All You Need for Gaming – AMD RDNA 4 and RX 9000 Series Reveal
For 25 years, AMD has powered your best gaming moments, driving the industry forward with pure, unrelenting performance. From the early days of first-person shooters, to the heart-pounding battles on the latest consoles, like Xbox & PlayStation, and handhelds like Steam Deck, AMD was there fueling your victories. Our commitment to the gaming community has always been strong. We advance adrenaline-fueled visual fidelity and push the limits with every generation of Radeon graphics, from the earliest breakthroughs to the most recent FSR launch, we've never stopped innovating. At AMD, gaming is at the core of our DNA, and the passion of the community is the fuel of our inspiration.
PC gaming captivates millions of players worldwide and has become a dominant force in the entertainment industry. It's about community, creativity, and the ever-expanding possibilities that gaming technology brings to the table. Whether you're gaming on a desktop, laptop, handheld, or console, AMD has been at the center of some of the most iconic gaming devices, from the early Microsoft Xbox 360 up to the latest Xbox Series X, the PlayStation 4 to the new PlayStation 5 Pro, or handheld gaming platforms like the Valve Steam Deck OLED, ASUS ROG Ally, or Lenovo Legion Go, even to the latest gaming PCs and laptops, AMD processors and graphics drive the heart of gaming. This year marks the 25th anniversary of Radeon graphics, which began with the R100 architecture offering hardware transform and lighting capabilities in an industry-leading 30 million transistor design. Over the years, we've poured our hearts and passion into shaping PC gaming and advancing the realism, immersion, and performance of Radeon graphics solutions.
We've transformed the render pipeline from separate pixel and vertex shaders to unified programmable shader units, with many groundbreaking products along the journey, like the much-loved Radeon 9700 Pro, to the first Teraflop GPU with the HD 4800 series, to the Radeon R9 Fury X, which was the first consumer graphics card with high bandwidth memory technology. In 2019, we introduced the advanced RDNA architecture with the RX 5700 XT, which is evolved to today's RDNA 3 RX 7000 designs, featuring hardware ray tracing and AI accelerators nearly 2,000 times more complex than our first Radeon GPU 25 years ago. Now, we're incredibly excited to share with you the next chapter in our Radeon journey. We began with some ambitious goals with RDNA 4, starting with an architecture that enables more players to access enthusiast-level gaming experiences. We focused on where gaming is going and built a significantly improved engine for ray tracing and machine learning technologies to meet the needs of games over the next several years.
And finally, we were laser-focused in every product decision to ensure that we deliver the best performance and capabilities that gamers want, while ensuring that we're accelerating our performance per dollar leadership. Today, we're going to take you through all the details of our next-generation graphics products, starting with the RDNA 4 architecture, with innovations to deliver enthusiast gaming experiences. We'll also share the software innovations built for our RDNA 4 products to enhance your gaming experience.
And, of course, we will dive deep into the Radeon RX 9000 series performance that we know you're excited to learn more about. First, let's hear from Andy, our lead for advanced graphics architecture, to tell us more about the heart of this new RDNA 4 graphics family. Thank you, Dave. It's great to be here today to talk to you all about RDNA 4.
There's a lot of exciting details to get through, so let's get straight into it. At its core, the Vision for RDNA has remained constant since we started in 2019. We always target the best possible performance that we can engineer for the workloads that will be important to gamers in the life of the architecture. But behind that simple statement lies so much complexity.
And as our architects sit down at the start of every new project, we have to navigate through all that we've learned and all that we see coming years into the future to drive our innovation and the areas of the architecture that we will focus on. From the Infinity Cache, high frequency and ray tracing that we drove in RDNA 2, to the advanced chiplet construction, matrix acceleration and physical design of RDNA 3, we aim to bring the very best of AMD's experience in computation, graphics and advanced construction techniques to bear on the problem of engineering the best products for our gaming customers. Now with RDNA 4, we are bringing the results of years of research and development to fruition, with the most powerful architecture we've ever designed for gaming graphics. We've targeted every step in the flow of rendering.
From the front end, with an enhanced command processor to more quickly handle the commands from the application, through the core where our updated compute units increase the efficiency of shading and double the performance of ray tracing and deliver even larger improvements in ML, to the memory interface where we improved our efficiency and compression. Our aim was to make your games run faster and smoother than ever before. Going beyond this, we made sure to also elevate other aspects of the visual experience for our users, with updates to our AMD Radiance Display and Media Engines. The streaming community has asked us for encoding quality improvements for popular services like Twitch, YouTube and Discord, and RDNA 4 delivers major improvements across H.264, H.265 and AV1. When using our H.264 encoder at 6000 kilobits per second, we are seeing up to 20% higher
V-math quality scores versus our previous generation. I'm incredibly proud of what the team has achieved, so let's take some time to dig in a bit further to what you as a gamer can expect from RDNA 4. Modern game rendering combines an ever expanding set of tools to deliver worlds of incredible detail and realism.
Rasterization has advanced to the point where we can explore incredibly rich worlds with near unlimited geometric detail, and modern techniques allow fully real time lighting, removing constraints in how dynamic or complex those worlds can be. AMD's philosophy is to embrace all of the techniques that graphics developers may use and build the best possible platform for them to deliver their vision to gamers. This can be seen in both our advanced RDNA 4 architecture and also our regular updates to our GPU open website, where we post our latest research and development for everyone to use, whether a game is using fast approximations to global illumination using compute shaders, or chooses to make use of our advanced ray tracing capabilities to go even further in quality. We have built RDNA 4 to adapt and deliver great performance.
The new compute unit includes a number of key optimizations for shading compared to RDNA 3. Perhaps the largest improvement lies in the new architecture for handling memory requests, allowing work to flow more efficiently through the machine for the complex shading in modern games. We've enhanced the register storage in the core with new dynamic register allocation techniques increasing efficiency, and we've added dedicated engines to accelerate register block moves to make function calls faster for modern shading languages.
We have also extended the instruction set of our scaler units, allowing our compiler to schedule more work on these units for better performance and lower power. Taken together, these changes make the compute units by far the most capable we've ever shipped in our consumer hardware. And the performance results that you will see with RDNA 4 hardware reflect that. Beyond these broad performance improvements, we also focused heavily on enhancing our ray tracing architecture. We attacked this problem at all levels, not only making RDNA 4 able to trace more rays more quickly, but also creating hardware that operates more intelligently to reduce the overall work.
Of the many advanced RT features introduced in RDNA 4, I will call out one key area here that drives the efficiency of ray traversal, which is the representation of the world geometry. The data structure that contains this geometry is called the bounding volume hierarchy, or BVH. In RDNA 4, we've taken an innovative approach to this with a new method for organizing that data that we call oriented bounding boxes. And we also develop better compression for the data itself, and these changes deliver much more efficient ray traversal through the geometry, and with a lower memory cost, making better use of capacity and bandwidth. Within the ray accelerator, we've added a second intersection engine, which doubles the performance for both ray box and ray triangle testing.
And based on our analysis of ray tracing workloads, we also created a dedicated ray transform block, which increases the performance of the hardware as rays are traversed into the lower levels of the BVH. With our updated ray accelerators, you will see a transformational change in the efficiency of ray tracing, making great performance available to more gamers at compelling price points, and delivering twice the ray tracing performance of RDNA 3 per compute unit, enabling smooth gaming experiences even with complex ray tracing effects. Now, let's take a look at our architectural advances for ML, as the impact of these technologies are being felt everywhere. With exciting announcements and new advances turning up on an almost daily basis, just recently we revealed a partnership with Sony Interactive Entertainment called Project Amethyst, where we've been collaborating for over a year now on research into ML model architectures for graphics enhancements.
And I'm happy to say we're already seeing results. Using RDNA 4's second generation of advanced AI accelerators, our users will be able to exploit all of the benefits of these new technologies, for gaming, content creation, or any of the other emerging areas that are appearing. RDNA 4 is able to efficiently process new advanced AI models much faster than was possible with previous generations of hardware, whether an AI model requires high precision with 16-bit floating point, or ultra-high performance with lower precision, such as 8-bit or even 4-bit, by adding more dedicated math pipelines and support for techniques such as structured sparsity. We have increased the peak performance per compute unit by as much as 8x compared to RDNA 3.
These improvements open a world of new possibilities for users of RDNA 4, and we're working now to bring technologies into games and other applications using these capabilities. So the performance is great. But what can RDNA 4 really do? Twenty years ago, we released a demonstration that showed the capabilities of our newest Radeon graphics when using Microsoft's DirectX 9 graphics API. This demo was called the toy shop, and I hope that back then it helped to open people's eyes to the kind of graphics that were now possible with advanced programmable shading. Now as we enter a new era of path-traced visuals combined with ML, we wanted to call back to this a little bit. So I would like to present the latest work of our graphics research teams being shown here for the first time in a new demonstration for RDNA 4, built with Unreal Engine 5 and enhanced with our latest techniques for neural radiance caching, super sampling and denoising.
And so, I give you our new toy shop. I hope that you enjoyed that demonstration. I am amazed at what the team has developed, and this is just a hint at what is to come in the future.
AMD is also working together with Microsoft to seamlessly integrate neural graphics techniques into future releases of DirectX that will light up the high-performance ML in our graphics hardware. As I wrap up, I wanted to end on one chart to give you a feel for the new levels of capability that RDNA 4 brings and the path of improvement that we've been on with RDNA as a whole. The journey of real-time graphics is just as exciting as ever, and we are not slowing down.
But of course, great hardware needs great software. And so to talk more about that, allow me to introduce Andrej, our Senior Vice President of GPU Technologies and Chief Software Officer. Thanks, Andy.
The evolution of Radeon Hardware has been accompanied by Radeon software innovations, unlocking the full potential of every AMD graphics card. We deliver complete and optimized experience, new features, more performance and robust ability. With the RDNA 4 family, we are leveraging AI everywhere, from hardware design to software feature development to enhancing our QA. With this vision in mind, our upcoming RDNA 4 adrenalin driver will set the new standard as the best launch driver in AMD history. Over the past few years, one of our top priorities has been software quality.
AMD has made significant progress in improving game compatibility, visual fidelity and driver reliability. To further enhance reliability, we have implemented AI-driven testing tools that identify issues early in the development. Our testing now covers a broader range of products and configurations, ensuring greater stability across different setups. However, we recognize that no testing environment can account for every unique user configuration.
This is why we introduce AMD Bug Report Tool, enabling users to submit issues directly to us and help us refine our products. Additionally, we launch a new beta tester program, inviting gamers and enthusiasts to try pre-release software and provide feedback. By working together with our community, we continue to push the boundaries of software quality and stability. Now, it's time to turn our attention to gaming performance.
To seamlessly bring great performance to more games, we have a number of technologies that enable gamers to get higher frame rates, smoother gameplay and a richer experience at any resolution. While FSR is an incredible in-game solution that delivers the best experiences, we know that there are thousands of other games out there where gamers want more performance. And this is where HYPR-RX comes in.
HYPR-RX automatically enables a suite of performance boosting and latency-reducing technologies. It works through the adrenalin driver and does not require deep integration into the game engine. Therefore, it is our most widely used AMD software feature available for more than 1,000 games. This year, we are adding new features for more performance and to improve image quality, with reduced ghosting and improved fine feature handling to make HYPR-RX even better. In some of the most demanding games at 4K, AMD HYPR-RX can deliver over hundreds of frame per second at 4K max settings.
In Star Citizen, gamers can experience up to 2.9 times better frame rates. Similarly, in Kingdom Come Deliverance 2, users can enjoy silky smooth game experiences at 4K resolutions with all settings turned on. It can also automatically enable latency-reducing technologies, like Radeon Anti-Lag, to reduce latency in games like Apex Legends by up to 42%. One of the fastest-growing software technologies is super-resolution or frame upscaling. In 2021, AMD released FidelityFX Super Resolution or FSR. This is a cutting-edge upscaling technology to improve game performance.
FSR has quickly become widely adopted across many games and platforms. Today, FSR is available in over 400 games, and our latest iteration, FSR 3.1, is already featured in over 70 games. The list continues to grow at an impressive rate, enabling all gamers to enjoy frame generation and upscaling. FSR has come a long way since introduction in 2021 as a spatial upscaler.
In 2022, AMD launched FSR 2, which incorporated temporal upscaling and motion vector data. This innovation dramatically improved image quality in motion-heavy scenes. Building on this momentum, we introduced FSR 3 in 2023, adding frame generation to interpolate frames and further boost frame rates.
Our latest version, FSR 3.1, had many quality improvements to reduce flickering and shimmering. But more importantly, FSR 3.1 introduced for the first time an upgradable FSR API. This was the basis to seamlessly unlock new capabilities in future versions of FSR, which brings us to today. AMD RDNA 4 owners will get access to a new version, FSR 4, featuring AMD machine learning powered upscaling, specifically developed for RDNA 4 architecture. We are delivering near-native image quality at four times lower resolutions.
We have improved ghosting, anti-aliasing, particles, fabrics, transparency effects, more effects. The result is much improved, near-native image stability with greater performance. In addition to these improvements, FSR 4 works seamlessly with our frame generation and latency reduction anti-lag technology. Everything we discussed is designed with gamers in mind. At the same time, we worked very closely with game developer community. Now we are taking it to the next level with an upgradable and flexible API that empowers developers to seamlessly integrate and deploy future FSR technologies.
FSR 4 can be enabled on all participating FSR 3.1 titles. This allows us to deliver FSR 4 on more titles sooner. This means FSR 4 is ready for the future of neural rendering with the same upgradable API.
FSR 4 upscaling is end-to-end AI solution developed on AMD consumer platforms with Ryzen processor and Radeon graphics. We trained our model using the latest AMD EPYC and Instinct accelerators and optimized it for the new FP8 ML acceleration in RDNA 4. Our new technology leverages a proprietary hybrid model resulting from extensive research across different types and combinations of neural networks and unique training techniques. FSR 4 uses the new FPA data type in the RDNA 4 architecture to balance quality and performance. The result is flexible solution that works with the existing FSR 3.1 framework.
Solution that enables a new ML option for upscaling alongside the frame generation and anti-lag technologies for a highly performant end-to-end solution. Here we are showing a close-up shot of Space Marine 2 running native 4K versus our previous generations of FSR 3. In the native mode you can see the series retain a lot of detail.
FSR 3 does a good job of retaining most of the detail but in a performance mode it does produce a softer image. By leveraging AI, FSR 4 is not only able to retain the details but even corrects the loss of details over the native TA mode. With that in mind, let's turn our attention to performance. In this video example we have Call of Duty Black Ops 6 running at max settings at 4K with variable rate shading turned off.
The minute FSR 4 and frame generation is enabled, frame rate increases 3 times. In the third person shooter, Ratchet and Clank: Rift Apart, FSR 4 delivers an incredible 3.7 times frame rate uplift with sharper visuals and in Warhammer 40,000 Space Marine 2 you see faster frame rate by enabling FSR 4. As you can see the resulting output is nearly indistinguishable from the native resolution. I'm happy to say that a lot of our partners and friends are excited about FSR 4. Gamers can enjoy the latest ML upscaling technology in over 30 titles at launch with over 75 titles planned by 2025.
Similar to the rest of our FidelityFX technologies, AMD will make FSR 4 for RDNA 4 available to all developers in the second half of this year, broadening the reach as much as possible. I'm sure that you are all excited to hear more about Radeon RX 9000 Series products. Let me turn things back to David.
Thanks Andrej. As we built the Radeon RX 9000 Series family, we focused on a number of imperatives. 85% of gamers purchased graphics cards below $700 and we wanted to deliver a family of products to level up the experience for more gamers than ever.
We wanted to build products designed for today's most popular games and the most anticipated games launching in the upcoming years. That means building for higher display resolutions as more gamers are playing at 4K and 1440p than ever before. But it's not just resolutions that are getting more demanding. New game engines and new technologies are pushing the settings used in the latest games. Ray tracing has become far more prevalent in the latest titles and gamers want to enjoy great frame rates with all of the new in-game technologies enabled.
And last but not least, we wanted to deliver a product family which makes the whole upgrade process as simple and seamless as possible. With reasonable power envelopes featuring standard 8-pin power connectors on many of our card designs and the ability to just plug and play into as many systems as possible. Today I'm so excited to introduce the Radeon RX 9000 Series graphics cards combining our next generation architecture along with the latest gaming technologies. With a significant performance increase over our prior generation, the Radeon RX 9000 series was built to deliver everything you need for 4K gaming at a 1440p price point. I'm thrilled to announce the Radeon RX 9070, the world's best $549 graphics card.
The 9070 is a great option for getting immersed in the latest titles with 56 RDNA 4 compute units, 16GB of memory, 2.5GHz boost clock, and over 1100 AI TOPS, all in an incredibly efficient 220W power envelope, starting at $549. Generationally at 4K resolutions, the 9070 is able to deliver on average 21% faster performance in over 30 of the most popular games, including both raster and ray traced titles.
Games like Cyberpunk and Dragon Age: The Veilguard are great examples where the 9070 is able to outperform the 7900 GRE by 25% or more. In ray tracing, those differences can be even greater, well over 30% in many cases. If you have a 1440p display, the uplift is similar at 20%, and all of this is with no special software enabled, just pure 1440p performance. For those gamers playing popular competitive esports titles at 1440p, the RX 9070 delivers great, ultra-high frame rates, meeting the capabilities of today's latest high refresh gaming displays.
In games like Counter Strike 2, Dota, you will see frame rates well over 300 FPS. But if you want to take your performance to the next level, then RX 9070 XT delivers even more, making it the world's best graphics card under $600. It delivers all you need for an ultra immersive gaming experience at 1440p and 4K. With 64 compute units, clock speeds boosting to nearly 3GHz, over 1,500 AI TOPS, all at a total board power of 304 watts, and best of all, a starting price point of $599. Looking at the same suite of games against the 7900 GRE, the 9070 XT delivers an awesome 42% gaming advantage at native 4K ultra. Again, this is pure gaming performance uplift with no additional software enabled.
But with the power of FSR4, the 9000 series graphics cards have the ability to deliver very high frame rates, well above 100 FPS, while maintaining near-native image quality. Compared to native 4K performance on the 7900 GRE, the 9070 XT with FSR4 can deliver 3-4X higher frame rates with incredible image quality and responsiveness. At 1440p, the performance of the 9070 XT is equally strong, averaging 38% faster gaming against the 7900 GRE. But gaming isn't the only place where the 9070 XT shines.
For creators, you'll see a nice step up in AI creator and generative AI workloads. With the 9070 XT outperforming the 7900 GRE by 40% and higher in generative AI and creator workloads. This is a massive leap forward in performance, advancing both your productivity and creativity. Now, let's look at the competitive performance of the Radeon RX 9070 XT at $599 versus the 5070 Ti at $749. The 9070 XT, powered by RDNA4, is able to deliver fantastic performance for $150 less. Across some of the most popular titles, the 9070 XT is the best choice for your next gaming PC when looking at raw gaming performance.
But when you factor in the difference in price point, the 9070 XT delivers an impressive gaming performance per dollar advantage over the competition's latest at 4K resolution. And if you're looking for the absolute best performance, our partners will feature a higher power 340 watt model with even more performance than a 5070 Ti. This makes the 9070 XT the most compelling upgrade for your system by far. The Radeon RX 9000 series is designed to deliver excellent gameplay at all of the key resolutions, especially 4K and 1440p, with next generation RDNA 4 compute units. Both cards feature 16GB of video memory, giving you more flexibility to increase in-game settings and play at higher resolutions.
Pair that with the latest DisplayPort 2.1A and HDMI 2.1B technologies and you're ready to drive the latest ultra-high refresh monitors. With standard 8-pin power connectors across most designs, the RX 9070 series is an effortless drop-in upgrade for your system. And with PCIe 5 capabilities, they're a perfect match with the latest Ryzen 9000 series processors and the AMD AM5 platform. The combination of Ryzen and Radeon, along with the advanced suite of AMD gaming technologies, ensures that you're ready to dive into your favorite titles, both today and tomorrow. Now looking at one of those future titles, our friends at Pearl Abyss are hard at work on their new game Crimson Desert.
And we're really excited about this open-world action adventure featuring incredible graphics, ray traced visuals and FSR4. Crimson Desert is expecting to launch this year and looks incredible on the latest AMD gaming hardware. Let's check in with Pearl Abyss on their progress. Wow.
Thank you, Kyung In Jung and the team at Pearl Abyss for the innovation you're bringing to PC gaming. I'm so excited to see what your team is doing and how you're advancing the Ryzen and Radeon gaming experience. With the latest Ryzen 9000 X3D processors and the all-new Radeon RX 9000 series GPUs, you'll be ready for your favorite games of today and tomorrow. Alongside our new FSR4 gaming technology, AMD has some of the best options for you with our 9000 series solutions. Our partners have been shipping these cards into retailers and e-tailers to ensure broad availability at launch.
I'm happy to say both the 9070 XT and 9070 will be available for sale starting March 6th. And in the second quarter of this year, we'll broaden the family even further with the RX 9060 products, which we'll tell you more about later. From all of us, thank you for joining us today and thank you for being a part of the AMD gaming community.
2025-03-03 00:29
