By: Per Hesselgren (perhesselgren.delete@this.yahoo.se),
Room: Moderated Discussions
anon.1 (abc.delete@this.def.com) on October 31, 2020 8:06 am wrote:
> Adrian (a.delete@this.acm.org) on October 30, 2020 12:37 pm wrote:
> >
> > It appears that the rumors about Zen 3 having increased the
> > number of 256-bit FMA units from 2 to 3 have been true.
> >
> > The benchmarks that have begun to appear show floating-point
> > increases around 45% over Intel CFL/CML and Zen 2.
> >
> >
> > Also the number of integer AVX2 execution units must have been increased
> > because there are speed-ups over 40%, or even over 60%.
> >
> >
> > https://www.sisoftware.co.uk/2020/10/29/amd-ryzen3-5800x-review-benchmarks-cpu-8-core-16-thread-performance/
> >
> >
> >
>
> Are you just seeing the effect of an effectively larger cache
> or do you know that these benchmarks are 100% compute bound?
Now they only show 5950X
5950X on Sisoft
DGEMM should be more depending on cache but most of the others not
> Adrian (a.delete@this.acm.org) on October 30, 2020 12:37 pm wrote:
> >
> > It appears that the rumors about Zen 3 having increased the
> > number of 256-bit FMA units from 2 to 3 have been true.
> >
> > The benchmarks that have begun to appear show floating-point
> > increases around 45% over Intel CFL/CML and Zen 2.
> >
> >
> > Also the number of integer AVX2 execution units must have been increased
> > because there are speed-ups over 40%, or even over 60%.
> >
> >
> > https://www.sisoftware.co.uk/2020/10/29/amd-ryzen3-5800x-review-benchmarks-cpu-8-core-16-thread-performance/
> >
> >
> >
>
> Are you just seeing the effect of an effectively larger cache
> or do you know that these benchmarks are 100% compute bound?
Now they only show 5950X
5950X on Sisoft
DGEMM should be more depending on cache but most of the others not
Thread (7 posts)
| Topic | Posted By | Posted |
|---|---|---|
| Zen 3 triple FMA units | Adrian | |
| So use AVX2 kernels and ignore AVX512 entirely? (NT) | anonymous2 | |
| So use AVX2 kernels and ignore AVX512 entirely? | anon | |
| Good luck ignoring AVX512 with kernels that need scatter (NT) | Heikki Kultala | |
| Zen 3 triple FMA units | anon.1 | |
| Zen 3 triple FMA units | Per Hesselgren | |
| Zen 3 triple FMA units | Adrian |



