Embree 4.0 Is Running Well On Intel 4th Gen Xeon Scalable "Sapphire Rapids"

The other testing I was curious about for Embree 4.0 with Sapphire Rapids was making use of its ISA argument to compare the AVX impact and performance going back to only SSE4.2 usage. By default Embree will make use of all available/supported instruction set extensions present while this ISA comparison shows the performance if limiting the Xeon Platinum 8490H 2P server to just AVX2 usage, AVX, or SSE4.2.

To no real surprise, there were healthy gains each step of the way as the greater Advanced Vector Extensions were employed...

One of the more interesting areas was wanting to look at the affect on the dual socket combined CPU power consumption at the different ISA levels with Embree 4.0. When making use of AVX2 and AVX-512 was actually a slightly lower power consumption average on the dual socket Sapphire Rapids server than when using AVX or just SSE4.2: ~608 Watts versus ~621 Watts. The peak 2P power consumption was similar.

On a performance-per-Watt basis, the AVX-512 results on Sapphire Rapids are in great shape. As illustrated previously in AVX-512 Performance Comparison: AMD Genoa vs. Intel Sapphire Rapids & Ice Lake, the AVX-512 performance with Sapphire Rapids is in much better shape than prior generations -- especially if comparing to the AVX-512 performance back to the Skylake era when there was significant power and thermal differences.

In fact, when it came to the CPU0 temperature with AVX2 and AVX-512 it was lower than when just using AVX or SSE4.2.

The other models tested with the Embree Pathracer ISPC all showed off well with their AVX-512 performance on Sapphire Rapids.

Across all of the Embree configurations tested, overall the CPU power consumption of the dual Xeon Platinum 8490H was on average at its lowest point when making use of the optimal AVX-512 path. This is great to see and jiving with what my prior AVX-512 Sapphire Rapids testing has shown with other workloads.

The CPU peak frequency when engaging AVX-512 for Embree 4.0 was also minimally impacted.

And no elevated CPU temperatures when running Embree with AVX-512 -- but rather the opposite with more efficient usage.

These Embree 4.0 AVX-512 results jive with my earlier Sapphire Rapids AVX-512 benchmarks across many different workloads with the involved thermal/power costs being minimal compared to older generation server processors. It's also great to see the speed-ups with Embree 4.0 compared to the earlier 3.13 series. Now next up to check out the Embree 4.0 SYCL GPU capabilities.