| Captain_Yuri said: RDNA3 GPUs only capable of 36tflops vs 61tflop marketing. https://www.reddit.com/r/hardware/comments/102veop/rdna3_gpus_only_capable_of_36tflops_vs_61tflop/ Take it with a grain of salt until we see more proof but essentially, while with Nvidia, you don't see the advertised theoretical FP32 gains in gaming, in rendering and such, you generally do see those types of gains. With Radeon, that doesn't seem to be the case, least for now. |
That has always been a poor indicator. For a different reason the same is true for Ampere since almost nothing runs on float alone and the extra FP32 pipeline can't execute instructions along with integer.
For instance, same website, the 3080 has a 163% advantage over the 2080 Super in TFs but scored only 47%-60% more in Fire Strike, 47% in Time Spy, 56% in Superposition.
The only synthetic benchmark that comes close to that 163% number is image processing in Sandra 2020... but then the same gap exists between the Radeon 5000 and the 6000 series. So yeah, poor indicator.
Essay Pro