AMD seems to have the edge over Nvidia when it comes to memory latency performance with current generation graphics cards. This conclusion was drawn by technical site Chips and Cheese, which performed comparisons between Big Navi and Ampere, using pointer lookup benchmarks (written in OpenCL) to evaluate memory latency in nanoseconds (ns) with the RX 6800 AMD's XT versus Nvidia's RTX 3090. And the results were interesting to say the least, considering that the Big Navi card has more layers of cache to pass through on the trip to memory, but manages to match the Ampere GPU. Nvidia works with a simple two-level cache scheme, L1 and L2, with the cache being small amounts of very fast built-in storage (much faster than actual video RAM) right there on the chip (as is). processors)). However, with Big Navi, AMD does things very differently, using multiple layers of cache: L0, L1, L2, and Infinity Cache (which is actually L3). This means there are more levels to go through, as we mentioned, so you think it might slow things down even more in terms of latency, but it doesn't. In fact, AMD's cache is incredibly fast with low latency across all of these multiple layers, while Nvidia has high L2 latency: 100 ns, compared to around 66 ns for AMD from L0 to L2. The result of all this is that the overall result is practically a dead end.
(Image credit: chips and cheese)
Amazing feat
As Chips and Cheese notes: "Surprisingly, RDNA 2's VRAM latency is about the same as Ampere's, although RDNA 2 looks for two more cache levels in the memory path." The site notes that fast L2 caches and AMD's L3 with its ultra-low latency could give Big Navi GPUs the edge in less demanding workloads, and this could explain why the RX 6000 graphics cards perform very well at lower resolutions (where the GPU is not). t being pushed with almost the same force). Of course, this is a tentative conclusion, and we should be careful not to read too much into a single test. Furthermore, it's unclear how such a memory latency advantage could translate into actual performance in games anyway. Chips and Cheese also notes that memory latency with the processors is much faster, of course, but that comparatively "GDDR6 latency itself is not that bad" on a general level. The best graphics card deals of the day
Through Tom's Hardware