AMD RDNA 4 Rumors Revealed
AMD’s RDNA 4 is rumored to be a significant upgrade over the current RX 7000 series, especially at the higher end. According to MLID’s source, the GPU will feature 48 CUs and use GDDR7.
Apparently, vector access latency will be lower as well, since the L0 vector cache is 32-way set associative compared to 4-way scalar caches. This will also reduce cache misses.
RDNA 4 is the next-generation graphics architecture that will power AMD’s Radeon 8000 series GPUs. It’s expected to be released sometime in 2024 and should offer significant performance improvements over the previous generation. During a recent interview with 4Gamer, AMD senior executives including the head of their Radeon Technologies Group talked about the development process behind this new architecture and what it can do for gaming.
One of the most interesting bits of information that came out of this is that they are focusing on optimizing the GPU compute pipeline with RDNA 4. It will also get a major upgrade with a 3rd-generation Infinity Cache and a new data prefetch system. This should help improve performance by up to 50% over the current gen. Additionally, they are working on a new matrix instruction set that will increase performance per CU by up to 2x.
Another area that could see improvements is the command processor and geometry engines. These will feature specialized box sorting algorithms to reduce instructions needed for BVH traversal as well as stack optimizations. Additionally, they are working on a WMMA V2 system that should reduce the number of cycles needed for matrix operations.
These are just some of the many improvements that are being worked on with RDNA 4. In addition, there is a new Dual Media Engine that supports AV1 encode/decode and AI-enhanced video decoding. The Radiance Display Engine is also getting some love with support for DisplayPort 2.1 UHBR 13.5 which can deliver up to 54 Gbit/s of bandwidth for high refresh rates and 4K/8K resolution.
It is unclear whether or not AMD will continue to use its GCN chiplet design with RDNA 4. However, they will likely move to TSMC’s 7nm node for this generation of GPUs to keep up with Nvidia’s RTX 5000 series.
AMD’s RDNA architecture has already proven itself in the market with a series of high-performance graphics cards that offer a competitive price-per-performance ratio. The RX 5700XT, for instance, shattered NVIDIA’s $1200 RTX 2080 in raw rasterization performance. And the company is continuing to improve its GPUs, bringing us the RDNA 4 which could be even faster than its predecessor.
While the rumors about AMD’s next generation of GPUs are still scarce, there have been some hints here and there. For example, YouTuber Moore’s Law Is Dead recently published a diagram of what looks to be a future GPU from AMD called “Navi 4C.” It’s not clear if this will be the top-tier GPU or the mid-range one, but it’s interesting to see that the chip has been optimized for multi-draw indirect operations, which should make it a good choice for RTX-based games.
Furthermore, MLID spoke to a source who claimed that the most elaborate designs of RDNA 41 had been canceled to accelerate RDNA 5 development. This suggests that Team Red’s engineers may have a skeleton crew left, but they are making good progress and have juicy performance projections for the next GPU generation.
Another interesting tidbit that MLID picked up from their interview with Rick Bergman is that the company has promised to evolve RDNA 4 to higher performance shortly. This should come as a relief to gamers who were disappointed by the lack of a higher-end Navi 41 GPU.
Moreover, MLID’s source also claims that the RDNA 4 GPUs will use chipsets, which is an interesting development since it marks the first time in a while that AMD will be using the design in their consumer-facing cards.
Features and Specifications
The RDNA 4 will be AMD’s first GPU architecture to support hardware ray tracing. It will also feature a new visual pipeline and a 3rd-generation Infinity Cache technology. According to a recent leak by RedGamingTech, the new GPU will be based on TSMC’s 5nm process node and utilize an 8-cluster design. It will have a high acceleration clock of 3.5GHz and the ability to utilize WMMA matrix instructions, which will double the performance per CU.
The new AMD RDNA will also be able to support DLSS and 4K resolutions. The company is expected to launch the RDNA 4 next year in Q3 2024. The RDNA 4 lineup will include mid-range and high-end graphics cards. It will compete against NVIDIA’s Ada Lovelace and Intel’s ARC Battle Mage or Celestial.
Earlier this year, AMD’s EVP of Computing and Graphics Business Unit Rick Bergman shared that the Radeon 8000 series would have significantly higher performance in the future. The upcoming generation will likely have better mid-range offerings than the current generation, but this doesn’t mean that there won’t be a high-end model.
The leaked roadmap from RedGamingTech claims that the upcoming RDNA will have Navi 41, 42, and 43 GPUs. These GPUs will have up to 144 CUs and use GDDR7 memory. The top-of-the-line Navi 41 will have 2x the performance of its predecessor, the 7900 XTX. On the other hand, Navi 42 will have 48 CUs and can deliver up to 43 TFLOPs of FP32 performance. This is a bit lower than the 7900 XTX’s 56 TFLOPs, but it’s still a decent amount of performance. This will be the most affordable model of the two.
AMD RDNA 4 is expected to offer significantly improved performance over the current RDNA 3 architecture. This is due to several factors including faster clock rates, variable rate shading, and ray tracing capabilities. It is also expected to have a higher core count and better power efficiency. Moreover, it is believed to utilize an advanced node such as 5nm or 6nm.
In a recent interview with the site 4Gamer, AMD’s EVP of Computing and Graphics Business Group Rick Bergman spoke about the company’s future GPU offerings. He mentioned that the new RDNA 4 graphics processor units will feature AI integration alongside continued performance and efficiency improvements. Moreover, it will be capable of processing mesh shaders on hardware level thereby improving the overall graphics performance.
David Wang, the head of AMD Radeon Technologies Group, has also made some interesting remarks about the company’s upcoming GPUs. He has stated that the upcoming GPUs will use a method known as Stable Diffusion to process MDI on a hardware level, thus extracting more performance than AMD’s RDNA 2 GPUs. This method is similar to the one used by Nvidia for their RTX 20 series GPUs.
Moreover, AMD might use multiple GCDs in its RDNA 4 GPUs. The company already uses the MCM or Multi-Chip Module technology that combines a Graphics Complex Die (GCD) with multiple Memory Cache Dies. Hence, it is not a stretch to expect high-end RDNA 8000 GPUs from the company. Lastly, it is worth noting that AMD might switch to TSMC’s 3nm process node to keep up with Nvidia’s RTX 5000 Series. This would, however, make the cards expensive. As a result, the RDNA 4 GPUs might only appeal to budget-conscious gamers.
AMD’s RDNA 3 architecture has proved to be a good competitor to Nvidia’s RTX GPUs. It has seen success with the RX 5700XT and the RX 7900 XTX, both of which have beat Nvidia’s RTX 3000 Series in price-per-performance. AMD’s budget-friendly RX 7600 has also been a great option for gamers who want to play the latest titles at a fraction of the cost of Nvidia’s cards.
However, recent comments from the company suggest that the next generation of AMD GPUs may focus on AI and not compete as aggressively in the high-end market. While this is not an entirely surprising move for the company, catastrophists are already predicting that this will be the beginning of the end for AMD’s desktop gaming GPUs.
Next year, the Radeon 8000 series will launch and rumors suggest that the lineup will only feature mid-range offerings. However, this doesn’t mean that Team Red fans will have to compromise on performance. Rick Bergman, EVP of the Computing & Graphics Business Group at AMD, has promised that the RDNA 4 lineup will offer better performance than the high-end offerings from the previous generation.
The leaks surrounding the upcoming Navi 4X graphics cards are not entirely reliable but should be taken with a grain of salt. A rumor from Kepler_L2 suggests that the new GPUs will not include any high-end configurations. This will be the first time in nearly four years that a generation of AMD GPUs has omitted any high-end cards from its lineup. This is similar to the Polaris and RDNA 1 generations which were designed for gamers who wanted more performance at a cheaper price point. The upcoming cards will use the ‘Navi 4x’ GPU compute die.
RDNA 4 VS RDNA 3 VS RDNA 2
AMD RDNA 4 is the next generation of AMD GPUs. It is rumored to improve performance and power efficiency compared to its predecessors.
RDNA 3-based cards have demonstrated that they can compete with Nvidia’s RTX series. This has created excitement among AMD fans.
The upcoming RDNA 4 architecture is rumored to have 2 GCXs with 96 CUs and 12,288 ALUs. This is a significant improvement over the current generation.
Difference between RDNA 4 and RDNA 3
AMD’s RDNA 3 architecture brought the company’s GPUs within striking distance of NVIDIA’s best, with cards like the RX 5700 XT beating the $1200 RTX 4080 in rasterization performance and price. RDNA 4 appears to take this a step further with a focus on AI.
At a Financial Analyst Day presentation, AMD detailed the features that will be coming in its RDNA 4 graphics processors. This includes a new AI matrix accelerator that can handle 32-bit floating point computing and INT4 WMMA Dot4 instructions (Wave Matrix Multiply Accumulate). AMD has also doubled L1 cache capacity to 256 KB and improved L2 cache latency to 1.5x compared to RDNA 2.
However, the company’s Infinity Cache regresses from 128 MB to 96 MB with a slight increase in latency. This is likely because the unified memory controller (UMC) is now on the chip instead of externally connected. This should allow the UMC to service more memory accesses without having to hit Infinity Cache as often.
AMD also improved ray tracing performance with improved BVH traversal and better pixel shader execution. It is not clear whether this will be a per-CU improvement or per-GPU basis, but the improvements should result in higher performance and lower power consumption.
Difference between RDNA 4 and RDNA 2
The RDNA 3 architecture has generated much anticipation among gaming fans due to its improved performance and power efficiency. Big Navi GPUs like the RX 7900 XTX have managed to compete with Nvidia’s RTX 4080, making it clear that AMD’s GPUs are a viable alternative.
RDNA 4 is expected to improve these figures further, thanks to the use of V-Cache chipset stacking technology and a more sophisticated memory hierarchy. It is also expected to have better BVH traversal, which will increase ray tracing performance.
In addition, RDNA 4 will have a more advanced manufacturing process, which will further improve performance and power efficiency. It is rumored to support hardware-accelerated ray tracing, which will make it more competitive with Nvidia’s latest GPUs.
Currently, we don’t know what the name of the new GPU will be, but rumors suggest that it could be called Navi 41 or Navi 42. However, these are just rumors at the moment, and only time will tell whether they turn out to be true or not.