Next Intel CPU It will also have the 512-bit AVX

[AndresAhumada]

Hi guys!!

Reading on the web I found these features to NEXT egenreacion of Xeons processors.
512-bit AVX


I want to ask a question to Vlado.

Could this benefit the math, or other technology provided in Vray?


regards

[benb32]

Fairly sure that Vray already supports this for certain operations including improving embree and light cache calculations. (Can't put my finger on it but seem to remember a thread or addition to the docs mentioning speed of light cache ~ 10%)

The 512 bit registers are already part of the current generation Xeon line, additional instructions are planned for future generations also.

[Paul Oblomov]

Well, this is how it looks now. This is how my cpu works with AVX (2699 v4). If you guys can find any avx test - lemme now and I will test it.
http://www.theregister.co.uk/2016/03...on_e5_2600_v4/

[vlado]

We do not use AVX anywhere in our code right now. From our experiments a few years ago with Embree, it didn't bring any performance improvements. In fact, on some processors AVX may actually slow things down, because it causes the CPU to consume more power and to reduce its clock speed as a result.

It is possible that things have changed since then, of course, so it might be worth investigating again. We are still using just plain SSE2, not even SSE4.1, so there's definitely place for improvement.

Best regards,
Vlado