The Xeon vs. EPYC Performance With Intel's oneAPI Embree & OSPray Render Projects

[starshipeleven]

Looking at those numbers....256 new threads and 8.57 FPS.

So, like, do those tools also measure in Frames Per Day for regular people with regular systems?

*week

Besides, it's not for real-time rendering, it's renderfarm software.

[lowflyer]

LuxCoreRender was also tossed in for this open-source physically based renderer. LuxCoreRender isn't developed by Intel but does make use of their Embree library and Threaded Building Blocks (TBB). LuxCoreRender ended up being faster on the Xeon Platinum server with those Intel components, but then again most LuxCoreRender usage is done with OpenCL on GPUs.

Something went wrong with the LuxCoreRender tests, the 7742 only got about 50% of the 7601!

It would be interesting to see the same test with a CPU that advertises itself as "Intel" but in fact is an AMD CPU. Intel is known for their "cripple AMD" function. Although Intel has lost unfair competition cases against AMD, the "cripple AMD" function is, according Agners Blog, still present to this day in Intels compilers.

[user556]

Something went wrong with the LuxCoreRender tests, the 7742 only got about 50% of the 7601!

[smitty3268]

A little strange to see the 7601 lose so badly in so many of the tests. Those are the results I'd expect of a single 7601, not two.

The older epyc's weren't very fast at AVX2 compared to the new ones or Intel chips. It only had 128-bit registers vs 256-bit now, so cutting performance in half is about what you'd expect.

[Michael]

Now some up to date LLVMpipe vs OpenSWR numbers, please. - GREAT work Michael!

Yes I hope to have OPenSWR numbers out in a few days.... still overloaded with benchmarking :/

[nuetzel]

Now some up to date LLVMpipe vs OpenSWR numbers, please. - GREAT work Michael!

[schmidtbag]

Well, yeah, it has more cores than threads. What kind of performance where you expecting?

lol whoops, not sure how I missed that one. Actually, I screwed up the c/t count of the Xeons too. That's what I get for not drinking my coffee first...

The Epyc 7601 setup is 64/128 and the Xeon is 56/112 and the Xeon runs ~0.8ghz faster. I'd reckon the extra cycles the Xeons put out is what's giving them the edge against the older Epyc.

Both CPUs have an all-core boost of 2.7GHz, so, assuming Michael's got sufficient cooling (which I think he does), they should both be about equal in speed. So, considering the 7601 also has 13% more cores, it's weird that it'd suffer such a drastic loss in performance. I know the IPC isn't as good on the AMD side but it's not more than 13% worse. But again, I know Intel has an advantage with AVX.

[cb88]

Well, yeah, it has more cores than threads. What kind of performance where you expecting?

The Epyc 7601 setup is 64/128 and the Xeon is 56/112 and the Xeon runs ~0.8ghz faster. I'd reckon the extra cycles the Xeons put out is what's giving them the edge against the older Epyc.

Also Zen 1 has an FPU that is half as wide...

[skeevy420]

I'm not questioning the performance of the newer Epyc; I'm talking about the old one. It has 64c/48t yet it's almost half the performance of the Xeon setup, which to my recollection has 48c/96t.
For the record, I'm not questioning the Xeon's performance (especially where AVX512 is in use).

Well, yeah, it has more cores than threads. What kind of performance were you expecting?

The Epyc 7601 setup is 64/128 and the Xeon is 56/112 and the Xeon runs ~0.8ghz faster. I'd reckon the extra cycles the Xeons put out is what's giving them the edge against the older Epyc.

[schmidtbag]

why? The newer epyc has double the cores.

I'm not questioning the performance of the newer Epyc; I'm talking about the old one. It has 64c/128t yet it's almost half the performance of the Xeon setup, which to my recollection has 56c/112t.
For the record, I'm not questioning the Xeon's performance (especially where AVX512 is in use).