TalkChess.com

So... any word on the implementation of BMI2, PEXT and PDEP?

If it's still as bad as Ryzen 2, these CPU's are out of the question for me. (Also, they are built on an old platform when using AM4; just like Intel 10-xx CPU's built on the X299 platform from 2017.)

I have no complaints about the 3970x. I am getting over 50m nps with my engine. I have seen it hit 100m in the endgame. I really don't think Intel has anything close in terms of price/performance. I also have a Ryzen 7 3700X system which has very nice performance for the price point it is at. It is 8 cores but is close in speed to my 24 core dual e5-2690x workstation (several years old).

Which is more beneficial for chess? BMI2 or AVX512?

I compile with both avx2 and bmi2. I am not using avx512. But really, neither one is worth more than a 10% speed increase in my experience.

Jhoravi wrote: ↑Fri Oct 09, 2020 5:35 am Which is more beneficial for chess? BMI2 or AVX512?

AVX for NNUE, BMI2 for classical

AVX2 can be magical for AMD threadripper, for the right engines.
I saw performance leaps that were startling.

yurikvelo wrote: ↑Fri Oct 09, 2020 8:27 am

Jhoravi wrote: ↑Fri Oct 09, 2020 5:35 am Which is more beneficial for chess? BMI2 or AVX512?

AVX for NNUE, BMI2 for classical

AVX2 to be precise (AVX is just floating point).

AVX512 so far seems to be a slowdown compared with AVX2 when used with many threads because it causes the CPU to clock down too much. This may or may not improve on future Intel CPUs.

mvanthoor wrote: ↑Fri Oct 09, 2020 3:11 am So... any word on the implementation of BMI2, PEXT and PDEP?

If it's still as bad as Ryzen 2, these CPU's are out of the question for me. (Also, they are built on an old platform when using AM4; just like Intel 10-xx CPU's built on the X299 platform from 2017.)

I was doing some thinking on why AMD has not done PEXT very efficiently and it is probably that INTEL has their implementation patented. I don't know if that is the case or not but if it is then someone show AMD this.
http://www.open-aurec.com/wbforum/viewt ... ain#p28725
It might give AMD more leeway in how they can implement their PEXT instruction as the inventor of the instruction, myself, gives AMD permission to implement their version based solely off of my description of the instruction from 2006.

mvanthoor wrote: ↑Fri Oct 09, 2020 3:11 am So... any word on the implementation of BMI2, PEXT and PDEP?

If it's still as bad as Ryzen 2, these CPU's are out of the question for me. (Also, they are built on an old platform when using AM4; just like Intel 10-xx CPU's built on the X299 platform from 2017.)

Don't count on it that Zen 3 will have a better implementation of PEXT/PDEP, besides this omission the Zen 2 is not a bad CPU at all.
Intels x299 platform is not bad either, it still has pcie 3.0 instead of 4.0, is this really a problem?
Pretty soon new HEDT platforms will all have DDR5 support, does this mean that DDR4 is bad?

I have both platforms over here, an Intel 10980XE and an AMD 3970X. For programming I prefer the Intel over the AMD because it is more reliable, has a better BMI2 implementation and has AVX512/VNNI. For bulk applications I prefer the AMD because it can handle almost twice the number of threads and has a somewhat better SMP behavior (probably due to it's large caches).

If the (AMD 5970X?) has a better BMI2 implementation (which I doubt) I can always switch the CPU or (if necessary) the CPU and Main-board.