The battle between AMD’s RX 6000 and NVIDIA’s RTX 3000 will not end with the latest releases, but in a few months it will continue in the lower end of the gaming market with the launch of the RTX 3050 by those of the green and the RX 6600 on the red side, whose Navi 23 GPU has a different Infinity Cache than expected. Van Gogh, on the other hand, is the codename for AMD Ryzen 6000 series SoCs which will be manufactured at 6nm, will use Zen3 + CPU cores and a GPU based on an RDNA architecture.
AMD Navi 23 GPU reduces Infinity cache
The Infinity Cache is an additional level of cache, actually an L3, which AMD added in RDNA 2 and which is essential in order to be able to achieve its high clock speeds. This is associated with the number of L2 cache partitions and GDDR6 memory channels of VRAM. Thus, the Navi 21 GPU with a 256-bit, 16-channel, 16-partition bus has a density of 128MB of Infinity cache. In the case of Navi 22, we have a 192-bit bus and 96MB Infinity Cache.
According to this logic, the expected Infinity cache amount for Navi 23 was 64MB, but the driver information shows us a cut to 32 MB. We can find out in the driver from the size of the cache line: 32 * 1024 in the L3 cache section. This can result in slower clock speeds as the location in the Infinity Cache is part of the PC’s RDNA 2 architecture scheme to achieve its high speeds.
Rumors suggest that AMD could launch this GPU on both laptops and desktops, so the potential reduction in clock speed from halving the Infinity cache could be due to this factor. . Another reason is to reduce the chip size to produce a higher GPU volume, as AMD aims to compete with the RTX 3050 with its AMD RX 6600s.
AMD Van Gogh Integrated GPU Details
The AMD Van Gogh SoC or APU will be the first time that AMD has implemented the RDNA architecture in its integrated graphics. All this after years where the AMD Vega has been the architecture of choice in the SoCs of the AMD Ryzen family, from Raven Ridge to the current Cézanne.
The first detail we find is that the number of GPU compute units is always 8, a figure which at first glance does not seem an improvement, but one cannot forget the improvements of the calculation units of GCN to those of RDNA. We are therefore talking about an obvious improvement in performance. AMD cannot fall asleep as Intel put the batteries in with its Intel Xe, which also received improvements in Alder Lake.
The other point on which the drivers inform us is the fact that there is no Infinity cache on the SoC, which is not surprising since it occupies a large size in RDNA 2 dedicated GPUs. In any case, we do not know if Van Gogh’s iGPU is RDNA or RDNA 2 since AMD considers RDNA 2 to be those of Xbox Series and PS5 consoles without having it hidden inside their iGPUs.