They use GD6, which is only half bandwidth compare to GD6X on 3080 3090. Maybe "Infinity cache" could helps in games, but not very useful for memory hunger application like mining ETH.
It's not clear if infinity cache will help compute performance or just CPU to GPU bandwidth which isn't a concern
for mining.
Cache will be interesting for RandomX...this might rival top CPUs for randomx...128mb can feed 64 cores so even if each core can compute less than a CPU you have 4x cores per GPU.
I doubt it will perform like a mid range CPU. Random-X is not only about cache, it's designed to perform on CPUs and nothing else.
Actually it is pretty much all about the cache, GPUs have had very little cache per core compared to CPUs, and thats why randomX was so good for just CPU mining, and AMD just upended that. There is also no RAM bottle neck like there is with CPUs on RandomX since GPUs have way more RAM bandwidth than CPUs. Sure the cores won't be as optimized for IPC as CPUs, but the core count + RAM speed should make up for it if someone designs an optimized randomx kernel for these. This will be the first time any GPU can load the whole scratchpad on the L3 cache like a CPU.
GPU cores are less capable and the clocks are about half the freq of a CPU.
GDDR is optimized for sequential access, it doesn't perform as well as DDR on random access.
I presume the infinity cache is also optimized for sequential access.
That could wipe out any numerical advantage in thread count and cache size.
I believe the real benefit of the infinity cache is increasing the throughput for streaming data which
would likley be read only once from the cache and discarded without being written to memory.