Unless I missed it, there is not nearly enough information about the benchmarks (and configurations) used to establish that number. Based on my experience I’d guess it’s a carefully cherry-picked selection. Maybe with an emphasis on very large memory scenarios.
If NVIDIA can be bothered to respond, I would expect their marketing department to counter by puzzling out a different set of cherry-picked scenarios that shows the A100 remains ahead.
Get a bucket of popcorn, have a seat on the couch, and enjoy the show :-)
Thanks @njuffa :)
Would be indeed very interesting to see NVIDIA’s official response.
I couldn’t find GB/s and Flops for the ICE Lake either… furthermore they say that the HPC loads are ~50% faster than their previous CPU gen, so that also makes the 1.3x claim even weirder.