gaming 1-Slot GPU Cooler

Alphacool International GmbH, headquartered in Braunschweig, has been one of the pioneers in the field of PC water cooling for over two decades. With one of the broadest product portfolios in the industry and extensive expertise, Alphacool is once again setting new standards: the product range is now being expanded with the new ES GeForce RTX 5090 Gaming 1-Slot Design GPU cooler.

The ES GPU coolers have been specially developed for use in server cases and workstations. Thanks to their slim 1-slot design and rear-mounted connectors, they can be easily integrated into densely equipped systems and stacked without blocking adjacent PCIe slots. This simplifies tubing and enables a space-saving, efficient integration into the water loop.

Qualcomm Unveils Snapdragon 8 Elite Gen 5 SoC With Impressive Performance Claims

As expected, along with the latest laptop-class Arm SoC, the Snapdragon X2 Elite series Qualcomm has officially unveiled its latest flagship mobile SoC, the Snapdragon 8 Elite Gen 5. The 8 Elite Gen 5 launch sees Qualcomm lean even further into AI workloads while adding a handful of spec upgrades to the new SoC. It features an Oryon Gen 3 CPU that has two Prime cores capable of up to 4.60 GHz and six Performance cores that run at up to 3.62 GHz, all of which is paired with 24 MB of cache split into 12 MB chunks—12 MB per core cluster. Qualcomm says this will increase single-core performance by 20% and multicore performance by 17%, compared to the likes of the Snapdragon 8 Elite Gen 2, that is. Based on Geekbench Browser scores and the information shown off by Qualcomm in the presentation, this means the 8 Elite Gen 5 performs about on-par with its true predecessor, the Snapdragon 8 Elite Gen 3, but early benchmarks from hands-on tests at the Snapdragon Summit tell a different story. It seems as though the 8 Elite Gen 5 delivers single-core Geekbench 6 results of around 3,800 points, with multicore scores coming in at around 12,400 points—a remarkable ~3,000 uplift over the current Geekbench leader, the Samsung Galaxy S25 Ultra and its Snapdragon 8 Elite Gen 3 for Galaxy.

Of course, as with much of the recent tech announcements, Qualcomm is focusing heavily on AI workloads with the launch of the Snapdragon 8 Elite Gen 5. AI performance improvements and NPU upgrades abound, with Qualcomm claiming the new Hexagon NPU is 37% faster and 16% more power efficient than the previous generation. The new Hexagon NPU now also works together with the ISP to process videos more efficiently and quickly. Similarly to how ISPs typically process images, the NPU in the 8 Elite Gen 5 will be leveraged to manage things lighting, color reproduction, and sharpness. Qualcomm also demonstrated generative image processing with the example of removing glare and artifacts from images "instantly." The new NPU will also enable ModelBest on-device agentic AI with onboard training and a memory of past conversations. Supposedly, this AI will be able to access a number of on-device apps, files, and functions, including location, stored images, and social media in order to automatically post, classify photos, and do other generative tasks.
Return to Keyword Browsing
Nov 22nd, 2026 17:28 CST change timezone

New Forum Posts

Popular Reviews

TPU on YouTube

Controversial News Posts