128-core AMD server

Discussion of chess software programming and technical issues.

Moderators: hgm, Rebel, chrisw

Leo
Posts: 1078
Joined: Fri Sep 16, 2016 6:55 pm
Location: USA/Minnesota
Full name: Leo Anger

Re: 128-core AMD server

Post by Leo »

jdart wrote:Some benchmarks (Stockfish 8, 16G hash):

32-core dual Opteron (2xOpteron 6376, 1/4 of the AMD server box, Linux):
26.3M nps

24-core dual Opteron (2xOpteron 6344, Linux):
21.8M nps

dual Xeon 2670 (16 cores, Linux):
24.0M nps

i7-6900k (8 cores, Windows):
14.08M nps
What about the argument that its time to depth that matters?
Advanced Micro Devices fan.
Leo
Posts: 1078
Joined: Fri Sep 16, 2016 6:55 pm
Location: USA/Minnesota
Full name: Leo Anger

Re: 128-core AMD server

Post by Leo »

jdart wrote:Rates in my area ramp up with usage and I have once in a while hit their "super overcharge now we're really serious" rate level, whatever they actually call it.

I am planning to get solar panels on the house next year.

--Jon
I have a 15 amp breaker that kept tripping. It was hot to the touch. I now have to wait until I can hook up the computer to 20 amps.
Advanced Micro Devices fan.
jhellis3
Posts: 546
Joined: Sat Aug 17, 2013 12:36 am

Re: 128-core AMD server

Post by jhellis3 »

What about the argument that its time to depth that matters?
TTD is a useless measurement for LazySMP.
Adam Hair
Posts: 3226
Joined: Wed May 06, 2009 10:31 pm
Location: Fuquay-Varina, North Carolina

Re: 128-core AMD server

Post by Adam Hair »

Leo wrote:
jdart wrote:Rates in my area ramp up with usage and I have once in a while hit their "super overcharge now we're really serious" rate level, whatever they actually call it.

I am planning to get solar panels on the house next year.

--Jon
I have a 15 amp breaker that kept tripping. It was hot to the touch. I now have to wait until I can hook up the computer to 20 amps.
Have your electrician check for a loose connection also. There could be a problem where the wire connects to the circuit breaker (small possibility) or where the circuit breaker plugs onto the electrical panel busbar (greater possibility). It is doubtful that you have a large enough overcurrent to cause the circuit breaker to be hot to the touch.

My guess is that the connection between the breaker and busbar is failing.
Leo
Posts: 1078
Joined: Fri Sep 16, 2016 6:55 pm
Location: USA/Minnesota
Full name: Leo Anger

Re: 128-core AMD server

Post by Leo »

I have read over and over that a powerful 8-16 core machine will defeat a 32-64 core machine. After 16 to 20 cores it scales very badly. Has something changed or is this still true?
Advanced Micro Devices fan.
Leo
Posts: 1078
Joined: Fri Sep 16, 2016 6:55 pm
Location: USA/Minnesota
Full name: Leo Anger

Re: 128-core AMD server

Post by Leo »

Adam Hair wrote:
Leo wrote:
jdart wrote:Rates in my area ramp up with usage and I have once in a while hit their "super overcharge now we're really serious" rate level, whatever they actually call it.

I am planning to get solar panels on the house next year.

--Jon
I have a 15 amp breaker that kept tripping. It was hot to the touch. I now have to wait until I can hook up the computer to 20 amps.
Have your electrician check for a loose connection also. There could be a problem where the wire connects to the circuit breaker (small possibility) or where the circuit breaker plugs onto the electrical panel busbar (greater possibility). It is doubtful that you have a large enough overcurrent to cause the circuit breaker to be hot to the touch.

My guess is that the connection between the breaker and busbar is failing.
Thanks for the info.
Advanced Micro Devices fan.
jhellis3
Posts: 546
Joined: Sat Aug 17, 2013 12:36 am

Re: 128-core AMD server

Post by jhellis3 »

I have read over and over that a powerful 8-16 core machine will defeat a 32-64 core machine. After 16 to 20 cores it scales very badly. Has something changed or is this still true?
It was never true? This has more to do with clock speed, heat generation, and power consumption than any sort of scaling issue.

Generally, the more cores per chip the lower the base and boost clocks because you can only dissipate so much heat per mm^2, unless you want to move to water cooling or something more exotic.

But the idea that more cores are somehow inherently bad is just bullshit. Are there diminishing returns, yes. Does performance per dollar drop drastically, yes. But more nodes searched at higher depths will always be better than nothing (provided equal clocks).
jwes
Posts: 778
Joined: Sat Jul 01, 2006 7:11 am

Re: 128-core AMD server

Post by jwes »

jhellis3 wrote:
What about the argument that its time to depth that matters?
TTD is a useless measurement for LazySMP.
So you're saying that LazySMP finds better moves at equal depths rather than finding the same move faster.
jdart
Posts: 4366
Joined: Fri Mar 10, 2006 5:23 am
Location: http://www.arasanchess.org

Re: 128-core AMD server

Post by jdart »

Arasan scales quite badly on 32 cores, at present. It is on my to-do list to look at soon.

But other engines do better: Stockfish for example.

Crafty also uses multiple cores efficiently and can hits amazingly high NPS, but I haven't tried it on this 32 core box.

--Jon
Dann Corbit
Posts: 12538
Joined: Wed Mar 08, 2006 8:57 pm
Location: Redmond, WA USA

Re: 128-core AMD server

Post by Dann Corbit »

jdart wrote:Arasan scales quite badly on 32 cores, at present. It is on my to-do list to look at soon.

But other engines do better: Stockfish for example.

Crafty also uses multiple cores efficiently and can hits amazingly high NPS, but I haven't tried it on this 32 core box.

--Jon
Texel scales brilliantly
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.