7 ms·
Good thing nobody actually uses AVX512
by hnuser123456 6y ago
Good thing nobody actually uses AVX512
- rbanffy 6y agoInterestingly enough, the AVX512 test was the easy win for Intel. Nobody uses it because it wasn't available and, when available, wasn't much faster overall. Now it is.
- Cullinet 6y agoiirc wasn't until recently available in the HEDT line, but the user base is equipped with scalable xeon boxes not awfully constrained desktop chips.
- rbanffy 6y agoI'm imagining its availability will have more impact outside the user base that's already well served by Xeon Scalable. Visicalc didn't make a big impact in mainframes.
- kcb 6y agoIs it though? The vast majority of PCs even on the enthusiast level aren't going to have coolers that can handle 300w for any sustained period.
- rbanffy 6y agoFrom eyeballing the chart, they'd only need to handle the load for 20% of the time compared to non AVX512 CPUs to be equivalent. Anything more is an advantage.
- kllrnohj 6y agoWho both buys consumer desktop CPUs (without ECC) and runs software that can use AVX512? Games won't use avx512 (not really simd math heavy work on the CPU, and also consoles don't have it). Web browsers won't use AVX512 (similar to games - not really SIMD-friendly raw math computation limited). Video & photo editing maybe will, but GPU acceleration is already more widely available & faster than avx512. What else does that leave?
- krapht 6y ago<-- this engineer's company workstation uses a consumer desktop CPU Most scientific and engineering computing applications use double precision floats.
- rbanffy 6y agoHow many of those workloads would benefit from it if it were more available? I'm not sure, but we haven't seen it being widely deployed just yet.
- 37ef_ced3 6y agoAVX-512 software is appearing. For example, for neural net inference: https://NN-512.com https://NN-512.com AVX-512 is a pleasure from the programmer's perspective, much nicer than writing software for GPUs or for earlier SIMD instruction sets like AVX2/SSE Versus a GPU, the AVX-512 machine model is just much simpler, and there are no drivers or proprietary compilers Versus earlier SIMD instruction sets, AVX-512 provides all the instructions you want (permutations, floating point conversions, etc.) and they're all masked so edge cases are easy. AVX-512 Foundation is a really nice instruction set and it has a bright future
- dogma1138 6y agoIsn’t the issue with AVX-512 is that different CPUs support different instructions? https://en.m.wikipedia.org/wiki/AVX-512 https://en.m.wikipedia.org/wiki/AVX-512 It might eventually get normalized but atm it’s a bit of a mess.
- wtallis 6y agoIt's a bit less messy if you ignore the GPU-like Xeon Phi (Knights Landing and Knights Mill) and the failed Cannonlake that Intel definitely would prefer everyone pretend never happened. But that still leaves you with a progression of four levels of expanding support, and the Cooper Lake mess of having BFLOAT16 but not the stuff added in Ice Lake and Tiger Lake. If you don't cut down the support table to eliminate irrelevant niche products, it's most of the way to being a Gray code.
- dogma1138 6y agoI think Icelake and Tiger lake not having BF16 is going to be a big issue especially for machine learning applications, I can’t find what Rocket Lake supports yet but I’m guessing it’s going to be similar to Tiger Lake. There is also the big unknown of what kind of support AMD will bring to the table. I think Intel is banking on people using OneAPI more than writing AVX-512 code using intrinsics or compiler magic. At that point compatibility might be not as much of an issue if you ignore potentially drastic differences in performance.
- Cullinet 6y ago37ef_ has covered why AVX512 is plenty in use internally everywhere I know. Shrink wrapped products, not so much.