AMD Unveils UDNA: A Unified GPU Architecture to Challenge NVIDIA's CUDA Dominance

AMD announces unified UDNA GPU architecture: combines RDNA and CDNA to compete against CUDA

Back in 2020, AMD announced it was splitting its post-GCN architecture into RDNA for gaming, with CDNA for its data center GPUs, with CDNA later being the architecture of its Radeon Instinct AI accelerators... and now, they're merging into UDNA. In a chat with Tom's Hardware, senior vice president and general manager of the Computing and Graphics Business Group of AMD, Jack Huynh, said: "So, part of a big change at AMD is today we have a CDNA architecture for our Instinct data center GPUs and RDNA for the consumer stuff. It's forked. Going forward, we will call it UDNA. There'll be one unified architecture, both Instinct and client (consumer). We'll unify it so that it will be so much easier for developers versus today, where they have to choose and value is not improving". AMD simplifying into the UDNA architecture means we'll see a future where developers only need to focus on a single system, no matter if they're building next-gen AI GPU architectures, or next-gen GPU architectures for Radeon in the form of RDNA 5 in the future. AMD is now planning 3 entire generations ahead: RDNA 5, UDNA 6, and UDNA 7 with a goal of maintaining optimizations without requiring a reset reach time AMD changes its memory hierarchy... some positive changes for the GPU business inside of AMD. There is no UDNA plan rolled out to the public by AMD yet, so we don't know if this fusion of its GPU architectures will start with RDNA 5... but it would be the perfect point to start this new venture. RDNA 4 will only be aimed at the mid-range on the consumer Radeon GPU side, with a clean-sheet "Zen moment" with RDNA 5... UDNA sounds perfect for that. Let's go, AMD.

AMD To Integrate Consumer & Datacenters GPUs Into Unified "UDNA" Architecture, RDNA & CDNA Coming To An End

AMD is merging its consumer "RDNA" and Data Center "CDNA" GPUs under one unified architecture with the announcement of UDNA. AMD Plans To Bring Some Serious Competition In Next-Gen Markets, Merges RDNA & CDNA Families Into UDNA Architecture For The Future Team Red is transitioning its business model to something much more massive, likely targeting a larger market share. AMD has looked determined towards its upcoming RDNA 4 lineup, as this is not just based on the firm's statements; previous leaks have indicated that the firm has "sharpened the blade" and is ready to hit the mainstream markets more aggressively than ever. In a statement by AMD's Jack Huynh, it is now revealed that the firm plans to merge its consumer GPUs and data center architectures to increase market relevance. The "unified" approach isn't entirely new for AMD, given that the firm had its product portfolio revolving around the GCN architecture until 2019, after which the RDNA & the CDNA architectures came into existence. This was an attempt to navigate its way through the newly emerged markets, and having separate lineups meant the development was consistent across both platforms. Still, in modern times, the actual race lies in being the very best, and with having a single architecture, AMD plans to increase development massively. So, the new architecture will now be called "UDNA," which is said to reshape AMD's business structure. Interestingly, NVIDIA employs a similar approach with its modern-day consumer and AI GPUs, such as the "Blackwell" architecture, where the AI lineup and the rumored GeForce RTX 50 series GPUs are named under a single label. So, part of a big change at AMD is today we have a CDNA architecture for our Instinct data center GPUs and RDNA for the consumer stuff. It's forked. Going forward, we will call it UDNA. There'll be one unified architecture, both Instinct and client [consumer]. We'll unify it so that it will be so much easier for developers versus today, where they have to choose and value is not improving. So, going forward, we're thinking about not just RDNA 5, RDNA 6, RDNA 7, but UDNA 6 and UDNA 7. We plan the next three generations because once we get the optimizations, I don't want to have to change the memory hierarchy, and then we lose a lot of optimizations. So, we're kind of forcing that issue about full forward and backward compatibility. We do that on Xbox today; it's very doable but requires advanced planning. It's a lot more work to do, but that's the direction we're going. - AMD's Jack Hunyh via Tom's Hardware Team Red says that UDNA's naming scheme will sound something like "UDNA 6 and UDNA 7", and the firm is working on backward compatibility with existing RDNA and CDNA architectures as well. One important factor of having a unified architecture is how accessible the UDNA platform will be to developers, and in the longer term, Huynh believes that they plan to compete with NVIDIA's CUDA, eventually leveling its open-source ROCm software stack, but this is uncertain for now, given that the firm has a lot of stuff to catch up on, both on hardware and software levels. Just recently, AMD also announced that they are looking at focusing on GPU scalability, which means that the firm is eager to target a wider segment of the markets rather than being involved in the race for flagship GPU models. So, while the firm does look to be heading in the right direction, we can't conclude how these decisions pan out for AMD, given that the markets are often uncertain and full of surprises.

'From the developer's standpoint, they love this strategy' -- AMD's plan to merge its RDNA and CDNA GPU architectures to a unified system called UDNA

Ever since it bought GPU firm ATI Technologies in 2006, AMD has experimented with all kinds of graphics architectures. Its current approach, one for PC clients and the other for data and AI servers, will be swapped in favour of a completely unified architecture, bringing the best of both systems into one developer-friendly architecture. That's according to AMD's senior vice president, Jack Huynh, who announced the company's GPU plans in an interview with Tom's Hardware. At the moment, there's not an awful lot of detail as to what the plan will entail or when we can expect to see the first products based on the architecture, but we do know what it will be called, and it's UDNA. Most of today's PC and console gamers will only be familiar with AMD's RDNA design, which first appeared in 2019 with the Radeon RX 5700 XT. It was a comprehensive overhaul of AMD's GCN (Graphics Core Next) architecture, greatly improving the IPC, efficiency, and overall performance. From a game developer's point of view, it was a much-needed improvement as the design was very focused on gaming. GCN dates all the way back to 2012 and despite evolving through multiple iterations, it was still fundamentally the same compute-focused design. In many ways, when it first appeared, it was an architecture ahead of its time but as games became increasingly reliant on compute shaders to handle the bulk of the rendering workload, the issues with GCN (such as its low IPC) began to hold things back. However, AMD didn't abandon it altogether when it turned to using RNDA in its GPU. Instead, it was reborn as CDNA and the latest revision of it is used to power the likes of the enormous Instinct MI300 accelerator. Nvidia took a somewhat different approach, though, and for a very long time used the same architecture in its gaming chips as it did for its workstation and server processors. That's still the case today, although there are some differences between client-focused and server-focused GPUs, such as the amount of L1 and L2 cache. Fundamentally, though, and certainly in terms of software, Nvidia's designs are pretty much the same. So it's not too surprising that AMD feels the time is right for it to merge its GPU architectures into one system that's equally at home in any application. Talking to Tom's Hardware, Huyhn said "[s]o, part of a big change at AMD is today we have a CDNA architecture for our Instinct data center GPUs and RDNA for the consumer stuff. It's forked. Going forward, we will call it UDNA. There'll be one unified architecture, both Instinct and client. We'll unify it so that it will be so much easier for developers versus today, where they have to choose and value is not improving." One significant change this should bring is the implementation of dedicated matrix-calculation cores in all AMD GPUs. Both Intel and Nvidia have these already (the former calls them XMX AI Engines, whereas the latter calls them Tensor cores) and they're put to good use in upscaling and frame generation. While current RDNA GPUs can also do such calculations, they're done on the standard shader units; only the hulking CDNA GPUs have matrix cores. But if you're hoping to see UDNA soon then you're going to be disappointed. When Tom's Hardware asked them how long it would take, Huynh replied "[w]e haven't disclosed that yet. It's a strategy. Strategy is very important to me. I think it's the right strategy. We've got to make sure we're doing the right thing." I doubt that AMD has only just made this decision but even if it started the ball rolling at the beginning of this year, it could take two or three years before we see a UDNA-powered Radeon graphics card. At the very least, we'll have a generation of RDNA 4 GPUs and probably RDNA 5 before UDNA makes an appearance. Perhaps the first platform to showcase UDNA won't be a graphics card but rather a console. A healthy chunk of AMD's revenue comes from selling its semi-custom APUs to Microsoft and Sony, and the Series X/S and PlayStation 5 were the first chips to sport AMD's ray tracing technology. A fully unified GPU architecture should make it easier for developers to create games that work just as well on a console, as they do on a handheld gaming PC or full desktop rig. With luck, AMD will share more details about the plan in the coming months.

AMD will unify Radeon and Pro graphics to battle Nvidia's dominance

AMD plans to unify the architectures used by its datacenter and consumer graphics architectures in what one company executive is calling a "new DNA," or a "unified DNA," that combines its CDNA and RDNA architectures. The goal is to bring more developers into the fold, a complement to AMD's adjustment to its consumer Radeon GPU plans to attack mainstream but not flagship enthusiast products. Like Steve Ballmer's famous push to land developers, so too is AMD trying to land a core group of software developers that can write apps for Radeon silicon. If AMD concentrates on the small number of customers who will buy a flagship GPU, it won't attract as many developers as pushing mainstream products, Jack Huynh, AMD's senior vice president and general manager of Computing and Graphics told a Friday breakfast meeting of reporters at the IFA show in Berlin. Huynh first called the new unified effort "new DNA," but also referred to it as a unified architecture, or UDNA. "Part of the big change of AMD is today we have a CDNA architecture [for datacenters] and RDNA for the consumer side," Huynh said. "Going forward, we'll call it New DNA. There will be one unified architecture, both Instinct [AMD's enterprise GPU architecture] and commercial side, so that we unify it. It will be so much easier for developers. Today they have to choose, and it's not accruing value. "The reason why we forked it, because then you get the sub-optimizations and recommended issues, but then it's very difficult for these developers, especially as we're growing our data business. I mean, now we may introduce it to unify it. That's going to help it quite a bit. Because remember, I said earlier, I'm thinking about millions of developers, that's where we want to get to." The process will take time, Huynh said. "But that's why I'm telling the team right now," he said. "It's that scale we have to build now." Huynh said that AMD had made "mistakes" on the development of RDNA. "Each time we change the memory of a subsystem, that has to reset the matrix on the optimizations," he said. "I don't want to do that." When developing future generations of RDNA, Huynh said that he didn't want to change the memory hierarchy and then lose all of the optimizations. "It's very doable, it just requires advanced planning, and a lot more work to do it. But that's the direction." Toms Hardware noted that AMD's consumer GPUs have lacked strong AI support, and that the new revisions might allow AMD to more directly attack Nvidia's CUDA architecture and DLSS. Nvidia has been a favored architecture in consumer AI generation, such as AI art, both because of CUDA's deep developer support as well as the sheer horsepower of Nvidia GPUs. But when will this happen? And will it be effective? This feels like a plan AMD will begin explaining more in the weeks ahead.

AMD reveals plans to unify its data center and consumer GPU architectures - SiliconANGLE

AMD reveals plans to unify its data center and consumer GPU architectures Advanced Micro Devices Inc. will unify the architectures that power its data center and consumer graphics cards. Jack Huynh, the executive who leads the AMD unit that develops the chips, disclosed the plan in a Tom's Hardware interview published today. AMD offers two sets of graphics processing units. The first product line, which targets the enterprise market, is based on an architecture called CDNA 3. The architecture is optimized to run artificial intelligence software with a particular emphasis on large language models. Some of the calculations that an AI model performs while making a decision involve data points with a value set to zero. CDNA 3 can compress such data into a more compact form that takes less GPU capacity to process, which speeds up calculations. The architecture also includes several other optimizations designed to speed up AI workloads. AMD's consumer graphics cards, in turn, are geared towards running video games rather than LLMs. They're based on an architecture called RDNA 3. One of the features that sets the design apart from CNDA 3 is its better support for ray tracing, a technique used by many video games to render lighting and shadow effects. In today's interview, Huynh detailed that AMD plans to merge RDNA and CDNA into a single chip architecture. That future technology is set to underpin all the company's GPUs. He cited three main reasons for the decision to adopt a unified design. The first reason is that developing one GPU architecture instead of two will enable AMD's engineering organization to operate more efficiently. According to Huynh, another factor behind the change in direction is that RDNA, the chipmaker's consumer GPU architecture, has certain design limitations. Those limitations have made it challenging for AMD's engineers to upgrade the architecture's memory components. Huynh said that a unified chip architecture will benefit third-party developers as well. Optimizing an application for one chip architecture is simpler than for two, which should ease software teams' work. Huynh also provided an update about AMD's plans for the top-end consumer GPU market. The segment is currently led by rival Nvidia Corp., which has an estimated 88% share of chip shipments. AMD accounts for the remaining 12%. According to Huynh, AMD will prioritize lower-cost graphics cards going forward. The company plans to grow its presence in that market before switching its focus back to competing with Nvidia's top-end consumer GPUs. Huynh explained that the decision has to do with developer adoption. The level of demand for a consumer GPU is heavily influenced by the number of video games that are optimized to run on it. AMD believes that increasing its share of the lower-end GPU market will convince more developers to optimize their games for its chips. Once it has built that go-to-market foundation, the chipmaker will be in a better position to take on Nvidia in the top-end GPU segment.

AMD announces unified UDNA GPU architecture -- bringing RDNA and CDNA together to take on Nvidia's CUDA ecosystem

Here in Berlin, Germany, at IFA 2024, AMD's Jack Huynh, the senior vice president and general manager of the Computing and Graphics Business Group, announced that the company will unify its consumer-focused RDNA and data center-focused CDNA architectures into one microarchitecture, named UDNA, that will set the stage for the company to tackle Nvidia's entrenched CUDA ecosystem more effectively. The announcement comes as AMD has decided to deprioritize high-end gaming graphics cards to accelerate market share gains. When AMD moved on from its GCN microarchitecture back in 2019, the company decided to split its new graphics microarchitecture into two different designs, with RDNA designed to power gaming graphics products for the consumer market while the CDNA architecture was designed specifically to cater to compute-centric AI and HPC workloads in the data center. Huynh explained the reasoning behind the split in a Q&A session with the press and the rationale for moving forward with a new unified design. We also followed up for more details about the forthcoming architecture. Here's a lightly edited transcript of the conversations: Jack Huynh [JH], AMD: So, part of a big change at AMD is today we have a CDNA architecture for our Instinct data center GPUs and RDNA for the consumer stuff. It's forked. Going forward, we will call it UDNA. There'll be one unified architecture, both Instinct and client [consumer]. We'll unify it so that it will be so much easier for developers versus today, where they have to choose and value is not improving. We forked it because then you get the sub-optimizations and the micro-optimizations, but then it's very difficult for these developers, especially as we're growing our data center business, so now we need to unify it. That's been a part of it. Because remember what I said earlier? I'm thinking about millions of developers; that's where we want to get to. Step one is to get to the hundreds, thousands, tens of thousands, hundreds of thousands, and hopefully, one day, millions. That's what I'm telling the team right now. It's that scale we have to build now. Tom's Hardware [TH], Paul Alcorn: So, with UDNA bringing those architectures back together, will all of that still be backward compatible with the RDNA and the CDNA split? JH: So, one of the things we want to do is ...we made some mistakes with the RDNA side; each time we change the memory hierarchy, the subsystem, it has to reset the matrix on the optimizations. I don't want to do that. So, going forward, we're thinking about not just RDNA 5, RDNA 6, RDNA 7, but UDNA 6 and UDNA 7. We plan the next three generations because once we get the optimizations, I don't want to have to change the memory hierarchy, and then we lose a lot of optimizations. So, we're kind of forcing that issue about full forward and backward compatibility. We do that on Xbox today; it's very doable but requires advanced planning. It's a lot more work to do, but that's the direction we're going. PA: When you bring this back to a unified architecture, this means, just to be clear, a desktop GPU would have the same architecture as an MI300X equivalent in the future? Correct? JH: It's a cloud-to-client strategy. And I think it will allow us to be very efficient, too. So, instead of having two teams do it, you have one team. It's not doing something that's that crazy, right? We forked it because we wanted to micro-optimize in the near term, but now that we have scale, we have to unify back, and I believe it's the right approach. There might be some little bumps. PA: So, this merging back together, how long will that take? How many more product generations before we see that? JH: We haven't disclosed that yet. It's a strategy. Strategy is very important to me. I think it's the right strategy. We've got to make sure we're doing the right thing. In fact, when we talk to developers, they love it because, again, they have all these other departments telling them to do different things, too. So, I need to reduce the complexity. [...]From the developer's standpoint, they love this strategy. They actually wish we did it sooner, but I can't change the engine when a plane's in the air. I have to find the right way to setpoint that so I don't break things. [End of Huynh's comments] Yes, high-end silicon can build markets, but ultimately, software support tends to declare the winners and losers. Nvidia has taught the master's class of how to build a seemingly impenetrable moat with its unparalleled proprietary CUDA ecosystem. Nvidia began laying the foundation of its empire when it started with CUDA eighteen long years ago, and perhaps one of its most fundamental advantages is signified in the 'U' in CUDA, the Compute Unified Device Architecture. Nvidia has but one CUDA platform for all uses, and it leverages the same underlying microarchitectures for AI, HPC, and gaming. Huynh told me that CUDA has four million developers, and his goal is to pave the path for AMD to see similar success. That's a tall order. AMD continues to rely on the open source ROCm software stack to counter Nvidia, but that requires buy-in from both users and the open source community that will shoulder some of the burden of optimizing the stack. Anything AMD can do to simplify that work, even if it comes at the cost of some micro-optimizations for certain types of applications/games, will help accelerate that ecosystem. AMD has taken its fair share of criticism for the often scattered efficacy of the ROCm stack. When it bought Xilinx in 2022, AMD even announced that it would put Victor Peng, the then-CEO of Xilinx, in charge of a unified ROCm team to bring the project under tighter control (Peng recently retired). That effort has yielded at least some fruit, but AMD continues to receive criticism for the state of its ROCm stack -- it's clear the company has plenty of work ahead to fully put itself in a position to take on Nvidia's CUDA. The company also remains focused on ROCm despite the emergence of the UXL Foundation, an open software ecosystem for accelerators that is getting broad support from other players in the industry, like Qualcomm, Samsung, Arm, and Intel. What precisely will UDNA change compared to the current RDNA and CDNA split? Huynh didn't go into a lot of detail, and obviously there's still plenty of groundwork to be laid. But one clear potential pain point has been the lack of dedicated AI acceleration units in RDNA. Nvidia brought tensor cores to then entire RTX line starting in 2018. AMD only has limited AI acceleration in RDNA 3, basically accessing the FP16 units in a more optimized fashion via WMMA instructions, while RDNA 2 depends purely on the GPU shaders for such work. Our assumption is that, at some point, AMD will bring full stack support for tensor operations to its GPUs with UDNA. CDNA has had such functional units since 2020, with increased throughput and number format support being added with CDNA 2 (2021) and CDNA 3 (2023). Given the preponderance of AI work being done on both data center and client GPUs these days, adding tensor support to client GPUs seems like a critical need. The unified UDNA architecture is a good next logical step on the journey to competing with CUDA, but AMD has a mountain to climb. Huynh wouldn't commit to a release date for the new architecture, but given the billions of dollars at stake in the AI market, it's obviously going to be a top priority to execute the new microarchitectural strategy. Still, with what we've heard about AMD RDNA 4, it appears UDNA is at least one more generation away.