kevboard
Member
Ampere SM improved upon Turing SM's integer CUDA cores with floating point (FP) capabilities without increasing TMU (texture management units). AMD executed a similar Ampere-like FP double rate increase per SM with RDNA 3.0 CU hardware generation. PC's RDNA 4 CU and RDNA 3.5 CU have double texture sampling rate improvements.
okay... not sure why you're saying all that... especially when we are talking about RDNA2, which has no dual issue FP32.
but just to reiterate my comment, I am simply going by real world performance... and by real world performance my RTX3060ti for example, which is a 16.2 TFLOPS card, is about as fast in raster performance as the PS5's 10.2 TFLOPS RDNA2 GPU... maybe a tiny bit faster in some games.
so that gives me a decent comparison point for Ampere and AMD, and specifically RDNA2.
Last edited: