AMD Ryzen Threadripper 1950X 16-Core And 1920X 12-Core CPUs

[Dann Corbit]

It is still very difficult to know if this chip will good for chess.... and how will it compare to the i9 Chip?

Why is it difficult? it's basically two Rizen 7 mashed together, and we know how that one compares to... everything:
http://www.hardware.fr/articles/965-2/p ... tives.html

Thanks, so should we expect something in the range of 32'000 kn/s with stockfish 8 which would be very close to an i9 7980XE ?

It should be north or 30 million, for sure, how that will compare to an unreleased CPU which base frequency remains unknown, is anyone's guess.

My best estimate - simply from extrapolating from data found on the internet is that SF would bench around ~40 M nps on i9 7980xe. Very rough - back of napkin type calc- could easily be off by more than 10% eitherway - there are no true benchmarks out there. There are benchmarks out there for i7900 - but until you see it you never know for sure.

When you spend this much on a chess machine, might as well go the whole hog and buy a Xeon system.

The dual Epyc is not too shabby:
http://www.ipmanchess.yolasite.com/amd- ... -bench.php
86,375,777 NPS

I would normally be impressed with the 64 core Epyc but I have so much about engines not scaling beyond 16 cores that I can't get excited about it. I also wonder about its time to depth.

NPS and depth are both crap.

All that matters is Elo.

[Leo]

PC World headline: "AMD Threadripper prices undercut Intel's Core i9 by as much as $1,000"

[Dann Corbit]

PC World headline: "AMD Threadripper prices undercut Intel's Core i9 by as much as $1,000"

Good time to buy AMD stock

[shrapnel]

I would normally be impressed with the 64 core Epyc but I have so much about engines not scaling beyond 16 cores that I can't get excited about it.

Exactly ! Anything above 12 Cores is pointless because all Chess Engines are still practically in the Stone Age.... they are inherently incapable of utilizing the full power of modern Computers in a meaningful way.
As has been demonstrated many times in online Matches, any reasonably decent 4-Core CPU using latest Stockfish Development version and an average Book, is fully capable of holding even the fastest Dual-Xeons to Draws.
So, what's the point in spending so much money on fast computers (to say nothing about the huge Power Bills) if its only for chess purposes ?

[Dirt]

So, what's the point in spending so much money on fast computers (to say nothing about the huge Power Bills) if its only for chess purposes ?

For testing they're great! You can have your own mini-cluster.

For analysis they are difficult to use effectively.

[Dann Corbit]

So, what's the point in spending so much money on fast computers (to say nothing about the huge Power Bills) if its only for chess purposes ?

For testing they're great! You can have your own mini-cluster.

For analysis they are difficult to use effectively.

Asmfish has special NUMA code and it scales pretty well to fairly high counts (e.g. 60 cores or so before things fall off the cliff).

Most others not so much.

[Modern Times]

Exactly ! Anything above 12 Cores is pointless because all Chess Engines are still practically in the Stone Age.... they are inherently incapable of utilizing the full power of modern Computers in a meaningful way.

Shh! Don't tell the TCEC organisers.

[corres]

The main issues with CPU having many cores are:
1, The more cores the less clock frequency of CPU are.
2, To make parallel the code of chess engines are a very limited thing.
I think the ideal number of cores for a CPU to analyze chess positions
are 16, nowadays.
In my opinion 12 are rather small for cores because Elo change on power of CPU ~linerarly if number of cores enhanced as power of 2 (1x-2x-4x-8x-16x..).

[shrapnel]

Asmfish has special NUMA code and it scales pretty well to fairly high counts

In my opinion, Asmfish is over-rated.
It reminds me of a guy running very fast on a Treadmill....runs very fast, very high nps, but gets exactly NOWHERE, like the guy on the Treadmill !
In other words, its results are no better or no worse than other Stockfish derivatives/versions that I've observed playing on Infinity Chess.

[shrapnel]

In my opinion 12 are rather small for cores because Elo change on power of CPU ~linearly if number of cores enhanced as power of 2 (1x-2x-4x-8x-16x..).

Not true.
I don't know the Math, but I've observed that a heavily-overclocked, well-cooled 4-Core ( or 6 or 8) is more than capable of repeatedly holding even the fastest dual-xeons to draws.
The NPS may still be lower than the dua-xeons, but it make no difference to the final results.
Perhaps something to do with the increasing number of minute errors creeping into the Engine calculations with increasing number of Cores, whether Physical or Virtual.
Actually, come to think of it, maybe this is one of the main reasons why all Chess engines are relatively primitive, as I postulated earlier.
The day chess engines progress enough to fully utilize the resources of dual-xeons to win consistently, I'll go out and buy one, money no object !