Per Hesselgren (perhesselgren.delete@this.yahoo.se) on October 9, 2020 7:29 am wrote:
> Adrian (a.delete@this.acm.org) on October 8, 2020 11:53 am wrote:
> > ⚛ (0xe2.0x9a.0x9b.delete@this.gmail.com) on October 8, 2020 11:22 am wrote:
> > > CPU) then it is likely there exists at least one architectural
> > > feature within the new CPU that has alone/by-itself
> > > 50-100% higher performance in synthetic benchmarks designed
> > > to stress that particular CPU feature. The question
> > > is: What Zen 3 features with 50-100% improvement are there? The only such Zen 3 feature disclosed so far is
> > > that L3 cache directly accessible by a CPU core has increased from 16 MiB to 32 MiB.
> > >
> > > -atom
> >
> >
> > As I have already mentioned in another post, there is an ancient rumor that
> > Zen 3 has a better floating-point throughput per cycle than Zen 2 by 50%.
> >
> > The wider integer execution units that were mentioned in the presentation might
> > also provide a similar increase in throughput for certain operations.
> >
>
> I saw that 5950X is +27% faster than 3950X on CAD SolidWorks 2019. This is
> high compared to for example +12% on Raytracing V-Ray 4.10. Any ideas why?


On the Solidworks site the benchmark is said to be influenced mostly by single-thread processor performance.

27% is just a little above the average improvement in ST (19% IPC + 5% clock frequency), so Solidworks is a typical case, while Raytracing is below average.

Maybe Raytracing is not limited less by the core, but more by memory latency or by anything else.

Or else if Raytracing is heavily multithreaded, then the improvement of 5950X is less than for ST (claimed to be at most 20%).