Best CPU for chess engine game analysis

Sort:
Canutus

Apols for resurrecting an old Thread, but I'm trying to get into correspondence/centaur chess more seriously'

In 2020 what do you need as a chess UI?

Is Arena sufficient and how does it compare to Fritz or Aquarium?

Any insights would be appreciated.

Canutus
pfren wrote:
Canutus έγραψε:

Apols for resurrecting an old Thread, but I'm trying to get into correspondence/centaur chess more seriously'

In 2020 what do you need as a chess UI?

Is Arena sufficient and how does it compare to Fritz or Aquarium?

Any insights would be appreciated.

 

GUI is not very relevant.

I know many ICCF players use Aquarium because of IDeA, while I do not bother at all- I mostly use SCID vs PC, CorrChess (Stockfish tuned for LTC) and my brains.

You should be very careful during the opening, where the engine is useless- you will need a very good and fully updated chess database, including games from most correspondence servers.

Thanks for the reply, I have an old version of Fritz which I use for databases but when analysing with modern engines like Stockfish my rather ancient computer slows to a crawl and can't be used for anything else.

I haven't tried Scid v PC, but was thinking of initially trying Arena to see if it was less resource hungry than Fritz.

How important is hardware, do I need a new machine?

NTBT_DHS

How about AMD Threadripper 3970X? I am using it 32 cores and 64 threads. Which engine is best for it?

drmrboss
NTBT_DHS wrote:

How about AMD Threadripper 3970X? I am using it 32 cores and 64 threads. Which engine is best for it?

 

Spitting your money on both CPU and GPU is the best idea!!

 

The strongest engine is the combination of Leela on 2080 Ti plus Stockfish on 16 cores cpu ( as Stockfish SMP is poor beyond 16 cores, you dont get double benefits from 16 to 32 corrs  32 to 64 cores, instead +30 %, +20% etc). And Leela has been praised by 2015 iccf champ as if perfect Tablebase in opening.

 

Here is also one test who split the phase of two engines.

drmrboss
Howhorseymove wrote:
I like to look at this topic from a potential coders point of view. Suppose you decide that you are going to spend 3 or 4 months to optimize the code so that it works great with an 8 core system. So I spend 3 or 4 months to have it optimized to work on an 8 core system and ignore other improvements that could be made.

After 3 or 4 months you release the code and you get a lot of praise but find that only 3 percent of your user base can make use of that feature. That was not time well spent.

Coders are usually trying to find a way to make their code do more in less time while trying to avoid creating a bugs or glitches as a result of reworking the logic of a program.

I don’t think chess programs are focused on using multiple cores as much as trying to optimize for the most common features that a user might have on their system.

Stockfish codes had been modified  40,000 + times and tested  2 billions + games, one of the most fine tuned program in the world.

 

It is not due to the weakness of program, it is due to nature of search.  In initial position, there are 20 possible moves, but SF search mostly 2-3 promising moves ( branches) and keep searching deeper and deeper. 

 

If you have 20 cpu, 3 cpu will be searching for the most promising moves like 1. e4 or 1. d4 or 1. Nf3 ( the other 17 possible moves like 1. g4 1. f3 etc will be almost excluded in search from AB pruning except a small short nodes search).

 

Because of Alpha Beta pruning system, parallel search in massive CPU is not that useful. 

drmrboss

And also due to the nature of chess, the search trees will  be overlapping somewhere. For example,

CPU A is searching 1. e4 c5 2.Nf3

CPU B is searching 1. Nf3 c5 3. e4

 

And now you see the two search trees become overlap, wasting the power of 1 CPU.

anonym_anonym1

 Xeon 2697 v2(12 core CPU) is still strong for chess?

anonym_anonym1
pfren написал:
anonym_anonym1 wrote:

 Xeon 2697 v2(12 core CPU) is still strong for chess?

 

Just average. Slightly weaker than a $200 R5-3600, while consuming about double energy. Still, not bad at all for a CPU aged 7-8 years.

whats about  AMD 2700X it is better  then Xeon  2697v2... seems I will buy AMD ,thanks for Advise! 

 

play4fun64

Which is better for Stockfish? Ryzen 9 5950 or Intel 12900K?

blueemu
pfren wrote:

But you can get excellent results (easily some 14M nodes per second) with a cheap FX-8350 for just $190. Sure enough the FX-8350 is a weaker CPU than i7 3930k, but the difference when it comes to chess is too small to justify the huge price gap (the 3930 costs 2.5 times more).

The AMD FX series could be the ideal budget choice for chess servers, but unfortunately their architecture does not allow more than one CPU per rig (same applies for the expensive i7's).

I had an 8-core AMD Bulldozer back when I was working at the Tactics School, and I LOVED it.

For the price, it's almost impossible to beat.

It eventually blew a power supply and I bought a new i5 rig, but I still remember that 8-core AMD with fondness.

play4fun64
play4fun64 wrote:

Which is better for Stockfish? Ryzen 9 5950 or Intel 12900K?

Someone answer please.

anonym_anonym1
play4fun64 написал:
play4fun64 wrote:

Which is better for Stockfish? Ryzen 9 5950 or Intel 12900K?

Someone answer please.

better and fast intel 12900k

Fluffster_56-inactive
mldavis617 wrote:

I think the basic problem is that most chess engines are written by amateurs, a few of them being commercialized. You'll usually see the programmer's name associated with any engine you find, including Houdini. The problem with off-loading calculations to the GPU is that the program must have been written to accommodate that. It isn't, as far as I know, simply a function of the CPU seeing all that GPU horsepower out there and sending the load away for faster computation. The program must be written to utilize the GPU, and not knowing what GPU is available since it varies with the system, I don't think our engines are (yet) capable of speeding up with a powerful GPU onboard.

This is the primary function of the engine framework for the neural networking of Lc0. The search protocols utilize the GPU for calculations. 

tygxc

Carlsen, Caruana, and Nepo rented cloud engines to prepare for their World Championship matches. A cloud engine reaches 10^9 nodes/s, that is 1000 x faster than your desktop and 100 x faster than the top CPU discussed here.

play4fun64

Cloud engines seems more economical than buying expensive CPU like Threadripper. The TR is Overkill for  office, home and gaming application..

xor_eax_eax05

 I dont think anybody needs a Threadripper to analyse their games at amateur level. When you carry out engine analysis on your games, what you want is depth. 

 

 And the processor determines how fast you get to that depth. It's not that the lines Stockfish 15 will offer you at 35 depth will be lower quality if you were using a dual core processor compared to a Threadripper. It's just it will take longer for an ancient dual core to reach 35 depth, that is all. 

 

 For example I have recently upgraded my gaming rig to a 5950X from AMD, and it's great for reaching higher depths in shorter time, I use it a lot for my Daily chess analysis from Gameknot. Previously, a 3900X was good enough too. 

 I also own a 3950X in my Debian Linux box, and it is just as good for chess analysis.

 

 

play4fun64

Just curious. Anyone have 6 men EGTB installed in their PC and phone. How long does it takes to calculate a 50 move checkmate?

tygxc

#46
To analyse amateur games an ordinary laptop or desktop is enough, but for ICCF correspondence chess or grandmaster preparation a cloud engine may be most economical: rent a top engine instead of buy a mediocre engine.
#47
You can access cloud syzygy 7-men endgame table bases from PC and phone.

tygxc

#49
They play 5 days per move in the ICCF world championship.
One such tournament means 16 games at the same time and lasts for about 2 years.
With 3 tournaments at the same time that may lead up to 48 games.
They use more than one engine for one game for different perspectives.
So they run several threadripper style engines in their basement.
Cloud engines may be more efficient and less costly.

play4fun64

@tygxc

What is the recommended CPU and GPU for Video Editing and CAD? Surely they don't need Threadripper and RTX 3090Ti.