NVIDIA has developed a new algorithm for path tracing that will make it 2-3 times faster, deliver better visual quality, and improve the robustness of real-time path tracing. Scheduled for a technology demo at the ACM conference on Computer Graphics and Interactive Techniques in May, NVIDIA has found a way to make real-time path tracing a fully realistic rendering process, moving beyond the usual emulation, software denoising, upscaling, and other techniques we use today. Since real-time path tracing is extremely compute-intensive, modern GPUs are still not powerful enough to execute the full path tracing technology in all its potential. Instead, we currently rely on emulations in modern games, implemented across many titles to address this computational demand. Below is the abstract described on NVIDIA's paper page, though the paper itself is not yet available as its publication is scheduled for May. However, note that the technology is still in development, and we have to wait more for it to hit commercial games.

NVIDAAlgorithms leveraging ReSTIR-style spatiotemporal reuse have recently proliferated, hugely increasing effective sample count for light transport in real-time ray and path tracers. Many papers have explored novel theoretical improvements, but algorithmic improvements and engineering insights toward optimal implementation have largely been neglected. We demonstrate enhancements to ReSTIR PT that make it 2-3x faster, decrease both visual and numerical error, and improve its robustness, making it closer to production-ready. We halve the spatial reuse cost by reciprocal neighbor selection, robustify shift mappings with new footprint-based reconnection criteria, and reduce spatiotemporal correlation with duplication maps. We further improve both performance and quality by extensive optimization, unifying direct and global illumination into the same reservoirs, and utilizing existing techniques for color noise and disocclusion noise reduction.

AMD is planning a significant return to establishing new manufacturing ties with its former silicon manufacturing venture, which it sold back in 2008. According to recent reports, AMD intends to collaborate with GlobalFoundries for new Co-Packaged Optics (CPO) for the next-generation Instinct MI500 series of AI accelerators. CPO technology will help AMD remain competitive in the AI data center space, where signal transfer is limited by traditional copper-based wiring. By harnessing the power and speed of light, CPO technology will enable data transfer between multiple nodes, and even multiple facilities, without the speed loss and increased latency associated with copper-based connections. Consequently, AMD is investing considerable effort to secure Photonics Integrated Circuit (PIC) manufacturing with its former partner, GlobalFoundries, which will handle PIC manufacturing, while the actual packaging will be managed by ASE to finalize the CPO design. This approach allows AMD to leverage the strengths of both companies, likely resulting in the best solution for its systems next year.

The Instinct MI500 series of AI accelerators is scheduled for release in 2027, as this year's lineup focuses on the Instinct MI400 series, which includes multiple SKUs for AI and HPC workloads. While these designs are already among the industry's leading AI accelerators, AMD aims to enhance performance further with the 2027 release using CPO technology. This will significantly contribute to delivering a solution that uses less power and offers much higher overall bandwidth compared to traditional copper data transfers. However, AMD is not alone in this CPO push. NVIDIA is also collaborating with semiconductor makers to develop a CPO system for "Vera Rubin," particularly the "Rubin Ultra" variant.

It has been a while since one of Microsoft's Windows 11 updates wreaked havoc among PC enthusiasts, and the April update seems to be another in line for IT administrators. In an updated support document on the Windows 11 KB5083769 cumulative update, Microsoft officially confirms that some users are encountering an unexpected BitLocker recovery prompt after applying the update. However, the good news is that this issue requires a specific BitLocker Group Policy configuration that Microsoft refers to as "unrecommended," which is less likely to be found outside of managed IT departments. This includes systems managed through group policy settings, making it less likely to affect PC enthusiasts' builds. For the issue to occur, a few conditions must be met. First, BitLocker must be enabled on the PC. Second, the BitLocker Group Policy setting "Configure TPM platform validation profile for native UEFI firmware configurations" must be configured with PCR7 (a TPM 2.0 module register) included in the validation profile.

Once these conditions are met, running system information, or msinfo32.exe, shows that the Secure Boot State PC47 Binding is "Not Possible." Additionally, the Windows UEFI CA 2023 certificate must be present in the device's Secure Boot Signature Database (DB), making the device eligible for the Windows Boot Manager signed in 2023 to become the default. Finally, the device must not be running this 2023-signed Windows Boot Manager that it supports. Only by fulfilling this unusual set of conditions does the BitLocker Recovery environment trigger, leading to a password prompt. After this, subsequent startups will not trigger it again.

NVIDIA's GeForce RTX 3060 12 GB "Ampere" GPU, which is two generations old, is expected to make a comeback in June this year. Meanwhile, the rumored RTX 5050 9 GB edition is reportedly on hold. According to well-known hardware leaker MEGAsizeGPU on X, NVIDIA is pausing the transition from its 8 GB RTX 5050 "Blackwell" version to a 9 GB model due to the reintroduction of the GeForce RTX 3060 12 GB "Ampere" GPU. Since both of these GPUs compete in the budget segment, the company will reportedly only release the older GeForce RTX 3060 12 GB SKU with a 192-bit wide memory bus. For this, NVIDIA will once again use Samsung's 8 nm DUV node, as it has in the past. The entire NVIDIA "Ampere" architecture lineup was produced on the 8 nm DUV node, and its return after several years was unexpected. However, since NVIDIA has transitioned to TSMC for manufacturing its "Ada Lovelace" and latest "Blackwell" GPUs, becoming TSMC's largest customer using the 5 nm node, this move is intriguing.

The NVIDIA GeForce RTX 3060 features 3,584 CUDA cores and 12 GB of GDDR6 memory. In contrast, the GeForce RTX 5050 has newer CUDA cores on a smaller node with 2,560 cores on the GB207 die, but it has less memory, with only 9 GB of the rumored GDDR7. Since the resurrected GeForce RTX 3060 will use the older GDDR6 memory, sourcing GDDR6 may have been easier for NVIDIA, leaving the more expensive GDDR7 for its other GPUs.

One of the senior executives from Samsung Foundry, Shawn Han, has officially joined Intel Foundry to help secure new customer deals. With over 30 years of experience in the semiconductor industry, Shawn Han has worked through many node designs, transitions to new technologies, and has observed shifts in customer behaviors and trends. He joined Samsung in 1996 as a senior engineer in the Technology Development team, working on multiple nodes, and holds a PhD in Material Science and Engineering from Iowa State University. Over the years, he rose to the position of Executive Vice President at Samsung, where he was responsible for Samsung's U.S. Foundry business, focusing on attracting new customers and securing deals. Now, Intel Foundry will benefit from a significant new team member with decades of experience, deep expertise, and strategic relationships to help secure new customers. Shawn Han will report directly to Naga Chandrasekaran, Executive Vice President and General Manager of Intel Foundry, and will contribute to elevating Intel Foundry to new levels.

Intel today released its latest Arc GPU graphics drivers with the version 101.8724 WHQL bringing support for new hardware and new games. Yesterday, the company launched its Core Series 3 "Wildcat Lake" processors with up to six CPU cores and two Xe3 GPU cores. Expected to land in over 70 different laptop designs, Intel today enabled full gaming support with the Arc GPU driver update. In addition to new hardware, the new driver release adds game-ready support for Pragmata and fixes a lot of the old issues that persisted with the previous releases. These include fixes for flickering in Crimson Desert running DirectX 12, No Man's Sky experiencing corruptions on certain terrains, and importantly issues with Adobe Premier Pro crashing when exporting media based on HEVC codec has been fixed. For a complete picture of fixes and remaining issues, we have posted the changelog below.DOWNLOAD: Intel Arc Graphics Driver 101.8724 WHQL

Intel has officially unveiled its "Wildcat Lake" Core 300 series of processors for the entry-level PC segment. These CPUs are designed for value-oriented buyers looking for good CPU performance and basic GPU output in small form-factor commercial and edge AI PCs. Officially rated at 40 TOPS, these processors are capable of Copilot+ AI PC certification, meaning Intel has managed to provide basic local AI processing functionality to entry-level buyers. As part of the Core 300 series non-Ultra family, Intel's main selling point is all-day battery life in laptops with local AI processing. There are three segments, each featuring a hybrid core configuration, pairing two "Cougar Cove" P-cores with four LPE "Darkmont" cores.

The actual SoC package integrates two dies. The first and most important one, built on the 18A internal Intel node, features a 6-core CPU configuration, NPU 5 with 40 TOPS of INT8 data, Xe display and media engine, and a GPU that includes up to two Xe3 cores. This die also contains the memory controller and cache pool, supporting LPDDR5X memory running at 7,467 MT/s or up to 6,400 MT/s for DDR5. The memory-side cache includes 4 MB to help tasks "buffer" into this cache before and after accessing memory. Intel dedicates the second die to I/O handling, as the platform supports 6 PCIe Gen 4 lanes, two Thunderbolt 4 connections, two USB 3.2 connectors, and up to eight USB 2.0 connectors. For Wi-Fi, there is the Wi-Fi 7 standard, and Intel also includes Bluetooth 6.0.

Intel's upcoming "Crescent Island" series of GPUs, based on the newly developed Xe3P graphics IP, will be hitting the market this year. However, it seems this IP won't be featured in the desktop Arc gaming series of discrete GPUs. According to a reliable leaker, Jaykihn, Intel will not release this GPU generation as dedicated Arc gaming graphics cards. Instead, Xe3P-based GPUs will power "Crescent Island" data center and workstation GPUs, similar to the current Arc Pro series for professional users. This leaves Arc gamers without a clear indication of when, or even if, they will receive a new gaming Arc GPU based on the Xe3 or Xe4 IP in the coming months.

Intel recently announced the Arc Pro B70 and Arc Pro B65 graphics cards, which maximize the BMG-G31 "Battlemage" Xe2 silicon but are intended for professional users only. A recent driver update added the ability to play games on these cards, but they are still primarily designed for AI workloads and professional visualization. Gamers are still seeking clarity about future updates, and the lack of recent rumors regarding an additional Arc gaming discrete GPU is concerning. Even the higher-end Arc B770, which has been rumored for some time, seems to be on hold. Upcoming events like Computex might offer Intel a chance to at least provide a teaser of what the desktop Arc GPU family will look like or if there will be any updates soon.

Intel Foundry has been attracting significant attention from the industry lately, with some of the world's largest companies expected to announce new foundry commitments this fall, according to UBS Group. The investment bank has reported that Intel Foundry is on the brink of securing many new contracts scheduled for this fall, citing the release of version 1.0 of the 14A node process design kit (PDK) as a key catalyst. There have been numerous rumors that Apple, AMD, NVIDIA, Google, and Broadcom are considering utilizing Intel's silicon manufacturing, including the 18A, 18A-P, and 18A-PT nodes, as well as the upcoming 14A node. Apple is rumored to manufacture some of its M-Series "Apple Silicon" laptop processors in 2027 using the 18A-P node, while Google might leverage Intel's EMIB and Foveros 3D advanced packaging technology for some of its TPU designs.

Companies regularly evaluate leading industry nodes but often choose TSMC's manufacturing due to its reliability, high volume capacity, and advanced packaging, which has produced numerous high-performance and low-power designs at scale. However, Intel Foundry has been investing significant resources and logistics into attracting external clients, leading UBS to anticipate multiple foundry commitments being announced this fall. Late last year, we learned that Apple was waiting for Intel to release the 18A-P PDK version 1.0 or 1.1, scheduled for Q1 and Q2 of 2026, respectively. As we are now in Q2, we are awaiting further confirmation to see if Apple proceeded with the deal, but UBS expects it did.

NVIDIA today announced the world's first family of open source quantum AI models, NVIDIA Ising, designed to help researchers and enterprises build quantum processors capable of running useful applications. To achieve useful quantum applications at scale, significant breakthroughs are needed in quantum processor calibration and quantum error correction. AI is key for turning today's quantum processors into large-scale, reliable computers. Open models empower developers to build high-performance AI while maintaining total control over their data and infrastructure.

Named after a landmark mathematical model that dramatically simplified the understanding of complex physical systems, the NVIDIA Ising family provides high-performance, scalable AI tools for quantum error correction and calibration — two of the most critical challenges in building hybrid-quantum classical systems. Ising models run the world's best quantum processor calibration and enable researchers to tackle much larger, more complex problems with quantum computers by delivering up to 2.5x faster performance and 3x higher accuracy for the decoding process needed for quantum error correction.

Valve is implementing a number of new changes to its Steam platform, all aimed at enhancing the quality of the user experience for gamers. One of the latest updates involves adding a 30-day price history to all listed games. This feature will allow buyers to see if a game's price has changed recently, promoting greater transparency. Previously, gamers had to rely on third-party tools like SteamDB to track price fluctuations over time. Now, the Steam platform itself is set to include price tracking, integrating this functionality directly. This update will also highlight when a game is on discount, the percentage drop compared to its launch price, and how it compares to its price over the last 30 days. Gamers will be able to use these metrics to decide whether to wait for a Steam Sale or purchase the game immediately if it's a good deal.

Additionally, Valve is reportedly developing the SteamGPT AI system to assist with customer support queries, covering issues such as refunds, platform problems, and payment processing, among others. There is also talk of a new "Frame Estimator" tool that can predict your PC's performance before you buy a game. While Valve hasn't confirmed these features or provided a timeline for their release, these developments suggest preparations for future updates, which could be unveiled soon. Below is a sample price tracker from SteamDB for NBA 2K26 game, as we could see something similar on the main Steam platform soon.

Elon Musk announced today on X that Tesla has completed the design of a new generation of its AI chips for Full Self-Driving (FSD). The CEO also mentioned earlier that the AI5 chip will offer performance comparable to NVIDIA's "Hopper" architecture, with two AI5 units matching the power of a single "Blackwell" processor. In late 2025, reports indicated that Samsung had achieved a significant win for its previously struggling foundry business, as Tesla decided to split the manufacturing of its new AI5 accelerator between Samsung and TSMC. The chips will be produced at Samsung's plant in Taylor, Texas, and TSMC's facility in Arizona. This decision is part of a strategy to keep the supply chain diversified and maintain chip supply under control for any demand scenario.

Samsung and TSMC are not the only partners in this project. Tesla is also sourcing DRAM chips from SK hynix, which appear to be LPDDR5X memory integrated into the package. On both the left and right sides, there are two rows of SK hynix LPDDR5X memory modules, each with three modules. This totals 12 LPDDR5X memory modules per AI5 chip. With 16 GB per module, this results in 192 GB of LPDDR5X memory per single AI5 SoC.

In the latest Q1 2026 quarterly figures, ASML announced plans to ship over 60 EUV units this year, including both High-NA and Low-NA EUV lithography scanners. This is a significant increase from the 48 units shipped in 2025, but it doesn't break any records compared to the company's expectations for 2027. ASML predicts shipping about 80 EUV systems in 2027, driven by massive demand from memory makers. As memory has become the most sought-after commodity in AI data centers, there is simply not enough supply to meet demand. In Q1 2026, ASML reported that nearly half—45% to be precise—of its revenue came from South Korea.

While ASML hasn't provided specific data per manufacturer, we know that South Korean semiconductor makers primarily focus on memory, storage, and to a lesser extent, logic production. More than half of the quarterly shipments, at 51%, are dedicated to memory production. Companies like SK hynix and Samsung are purchasing many EUV systems to support their growing memory and storage businesses, which include the latest technologies like GDDR6, HBM3, HBM4, and DDR4/DDR5, all consumed by AI data centers worldwide. SK hynix plans to install 20 Low-NA EUV units in the next two years, all designed for HBM memory and advanced storage solutions, making SK hynix one of ASML's largest customers.

AMD's China store on JD has just started accepting pre-orders for the upcoming Ryzen 9 9950X3D2 Dual Edition desktop processor. Priced at an MSRP of $899, this model is $200 more expensive than the launch MSRP of the Ryzen 9 9950X3D. In China, AMD's store requires pre-orders with a deposit of 100 RMB, which counts as 300 RMB towards the final payment due to the "early bird" pre-order benefit. The final payment will be made on April 22, the planned retail launch date. Additionally, at launch, the processor will be available both as retail PIB (processor in box) packages and as part of pre-built gaming desktops by popular OEMs and system integrators. Currently, no Western stores offer pre-orders for this CPU, either in the United States or Europe. Most likely, these stores will start accepting orders on April 22. Some, like B&H, have a placeholder to notify customers when it becomes available.

The Ryzen 9 9950X3D2 Dual Edition is a 16-core, 32-thread processor built on the "Zen 5" microarchitecture and features 3D V-Cache technology, with stacked 3D V-Cache on both of its 8-core chiplets. This configuration results in an on-package L3 cache of 192 MB and a total cache (L2+L3) of 208 MB. In comparison, the 9950X3D has 3D V-Cache on only one of its two CCDs and relies on software-based OS scheduler guidance to ensure gaming workloads are directed to the CCD with 3D V-Cache. As a result, this new processor is more suited for productivity and workstation tasks, while the 9950X3D remains focused on gaming. Aside from the cache differences, the 9950X3D2 Dual Edition is mostly similar to the 9950X3D, except its maximum boost frequency is 5.60 GHz, which is 100 MHz lower than the 9950X3D, and it has a higher TDP of 200 W compared to the 170 W of the 9950X3D. While technically compatible with any Socket AM5 motherboard, support requires UEFI firmware updates from motherboard vendors.

During a media tour at Lexar HQ in China, Digital Foundry spoke with Lexar's Europe General Manager, Grace Su, who explained what gamers are buying and where they are willing to compromise. Interestingly, Lexar found that, despite the pricing of DRAM and NAND Flash, users could purchase lower-capacity RAM kits without issues, while SSDs with capacities lower than 512 GB weren't selling well. Although this is not a direct comparison, when building a new PC system, you need both a RAM kit and an SSD. Based on their budget, enthusiasts often have to decide between a larger SSD or a bigger RAM kit. However, Lexar reports that users are so committed to having larger SSD space that they are even willing to revert to HDDs rather than buy an SSD smaller than 512 GB. This is understandable, considering that modern games can take up hundreds of gigabytes of space.

Currently, high RAM prices are driving PC gamers to invest considerably more in their gaming rigs to play smoothly. As 16 GB is somewhat considered the minimum for a Windows 11 system, some games and tasks are so demanding that PC DIY enthusiasts need to invest in 32 GB kits at prices that were unimaginable just a year ago. For example, we reported on the DDR4 spot market pricing, which saw the cost of a single 16 Gb module increase by about 2,200% over the past year, with only a minor 5% correction in March. We can only hope for more corrections soon so that PC gamers can finally see signs of relief, but with AI demand projected to continue rising, that seems far from reality.

Yesterday, SemiAccurate posted a rumor that NVIDIA was planning to acquire a large PC maker, which would have reshaped its position within the PC space. However, it turns out the rumor is not true. NVIDIA issued a statement to CNBC saying, "The media report is false; NVIDIA is not engaged in discussions to acquire any PC maker." This brief statement effectively shuts down the rumors, suggesting that NVIDIA does not plan to go through with such an acquisition, even if it might have been considered in the past. Large corporations, especially with NVIDIA's $4.6 trillion market capitalization, are always looking to expand their business in areas where they are dominant and enter new markets where they could capture a significant share. NVIDIA holds a strong position in the dedicated GPU market, with about 94% market share among AIB GPU shipments.

The company is also looking to enter the laptop space with the upcoming N1/N1X SoCs, which would allow NVIDIA to tap into a completely new segment among PC users. In this context, acquiring a PC company would be a logical move, but it would present significant regulatory hurdles, as the company might not get approval from market regulators in the United States, European Union, and other regions. Now that we know the rumor is not true, we are waiting to see if any more rumors emerge about which company might have been the potential target. For now, we are refraining from making definitive claims or pointing fingers at any specific company.

Intel's upcoming Core Ultra 400 Series "Nova Lake-S" desktop processors are set to launch in the second half of this year, marking what seems to be the company's most ambitious lineup yet. Recently, new rumors have emerged suggesting that Intel is also working on a "Nova Lake-S" desktop APU featuring 12 Xe3P graphics cores. This configuration is expected to deliver graphics performance similar to the Arc B390 integrated GPU found in "Panther Lake," albeit with a slight twist in the overall GPU core design. We've already covered a recent leak detailing the entire Core Ultra 400 Series "Nova Lake-S" desktop lineup, but it seems Intel is developing and testing more in its labs than initially anticipated. Readers may recall brief rumors about a variant called "Nova Lake-AX," which was rumored to feature a single compute tile with eight "Coyote Cove" P-cores and 16 "Arctic Wolf" E-cores, along with a four-core LPE island, totaling 28 cores. The iGPU configuration for this variant was said to include 48 Xe3 cores, potentially making it one of the most powerful APUs ever.

However, since the "Nova Lake-AX" appears to be on hold, Intel is preparing other solutions with less powerful GPU configurations. According to a post by Jaykihn on X, the new desktop APU would feature four "Coyote Cove" P-cores, eight "Arctic Wolf" E-cores, and four LPE-cores. This 4+8+4 configuration would be paired with 12 Xe3P cores, which are slightly different from the regular Xe3 cores used in the standard "Nova Lake-S" desktop processors. Intel's lineup is becoming somewhat confusing, but it seems the main lineup will consist of standard desktop series split into several SKUs with high core counts and bLLC, or the newly rumored SKU with more Xe3P cores and fewer CPU cores, similar to AMD's Ryzen G-Series of desktop APUs. Do note that the image below is just an illustration, and not the actual render.

Apple has informed its supply chain that the company now aims to produce a total of 10 million first-generation MacBook Neo laptops, as consumer demand has been phenomenal. Initially, with the MacBook Neo launch, Apple expected consumers to purchase between 5 and 8 million units throughout the lifecycle of the first generation. However, since demand has exceeded initial expectations, Apple is significantly increasing production to meet this demand. As the Cupertino-based company has access to a wide network of manufacturing partners, ramping up production is straightforward, provided the main component—the A18 Pro SoC—is consistently supplied by TSMC.

Additionally, Apple is already planning a second-generation MacBook Neo with major upgrades to the overall system, primarily due to the new A19 Pro SoC, which will come with 12 GB of RAM. The current MacBook Neo features the mobile A18 Pro chip and is limited to 8 GB of RAM. According to recent rumors, Apple may upgrade the MacBook Neo's internals in 2027, equipping it with an A19 Pro, the same SoC found in the latest iPhone 17 Pro smartphones. Inside the MacBook Neo, Apple has opted to reuse the iPhone 16 Pro's chip, which is produced by TSMC and includes 8 GB of LPDDR5X memory. This memory is directly attached above the A18 Pro SoC using Integrated Fan-Out Package on Package (InFO-PoP) technology, creating a 3D wafer-level fan-out package.

Epic Games launched its Epic Games Store in 2018 to challenge the dominance of Valve's Steam platform in the PC gaming space. Known for its promotions and free games, the storefront has attracted a large gamer base that quickly adopted the platform. However, two former Epic Games employees claim that the Epic Games Store saw only an initial spike in active users during the free game promotions, with gamers eventually returning to Steam. Many gamers find switching between multiple storefronts to be a cumbersome experience. While free game deals are an excellent way to attract customers initially, the Epic Games Store needs to make further improvements to retain those gamers.

Meanwhile, Valve's Steam platform continues to break records for concurrent gamers. Just a few weeks ago, the platform reached 42,318,602 players on Sunday, March 22 of this year. This set a new record, surpassing the previous impressive 42 million concurrent players on January 11 earlier this year. At the time of writing, Valve's Steam platform has 37,053,591 active players, indicating that a large majority keep returning, with no signs of slowing down.

Valve's Linux developer, Natalie Vock, has introduced a new method for handling memory management on GPUs with 8 GB or less VRAM. Since a large part of the gaming community uses systems with limited resources, improving performance and VRAM allocation is an excellent way to maximize frames per second. According to Vock's solution, GPUs with 8 GB of VRAM or less on Linux will now benefit from better background VRAM management, allowing more free space for games. Several Linux kernel patches have been submitted to guide how Linux manages resources when a game is loaded. Previously, when a game used too much VRAM on a low-VRAM GPU, the Linux Graphics Translation Table (GTT) would move data to system RAM to prevent the game from crashing. Since Linux doesn't prioritize programs, it could sometimes move the actual game from VRAM to system RAM to make space for something like a browser or another third-party app.

In Vock's testing before these new patches, running Cyberpunk 2077 resulted in the system using only 6 GB of the 8 GB of available GPU VRAM, with about 1.37 GB spilling over to GTT and being allocated in system RAM. This could lead to game stuttering and frame-pacing issues. Now, Vock has developed a real-time dmemcg-booster, which stands for Device Memory Control Groups, to inform Linux when a program running on the GPU VRAM needs to stay there without being moved to GTT and then to system RAM. With the new patch, running Cyberpunk 2077 now uses almost the entire available system VRAM, reaching 7.4 GB, while GTT allocation has been reduced to 650 MB, cutting usage by more than half. In the picture below, you can check the "GameThread" behavior before and after the patch is applied.