AMD Unveils FSR 4: AI-Powered Upscaling Technology Exclusive to RDNA 4 GPUs

AMD doesn't want to repeat past mistakes with FSR 4

Summary FSR 4 will be available in over 75 games by year's end, exclusive to new RDNA 4 GPUs. FSR 4 uses AI for better image quality, similar to Nvidia's DLSS, improving fine details. FSR 4 shows up to 3.7x performance improvement, launching alongside new GPUs on March 6. AMD's roll out of FidelityFX Super Resolution 3 (FSR 3) was a bit of a nightmare, but it looks like Team Red has learned from past mistakes. The upcoming FSR 4, which is exclusive to new RDNA 4 graphics cards like the RX 9070 XT and RX 9070, is coming to over 30 games at launch, and AMD says it'll be available in over 75 games by the end of the year. Related 3 reasons why AMD finally used AI for FSR 4 AMD's FSR 3 wasn't far behind the competition, but FSR 4 needed sweeping changes for truly impressive gains Posts Better game support Game support was a major issue for FSR 3. It took nearly a year for FSR 3 to release after it was announced, and by the time it showed up, it was only available in two games - Forspoken and Immortals of Aveum, both of which had devastatingly low player counts by the time FSR 3 was added. Support for FSR 4 is already much better. AMD says it'll be available in games like Marvel Rivals, Call of Duty Warzone, Kingdom Come Deliverance 2, and God of War Ragnarok at release. In addition, AMD says it has partnered with developers like Insomniac, Sega, and Naughty Dog to include future support for the feature. FSR 4 improvements Close AMD is able to support a broad range of games so quickly because of how FSR 4 is designed. It's built on top of FSR 3.1, which is already available in dozens of games. FSR 4 looks like a drop-in replacement for these games. Compared to older versions, AMD says FSR 4 produces a higher quality image due to its use of AI. Unlike previous versions of FSR, FSR 4 requires dedicated AI accelerators that are available on RDNA 4 GPUs. Similar to Nvidia's Deep Learning Super Sampling (DLSS), FSR 4 leverages the AI accelerators to produce much better detail. You can see an example of that in Warhammer 40K: Space Marine 2 above. Fine details like the pillars on top of the tower flutter and disappear with the algorithmic upscaling of FSR 3.1, but those details are preserved with FSR 4. These small details are one of the main reasons DLSS still wins over FSR when looking at image quality. Although the results are impressive, it's important to wait until FSR 4 is actually here before drawing any firm conclusions. The pitfalls of upscaling usually show up when a game is in motion, and the stability of fine details plays a major role in how well the upscaling works. Thankfully, you shouldn't have to worry about performance. As you can see in the chart above, AMD claims you can see upwards of a 3.7x improvement with FSR 4 upscaling and frame generation in games like Ratchet and Clank: Rift Apart. It isn't long until we can see how FSR 4 performs in action, and critically, how it stands up to DLSS 4. The feature will launch alongside the RX 9070 XT and RX 9070 on March 6.

AMD's FSR 4 is exclusive to RDNA 4, 30+ games at launch, 75+ more coming in 2025

TL;DR: AMD's FSR 4, launching March 6, 2025, enhances image quality and performance using AI, exclusive to RDNA 4 and Radeon RX 9070/9070 XT. It supports over 30 games initially, with 75 more adopting it in 2025. FSR 4 offers improved image stability and detail but is not compatible with older Radeon cards. With FSR 4, AMD's answer to NVIDIA DLSS, the company embraces AI and machine learning to enhance image quality and performance. It builds on the upgradable API of FSR 3.1, so like DLSS 4, there will be a driver-level option to enable FSR 4 in games that feature FSR 3.1 via a simple .dll file swap. Exclusive to RDNA 4 and the new Radeon RX 9070 and Radeon RX 9070 XT, launching on March 6, 2025, FSR 4 will be available in over 30 games on day one. AMD notes that games will also feature native support for FSR 4 on day one, with over 75 more games set to adopt the new AI-powered Super Resolution upscaling in 2025. Compared to DLSS, this still leaves a big gap between the number of games you can enable NVIDIA's new DLSS 4 in and those with native or driver-level FSR 4 support. The good news is that the upgradable nature of FSR 4 means that updates can be applied to all games with existing support. However, games with FSR 2 or FSR 3 will not be compatible with the driver-level FSR 4 override, as only FSR 3.1 has this capability. FSR 4 uses a custom in-house hybrid FP8 AI model trained on powerful AMD Instinct hardware and requires the advanced new hardware inside the Radeon RX 9070 and Radeon RX 9070 XT to run - which feature up to 779 TOPS of AI performance. As the Radeon RX 9070 XT features double the AI performance of the Radeon RX 7900 XT, previous-gen Radeon cards will unfortunately miss out on FSR 4. With less ghosting, improved image stability, and more object detail, FSR 4 is a significant improvement over FSR 3.1. AMD notes that FSR 4's Quality, Balanced, and Performance modes will look better than their FSR 3.1 equivalents. In fact, during a pre-brief with media, AMD singled out Warhammer 40K Space Marine II as an example where AMD FSR 4's Performance Mode in 4K offers better image quality than FSR 3.1's 'Native AA' mode - an impressive result. It sounds like a worthy DLSS competitor and something that will add definite value to RDNA 4 GPUs. However, we won't know the whole story (how it compares to DLSS 4) until we launch and see FSR 4 in action. Here's a list of titles that will be FSR 4-ready on day one. As previously mentioned, FSR 4 differs from the platform-agnostic FSR 3.1 because it is exclusive to RDNA 4 and the new Radeon RX 9070 and RX 9070 XT GPUs. However, NVIDIA's DLSS 4 update, which arrived with the GeForce RTX 50 Series, is compatible with more GPUs - including GeForce RTX 20 Series cards from 2018. As fantastic and impressive as FSR 4 sounds, it took AMD a long time to switch to an AI-powered solution, so there will be some catching up to do.

'Infused with AMD DNA': FSR 4 has been announced with a healthy dose of machine learning and support for 30+ games at launch

After months of speculation we've finally had our first look at AMD's next generation RX 9070 XT and RX 9070 GPUs. But the fun doesn't stop there, as AMD has also taken this opportunity to tell us more about FSR 4, the latest version of its previously compute-based upscaler. The headline news here is that it's finally getting a much needed machine learning infusion that AMD claims will boost image quality significantly -- but, as we found out back in January, much like Nvidia's RTX 50-series-dependent DLSS 4 Multi Frame Generation you'll need a brand new card to use it. Both the RX 9070 and the RX 9070 XT come with fully-fledged, matrix operation-supported AI accelerators, and it's these that are required to power FSR 4. AMD says the machine learning upscaling models have been trained on AMD Instinct data center GPUs, and are accelerated by RDNA 4's architecture to deliver much improved image quality. DLSS has long been admired for its machine learning-based approach, and previous versions of FSR have paled in comparison to its remarkable image quality, particularly at Performance settings. Taking a look at the screenshots provided by AMD, however, it appears that machine learning may have given a similar boost to FSR. Fine details that FSR 3.1 often ignores in Performance mode, like the spires in the Space Marine 2 screenshot below, look much crisper. I'm particularly impressed by the improvements to fine details like the foliage highlighted in the bottom right of the second screenshot, as traditionally FSR has struggled with similar details. The performance gains look impressive when combined with frame generation, too. Space Marine 2 running at 4K with FSR set to Performance is said to gain a massive 3.5x frame rate uplift -- and if the pretty screenshots translate to smooth visuals in motion, well, it looks like many of my previous critiques of FSR may finally be solved. That's a fairly big 'if', though. I've had a fair bit of experience playing with previous versions of FSR, and all of them have a tendency to introduce a lot of unwanted noise to images in Performance mode that doesn't translate well to screenshots, but is fairly obvious when introduced to some fast-paced motion. The proof is in the playing here, so I'll be anxious to see it running for myself when I get a chance to mess around with the new GPUs. All of this is somewhat elementary if FSR 4 support isn't included in a large number of games, as we've all got grim memories of FSR 3's two supported releases at launch. This time, however, AMD says that not only are there 30+ games ready for FSR 4 inclusion at launch (including Kingdom Come: Deliverance 2, God of War: Ragnarok and Marvel's Spider-Man 2) but over 75 more coming this year. AMD also says that FSR 4 utilises the upgradeable FSR 3.1 API, which should help smooth over the transition. But keep your fork, because there's more: AMD Fluid Motion Frames 2.1 has also been announced as part of an improved version of AMD's driver-based HYPR-RX suite. AFMF 2.1 is said to feature improved frame generation image quality with reduced ghosting and better temporal tracking, and will run on RX 6000, 7000 and 9070-series GPUs, along with AMD Ryzen AI 300 series processors. So even if you plan on keeping an older AMD card, there's still something to celebrate here. The HYPR-RX suite also includes AMD Radeon Anti-Lag 2, which is designed to significantly reduce latency in supported games like Apex Legends, Counter Strike 2, DOTA 2 and Ghost of Tsushima. AMD will even include a latency measuring tool as part of the AMD Adrenaline drive suite, so you can see exactly how much quicker your responses are being registered -- although it won't explain why you still keep getting killed so often in-game. Probably. Anyway, AMD says that with all the HYPR-RX goodies enabled (including AFMF 2.1), you'll be getting average performance gains of 2.7x in 2025 compared to 1.5x in 2023. Big number goes up, etc, but it's nice to see these useful features getting upgrades alongside FSR 4. I mean, it's not quite as big a leap as Multi Frame Generation can provide, but at least progress seems to be being made on the opposing team. The real question here is whether FSR 4's alleged improvements translate into increased sales of the RX 9070-series when it becomes available to purchase on March 6. The obvious benefits of Multi Frame Generation have been a huge part of Nvidia's push for gamers to upgrade to the RTX 50-series GPUs, and while AMD still looks a step behind, some much improved upscaling and frame generation tools might help to swing the needle a little further in AMD's direction. It still feels somewhat behind the pace at this point when compared to the monumental achievements of DLSS, but as someone that's put up with the, let's say quirks of AMD's previous upscalers, I genuinely can't wait to see whether FSR 4's machine learning upgrades have finally made it a comparable solution. It's not long until I find out, but here's hoping it provides some serious competition, at the very least.

AMD FSR 4 mimics DLSS to offer up to 3.7x gaming performance

AMD finally shared technical specifications and pricing for its latest RDNA 4-based Radeon RX 9070 series GPUs. After a rather dull announcement at this year's Consumer Electronics Show (CES), the company today hosted a special online event where it spoke about architectural details and new features that we can expect from the RX 9070 and RX 9070 XT graphics cards. Alongside the launch of the Radeon RX 9000 series, AMD also announced key updates to its upscaling and performance-boosting technologies. FidelityFX Super Resolution (FSR) 4 represents a significant shift in AMD's approach to image upscaling and frame generation, while HYPR-RX receives refinements to further optimize gaming performance. FSR 4 marks a departure from previous versions by fully leveraging the AI accelerators in AMD's RDNA 4 architecture. Unlike FSR 3.1, which relied on traditional spatial upscaling methods, FSR 4 introduces AI-driven enhancements for improved image clarity and motion fluidity. However, this new iteration will initially be exclusive to RDNA 4 GPUs, with potential -- but uncertain -- compatibility for RDNA 3 and RDNA 2 at a later stage. AMD notably has plans to continue offering FSR 3.1. One of the key features of FSR 4 is its integration with frame generation, allowing the AI cores in RX 9000-series GPUs to generate additional frames, similar to Nvidia's DLSS frame generation. While AMD has not disclosed whether FSR 4 relies on a transformer-based neural network like DLSS 4 or a convolutional neural network (CNN), early image comparisons indicate a noticeable improvement over FSR 3.1. The update also comes with AMD's claim that it is neural rendering ready, hinting at future advancements that could integrate Microsoft's Cooperative Vectors. As per AMD's claims, FSR 4 with frame generation can offer a performance uplift of up to 3.7x. Additionally, FSR 4 will be compatible with existing games that support FSR 3.1, though how this will be implemented remains unclear. It may require manual activation through driver settings rather than automatic integration. At launch, over 30 games will support FSR 4, with AMD aiming to expand this to more than 75 titles by the end of 2025. As with all frame-generation technologies, real-world performance and responsiveness will depend on individual implementation. Recommended Videos HYPR-RX, AMD's one-click driver-level performance enhancement suite, is also receiving improvements. This feature consolidates multiple performance-boosting technologies, including FSR/RSR, Radeon Anti-Lag, Radeon Boost, and AFMF 2, offering an all-in-one solution for users looking to maximize FPS without manual tweaking. A key component of this update is AFMF 2.1 (AMD Fluid Motion Frames), which brings refinements to frame generation by reducing ghosting, improving fine details, and enhancing overlay detection. While HYPR-RX can provide noticeable gains in select games, performance improvements will vary based on individual titles and system configurations. Get your weekly teardown of the tech behind PC gaming ReSpec Subscribe Check your inbox! Privacy Policy AMD's FSR 4 and HYPR-RX updates signal a stronger push into AI-enhanced gaming technologies, placing them in direct competition with Nvidia's DLSS and Reflex ecosystem. While the exclusivity of FSR 4 to RDNA 4 GPUs may limit its initial adoption, its integration into existing FSR-supported titles could encourage developers to adopt the new standard. Meanwhile, HYPR-RX's improvements could offer a convenient way for casual gamers to optimize performance with minimal effort. As these technologies roll out, real-world benchmarks will determine their effectiveness and impact on the gaming landscape.

AMD FSR 4 & Fluid Motion Frames 2.1 Ready To Supercharge Your Games: AI Infused Performance For 75+ Titles In 2025

AMD has also introduced its brand-new FSR 4 & Fluid Motion Frames 2.1 technologies that increase image quality by leveraging AI & coming in 30+ titles at launch. AMD has been increasing its FSR support, which now has over 400+ titles supported. The journey started in 2021 with FSR 1 and has since seen various updates, such as FSR 2 in 2022, FSR 3 in 2023 and FSR 3.1 in 2024. All of these have brought in noticeable improvements with new technologies such as Frame Gen and an open integration commitment in various game engines. There was one area where FSR lacked behind versus the competitors such as NVIDIA's DLSS and Intel's XeSS and that was AI-based upscaling using ML engines. But today, AMD finally launches its ML-powered upscaler, FSR 4. With FSR 4, AMD is promising: For FSR 4, AMD has been training several custom game ML models on its Instinct accelerators, which are then integrated within the FSR 4 AI Upscaling Algorithm. This data is then passed onto Radeon RX 9000 "RDNA 4" GPUs, which offer up to 779 TOPS of AI acceleration to deliver high-quality and AI-enhanced upscaled 4K output. The scenes are then passed to the frame-gen algorithm, which generates several frames within the game and delivers the final result, which is smooth and great to look at. The result is a scene which looks even better than a Native 4K output even with the FSR 4 Performance mode. The quality is also very similar to FSR 3.1's native AA preset and that's big. AMD also shares some performance figures of FSR 4 against native 4K. The upscaling mode alone delivers a nice uplift, delivering up to 2x gains and an average uplift of 60%. With frame-gen on, the lead goes up to an average uplift of 2.8x and up to 3.7x in games such as Ratchet and Clank with RT enabled. You can comfortably achieve over 150 FPS with Frame-Gen enabled. As for support, AMD's FSR 4 will be available in 30+ games at launch and over 75 titles are planned to support FSR 4 throughout 2025. Furthermore, developers have the flexibility to integrate FSR 4 with relative ease into titles that already support FSR 3.1. The following is the list of titles available on the launch day: The following are the game studios which are supporting AMD FSR 4: AMD's HYPR-RX suite has also been a game-changer, especially on the mobility and handheld side of things. The platform features a one-click boost across thousands of games which can leverage technologies such as FidelityFX Super Resolution, Radeon Anti-Lag, Radeon Boost, Radeon Super Resolution and Fluid Motion Frames 2. Today, AMD is introducing an updated version of AFMF called Fluid Motion Frames 2.1 which adds the following changes: In comparison, AMD showcases how AFMF 2.1 reduces ghosting, offers finer feature restoration and adds better overlay handling. As for performance, you are getting up to 3x the performance boost when enabling HYPR-RX with Fluid Motion Frames 2.1 on a Radeon RX 9070 XT GPU. Enabling Frame Generation also leads to increased latency and, with Anti-Lag 2, you get up to 42% reduction in latency in upcoming titles such as Apex Legends. But that's not all. AMD is promising even more features coming to its HYPR-RX suite soon, so expect updates throughout the year. AMD is also adding a slew of features to its Adrenalin Software which would enhance the overall gameplay and display experience. One of these new features is Radeon Image Sharpening 2, which is now updated to provide stronger, more responsive sharpening in more use cases, apply sharpening to games, videos or across the entire desktop, and comes with a 1-click solution which is no longer dependent on apps or APIs and features cutting-edge efficiency designed to give more performance and power than the last version of RIS. AMD's brand-new and re-architected Encoder also delivers faster performance on the Radeon RX 9000 "RDNA 4" GPUs with improvements across the board in image quality from upgraded motion estimation and multi-frame reference while supporting up to 8K @ 75 FPS (HEVC/AV1). The AMD Adrenalin software suite also includes a range of new AI tools, such as the AMD AI Apps manager, AMD Install Manager, AMD Chat and AMD Image Inspector. The AI Apps manager can launch several AI Apps directly from the software windows and reports AI Apps usage while displaying select AI Apps that have been installed by the user. The Install Manager makes managing AMD software installations easier by setting up AI and PC ecosystems and automatically keeps user software up to date with customizable installations. The AMD Chat bot is a personalized AI agent which provides a range of functionality such as: The GPU-accelerated model is backed by support for ROCm v6.3 and Microsoft DirectML v1.15.2. One of the interesting implementations of the Adrenalin Suite is the AMD Image Inspector, which leverages AI to drive game quality improvements. The model is trained to detect various sorts of corruption within the game and, when enabled, it captures and sends diagnostic data (text and images) to AMD, which then helps enhance the User Experience. Some of the artifacts that can be detected by the Image Inspector include: So at launch, AMD will be offering the Adrenalin Edition 25.3.1, which comes with all of these features along with the latest optimizations and enhancements for Radeon GPUs.

RDNA 4 versus RDNA 3 - all the significant improvements as outlined by AMD

TL;DR: RDNA 4 is AMD's latest GPU architecture, powering the Radeon RX 9070 and 9070 XT, with significant advancements in AI and ray tracing. It doubles AI performance, introduces 3rd Gen Raytracing Accelerators, and enhances memory and shader management. The architecture supports improved video quality and high-resolution display capabilities, making it a substantial upgrade over RDNA 3. RDNA 4 is the latest gaming GPU architecture, powering the new Radeon RX 9070 and Radeon RX 9070 XT. It paves the way for Neural Rendering to become significant for gaming in the coming years. RDNA 4 doubles the AI performance compared to RDNA 3, with 2nd Generation AI Accelerators supporting FP8, which powers AMD's new custom Super Resolution model for FSR 4. Unlike previous versions of FSR, FSR 4 is exclusive to RDNA 4 as it leverages the latest AI hardware to deliver a massive improvement to image quality when compared to the non-AI FSR 3.1 - a big step to leveling the playing field with NVIDIA DLSS. The other big update arrives with RDNA 4's 3rd Generation Raytracing Accelerators, which doubles the ray-tracing throughput per Compute Unit with dual ray intersection calculations and more optimizations like compressing BVH sizes and new technology like 'Oriented Bounding Boxes' that improves performance at the hardware level. This is excellent news because ray tracing is one area where Radeon has fallen behind. AMD was clear to confirm that double the throughput doesn't mean double the performance, but it does mean that the Radeon RX 9070 XT will be a better RT GPU than even the RDNA 3 flagship Radeon RX 7900 XTX. With RDNA 4, AMD's mid-range GPUs will finally be able to deliver impressive ray-tracing performance. For those wondering about non-ray-tracing improvements, RDNA 4's architecture includes an enhanced memory subsystem with 3rd Gen Infinity Cache, improved scalar units, dynamic register allocation, increased efficiency, and higher clock speeds than RDNA 3. RDNA 4 introduces new shader reordering similar to NVIDIA's Shader Execution Reordering (SER) for the GeForce RTX 40 Series. This offers better memory management by allowing shader requests to be carried out more efficiently and not in some strict order. AMD notes that this will improve RDNA 4's performance in many workloads, not just ray-tracing. The flagship Radeon RX 9070 XT is built on an optimized monolithic design using TSMC's 4nm process. It sports 53.9 billion transistors on a 356.5 square mm die. This makes it comparable to NVIDIA's GeForce RTX 5080, which features a 378 square mm die size and 45.6 billion transistors. AMD is also significantly improving the Radeon Media Engine with 20-25% better video quality, more efficient playback, and more advanced encoder and decoder hardware. For content creators, RDNA 4 is a big step up over RDNA 3, with AMD ensuring that these improvements will immediately be felt when firing up the popular streaming and capture app OBS. The new AMD Radiance Display Engine includes DisplayPort 2.1 and HDMI 2.1b support for up to 4K 12-bit HDR at 480Hz and up to 8K 12-bit HDR at 165Hz. We'll dive into RDNA 4's architecture in more detail at launch, so stay tuned.

AMD RDNA 4 Architecture Deep-Dive: New Compute Units, Upgraded Raytracing Cores, AI Enhancements, & Path Tracing

AMD has finally unveiled the full architectural details of its next-gen RDNA 4 GPU architecture, which is built from the ground up for the Radeon RX 9000 series. AMD RDNA 4 Is A GPU Architecture Designed From The Ground Up For Gamers: New Compute Units, Ray Tracing & AI Cores, Ready For Path Tracing AMD's RDNA 4 architecture has been highly anticipated since the launch of the previous RDNA 3 and its upgraded RDNA 3.5 variation. While the RDNA 4 architecture isn't going to see any ultra-enthusiast SKUs, it does come with brand-new changes that should elevate gaming performance since it is designed primarily for gaming audiences. As such, AMD has brought the following new changes to RDNA 4: Compared to RDNA 2, the RDNA 4 GPUs see almost a 2x uplift in rasterization, close to 2.5x uplift in raytracing and a 3.5x uplift in ML (FP16 dense matrix) workloads per compute unit. So next, we dive into the building blocks of the RDNA 4 architectural block diagram to see how the entire chip comes together. RDNA 4's New Core IPs The core building block of the RDNA 4 GPU architecture is the Compute Engine. The new Compute Units come with Dual SIMD32 Vector Units and Enhanced Matrix Operations, which include: RDNA 4 also carries new shading improvements with RDNA 4 shades allocating registers dynamically. They can request registers from the pool when needed. They can release registers back to the pool when they complete that work, and the software manages the conditions when there's a wait time for an allocation. This results in better handling of memory latency while overall efficiency of the shared core can increase significantly. On the scalar unit side, you get new Float32 operations while scheduling updates include Split & Named barriers, Accelerated spill/fill operations, and improved instruction prefetch. Then we have the 3rd Generation Ray tracing units offering doubled ray intersection rates, improved BVH compression, accelerated ray traversal and shading, and Oriented Bounding Boxes. These new ray tracing cores offer one of the biggest performance increases on the chip. Each Ray accelerator has also been improved with: These new ray tracing upgrades also result in much lower memory requirements for BVH. On average, RDNA 4 reduces the memory requirements to less than 60% versus RDNA 3 thanks to the 8-wide design. But that's not it. AMD has also implemented a new solution to reduce traversal costs by encoding a rotation with each box to more tightly bound the contained geometry, while aligning the box to the geometry can help remove much of the space, and the ray direction is transformed on entry to the box to match the encoded rotation. This results in fewer traversal steps, a reduction in peak cost by eliminating traversal hotspots and an improvement in traversal performance by 10%. The result of these changes is that RDNA 4 CUs offer 2x ray traversal performance compared to RDNA 3 at equal clock rates and bandwidth. There's also an improved Command Processor which features enhanced packet accelerators. The Cache is also seeing an upgrade, which is now more balanced with up to 64 MB of 3rd Gen Infinity Cache, 8 MB of L2 cache and 2MB Aggregate CU cache. On the memory side, the RDNA 4 GPU architecture retains GDDR6 support but has been upgraded to faster speeds of up to 20.00 Gbps with up to 16 GB capacity alongside a 256-bit bus interface. RDNA 4 also employs enhanced memory compression techniques to lessen the stress on the available bandwidth. For AI, AMD is leveraging its 3rd Generation Matrix Acceleration engine which comes with improved Tensor Dense Rates, New 8b float data types, Structured Sparsity support and ML-based upscaling or Super Resolution. Compared to RDNA 3, the RDNA 4 CUs offer a 2x boost in image generation performance (SDXL 1.5) in a normalized scenario with FP16. The Media Engine moves to a dual-width design with updated Encode/Decode engines, up to 25% quality improvement in AVC, H.264, H.265, Double the AV1 throughput and is optimized for low-latency streaming. Finally, there's the updated Radiance Display Engine which now supports DisplayPort 2.1a, HDMI 2.1b outputs and an updated scaling and sharpening engine. The RDNA 4 Block Diagram (Top Navi 48 Die) Next, we move to the RDNA 4 block diagram which represents the full Navi 48 GPU SKU. RDNA 4 GPUs are fabricated on the TSMC 4nm process node and feature up to 53.9 Billion Transistors and the SKU measures 356.5mm2. The chip is also fully compliant with PCIe Gen5. Now it's time to break apart the RDNA 4 chip. The Navi 48 GPU (Radeon RX 9070 XT) is composed of four shader engines and each of those houses several "Dual Compute Units", not WGPs. Each Dual Compute Unit features two Compute units and there are a total of 8 DCUs or 16 CUs per Shader Engine. That's a total of 32 DCUs or 64 CUs on the chip itself for a total of 4096 stream processors or shader units. Each DCU has two Ray Accelerator engines for a total of 16 RAs per Shader Engine or 64 RAs in total, while each DCU also packs 4 Matrix Acceleration Engines for a total of 32 MAs per Shader Engine and 128 MAs in total. Each Shader Engine also packs four RB+ blocks, a rasterizer engine & a Prim Unit block. There are four sections of 3rd Gen Infinity Caches and four 4x16-bit memory controllers on the outskirts of the chip. The L2 caches are right in the middle of the GPU, which also includes two Geometry processors, two ACE units, and one each, HWS & DMA. The chip is connected using Infinity Fabric. The Path Tracing Future Ahead For AMD Raytracing is often seen as an outdated term in the PC gaming space. Sure, it's one form of tracing rays to make scenes look more realistic and has only started to gain traction in the console space, but the competition is often seen using a different type of ray tracer, called Path Tracing. While Ray Tracing uses a single primary ray to cast reflections, shadows, and refractions on a source, path tracing uses all possible paths of light and is a more expensive technique. NVIDIA's Path Tracing expertise can be seen in games like Cyberpunk 2077 or Alan Wake II, which are regarded as some of the most graphics-demanding titles, and also look absolutely stunning. It was made possible to use Path Tracing through new techniques such as upscaling and frame-gen, but the Green team also invested in a brand-new technology called ray reconstruction, which helps achieve path tracing more efficiently by removing the in-engine denoisers and using AI/ML to help re-evaluate and reconstruct the image. It looks like AMD is also following that approach with its own Neural Supersampling and Denoising technique for RDNA 4's Path Tracing capabilities. Upgraded Media & Display Capabilities We can't end this deep-dive without talking about the Media and Display Engines. So, to start it off, we first have the new Media Engines which offer enhanced game streaming and recording through: The Display Experiences have also improved with enhanced FreeSync Power Optimization modes that deliver lower idle power in most 2-display configs, a hardware flip queue support for offloading video frame scheduling to the GPU and saving CPU power for video playback, while Radeon Image Sharpening 2 delivers high-quality images and scenes and works across all APIs through a single toggle. That's all for the deep-dive, you can also check out our coverage of the AMD Radeon RX 9000 graphics cards and FSR 4 technology at the respective links.

AMD's FSR 4 will be getting its own version of DLSS Ray Reconstruction for Path Tracing

Cyberpunk 2077's impressive RT Overdrive Path Tracing Mode running on a Radeon RX 9070 XT with a playable frame-rate? Yes, it sounds like it's on the cards. Cyberpunk 2077's RT Overdrive mode screenshot captured on a GeForce RTX GPU, image credit: Hodi. Path Tracing or Full Ray Tracing is the future of lighting in games; it takes the promise of ray-tracing and applies the concept of tracing light rays and bounces around a scene to calculate and render realistic shadows, reflections, indirect lighting, global illumination, and more. If a single real-time ray-tracing effect like reflections is hardware intensive, then Path Tracing is like jumping ahead a decade. Path Tracing in games like Cyberpunk 2077, Indiana Jones and the Great Circle, and Alan Wake 2 is like seeing remastered versions of these games from 2030. They're playable on modern hardware only due to AI - specifically DLSS Super Resolution, Ray Reconstruction, and Frame Generation. NVIDIA's DLSS-powered Ray Reconstruction is a key reason why these titles look incredible, as the AI-powered denoiser can restore or render noticeably more detail in everything from reflections to texture detail and lighting. It's like going from HD to 4K, so it's fantastic to learn that AMD's RDNA 4 architecture and FSR 4 have been designed to support Path Tracing. With AMD's new AI-powered Neural Supersampling and Denoising, the company is ensuring that its new RDNA 4 GPUs - the Radeon RX 9070 and Radeon RX 9070 XT - will be ready for Path Tracing. It doesn't stop there as AMD's Neural Radiance Caching will provide a similar AI-powered approach to NVIDIA's new RTX Neural Shaders by leveraging AI to infer an infinite number of ray bounces in a scene to boost performance. Paired with the new FSR 4 upscaler and AMD's Frame Generation, it will level the playing field with DLSS - so to speak - regarding software support and cutting-edge rendering. Unfortunately, AMD hasn't confirmed if Neural Supersampling and Denoising will be available as part of FSR 4 on day one or when we might see it up and running in titles, so stay tuned. AMD presented its new Neural Supersampling and Denoising technology as part of a deep dive into RDNA 4's architecture for media using a single rendered frame of a scene that wasn't from a game.