Comment by oofbey
11 hours ago
I think the Top500 benchmark is biased here. It’s a supercomputing benchmark which means it’s going to heavily weight SIMD and float64 performance, which are things that rarely matter for everyday computing. SIMD has uses in the real world - image or video editing, data crunching, graphics. But consumers basically never use fp64 any more, and so many modern chips have next to zero performance on them. That could easily skew the numbers in intel’s favor.
I run a number of other benchmarks for a better comparison (linked from my blog post: https://github.com/geerlingguy/sbc-reviews/issues/111).
The Intel chip holds its own for multicore in less bursty workloads, where the fans help.
An interesting thing there is how much better the Geekbench scores are on Linux than Windows. Makes you think how the Intel/AMD chips would compare to Apple when both running the same OS. Or even just when both not being held back by Windows.
It's funny, too — in my testing with Asahi Linux, some benchmarks ran better on Linux than macOS on Apple Silicon!
> consumers basically never use fp64 any more
They do in audio processing. ;)
I usually use fp32 for the sole reason I can fit twice as many ~24 bit samples per register.
What? Pretty much every general-purpose calculation on your computer that isn’t an integer is done with FP64. AI may not use it and games may prefer FP32 for many things, but your CPU certainly does a ton of FP64 work.
Which workloads?
Literally every workload not named above. It’s the default for a non-integer number.
* Web rendering and almost all web apps (see the JS number type)
* Most things in a game not done on the GPU
* Audio processing
* Most GUI processing/rendering
* Every miscellaneous/scalar calculation these days
Believe it or not, on a CPU this adds up to a lot more than the work you get doing AI processing.