AMD appears to be readying a long-rumored and eagerly awaited feature in graphics cards. The upcoming AMD RDNA 2 or Big Navi GPUs could have an early iteration of ‘Infinity Cache’ that would significantly reduce latency and boost bandwidth if implemented correctly.
AMD may have been finetuning Infinity Cache, a feature that holds the potential of significantly reducing the constant request and exchange of data that takes place between a GPU and the onboard VRAM. Although there’s no official confirmation from AMD, it appears the company is developing and thinking of embedding the new AMD Infinity Cache within its Big Navi or Navi 2X Graphics Cards.
AMD To Increase GPU Cache To Achieve Infinity Cache On Its RDNA 2 Graphics Cards?
A new leak suggests AMD could be implementing Infinity Cache, a technology that holds the potential of reducing latency, improving bandwidth, and even boosting the efficiency of the Graphics Cards. The technology wasn’t feasible in the previous generation of AMD GPUs, but the next-gen RDNA 2 or Big Navi GPUs appear to have the necessary hardware and architecture to support the feature.
The Infinity Cache name correlates with AMD ZEN technology called Infinity Fabric which is the company’s proprietary interconnect system architecture for CPU and GPU cores. However, the Infinity Cache could be a new technology coming to AMD Radeon graphics cards.
AMD INFINITY CACHEhttps://t.co/SLGJVY9m6t
— 188号 (@momomo_us) October 5, 2020
The above Tweet points to a trademark page where it clearly states ‘AMD Infinity Cache’. It is important to note that this is not a patent application. It is merely a trademark application, which indicates AMD could use the Infinity Cache trademark, quite possibly in the upcoming or future iterations of Graphics Cards. However, it is also likely that AMD is merely trademarking the name, and might not use the same.
Regardless of the same, the Infinity Cache does hold a lot of potential in the world of high-intensity and ultra-high-resolution graphics. The feature, if implemented correctly, can address several issues that are holding back GPUs, and forcing companies to deploy other solutions to address the common bottlenecks to bandwidth and latency.
What Is AMD Infinity Cache For Graphics Cards?
Traditionally, CPUs have had three types of caches: L1, L2, and L3. The L1 cache is extremely fast but very small, the L2 cache is proportionally larger, but slower. The L3 cache and RAM follow the same general trend. GPUs generally only have up to L2 cache, and no L3 cache. It appears AMD has decided or managed to embed the L3 Cache and is terming the same as AMD Infinity Cache. As this is a Cache Memory instead of the regular VRAM, the latency is significantly reduced, and there’s a positive impact on bandwidth.
Cache amplifies bandwidth, especially when compared with VRAM. This is because, instead of the GPU fetching data from VRAM memory, if the data needed happens to be in cache, games can fetch the data directly from there. That directly reduces the bandwidth demand because games are using the cache data link instead of the memory bus.
GPUs have traditionally had a rather low cache, in the range of just 4MB, but AMD is rumored to be embedding 128MB. This large increase in cache size should certainly ensure bandwidth demand can be reduced significantly from VRAM.
In order for the AMD Infinity Cache to work well, game developers would also need supporting algorithms to populate the cache. The whole juggling of data could be as complex as branch prediction algorithms for CPUs. Hence it is not clear if the upcoming Big Navi would have a finished and polished iteration of Infinity Cache.