General statistics
List of Youtube channels
Youtube commenter search
Distinguished comments
About
Luredreier
JayzTwoCents
comments
Comments by "Luredreier" (@Luredreier) on "This budget CPU doesn't know it's a budget CPU!" video.
@Phenom98 Quite a few people actually, but usually they just run a work around using two AVX2 instructions instead of 1 AVX512 and then merging the result into a essentially the same as a AVX512 output. It runs just fine on Zen, Zen+ and Zen2 processors, it's just slower then running AVX512 directly. That said, any code that's so reliant on AVX512 that it can't handle the AVX2 work around is probably better suited for GPU acceleration anyway... In Zen and Zen+ you kind of had the above work around built into the processor as it internally did 2x 128 bit instructions and merged them to get a AVX2 output internally, causing those instructions to take more clock cycles to finish then they do on Intel or Zen 2.
10
@marcovtjev Actually, I think you're mixing it up with the Zen 1 and Zen + processors. As the Zen 2 processors actually do use 256 bit pipelines.
4
Regarding AVX, I think you've mixing the instruction sets up a bit. AMD got AVX and AVX2 in their processors, they just don't have AVX512 (the third gen of instruction sets in the AVX family of instruction sets). With Zen 1 and Zen + processors their AVX performance is behind Intel when running pure AVX but the mixed AVX + non-AVX performance is superior to Intel while the AVX2 performance is about half of Intel, if I don't remember wrong. With Zen 2 AMD reached parity with Intel in AVX2 (if I don't remember wrong) while still doing better in mixed workloads then Intel with regular AVX. If you got a workload with AVX512 though then Intel can take advantage of that while you have to run a different codepath for Zen and Zen 2 using AVX2 instead of AVX512. Zen and Zen + is of course completely outclassed in AVX512 heavy workloads. Zen 2 of course does way, way better then Zen or Zen + in workloads with a heavy reliance on some form or other of newer AVX instructions (be they AVX2 or AVX512 based code with a AVX2/AVX based work around code path for CPUs that don't have AVX512 support)
1