Hold your horses, 27 is 35% faster than 20. That is significant.
How to optimize a build for a chess engine?
- Thread starter BillFoster
- Start date
But shouldn't a 13900k be more than 35% faster than a 9900k? Keeping in mind that Stockfish 15 is highly threaded.
Obvious question, are all the cores busy, and what clock speeds/temps are you seeing?
I'm fully utilizing 24 cores over 24 threads of Stockfish. The 13900k is thermally throttling, and is around 300W TDP. Stockfish is utilizing 44GB of RAM for the hash table. P-cores are running at 5.5Ghz.
It's a pretty insane level of computation.
Is it any better with 32 threads, to use 16 threads on the hyperthreaded P cores, and run another 16 threads on the E cores? I'd expect it to be more power efficient for the integer parts of the computation, which might help if you are thermally limited.
From what I've read, Stockfish doesn't do well with hyperthreading. I'm not sure about the technical details, but nobody does it so I assume it's not very performant.
It's possible, I suppose. But I'm using DDR5-4800. That's not super fast, but it's not like crazy slow, either.
Yet some benchmark sites (https://openbenchmarking.org/test/pts/stockfish) report a 13900k getting almost 68 million nodes/second. That makes no sense. Okay, maybe I'm not using super fast benchmark memory. But is going to DDR5-6000 going to give me 2.5x performance? I severely doubt it.
It just doesn't make sense to me.
EDIT: And I am running the memory in dual channel mode. It's 32GB x 2.
Well, according to the tests here, a 12900K is 25% faster than a 10900K at Stockfish 15. So your numbers are as expected.
Coming in realllllly late here.... Assuming you are a member of chess.com have you tried the Stockfish analysis if a game/position you upload?
Edit: in playing with this. it looks like it is running on my laptop in Javascript within the browser. What you have is most likely vastly better.
Edit: in playing with this. it looks like it is running on my laptop in Javascript within the browser. What you have is most likely vastly better.
Last edited:
I am a chess.com member, and I've used their Javascript Stockfish analysis. I mean it's better than nothing, but yes I can get something like 10x as many nodes/second running Stockfish though Chessbase.Coming in realllllly late here.... Assuming you are a member of chess.com have you tried the Stockfish analysis if a game/position you upload?
Edit: in playing with this. it looks like it is running on my laptop in Javascript within the browser. What you have is most likely vastly better.
EDIT: If it matters, I also use Chessbase because I often do parallel analysis through Stockfish and also LCZero. They occasionally differ, so I like to have both. And LCZero is really light on the CPU, heavy on GPU, so running them in parallel is not really a problem.
Could you please share which chest program you are using to run stockfish on your computer? I'm asking cuz I also like chess.
I use Chessbase 17. I think it's pretty good. I know people who complain that it's too expensive but... I dunno. It's all in how much you use it, right?
I use Chessbase 17...every day. Largely for analysis of games I've played on Playchess, chess.com or lichess. Love it!
I am also about to get a new computer. I am seriously thinking of the AMD Ryzen 7945HX, the laptop 'equilvilant' of the desktop 7950x. I want the lower power draw and many reviews pegs things like Cinebench 23 close to the desktop CPU version even when using a lot less power/heat. As with the Intel, diminishing returns set in real quick in these high end processors and you are mostly burning energy for very little benefit.
I considered the (non X) version of the 7900 desktop as with it you get the efficiency goodness ...and could always overclock (with proper cooling- don't need a watercooler with it out of the box) if one wished, but kind of miss having a laptop.
I am also about to get a new computer. I am seriously thinking of the AMD Ryzen 7945HX, the laptop 'equilvilant' of the desktop 7950x. I want the lower power draw and many reviews pegs things like Cinebench 23 close to the desktop CPU version even when using a lot less power/heat. As with the Intel, diminishing returns set in real quick in these high end processors and you are mostly burning energy for very little benefit.
I considered the (non X) version of the 7900 desktop as with it you get the efficiency goodness ...and could always overclock (with proper cooling- don't need a watercooler with it out of the box) if one wished, but kind of miss having a laptop.
Hi I just bought ASUS ROG Strix G15 laptop - AMD Ryzen 7, 6800, 16 GB Ram. I bought it for other games and not only chess but I am getting only around 7000 Kn/s on my Chessbase 15 with Stockfish 16. Is this normal. I was expecting something like 20 000 Kn/s. What can I do to bring this up?
hi, how many kn/s do you getting in fritz or in chessbase on stockfish 16?
Since this thread was resurrected anyway, I'll go ahead and reply to this 18-month-old comment: my guess would be that your cooler isn't keeping up. You mention that you're thermally throttled, so a better cooler is certain to improve performance. How much performance, I can't tell, but you'll gain at least something if you can keep the chip under its throttle temp.
That plus faster DDR5 might improve your benchmark results, maybe even dramatically, but I strongly doubt you'd triple performance. That person might be using liquid nitrogen cooling or something similarly weird. Or they could just be cheating. People are weird like that.
Also have you checked which CPU compiled version of stockfish you are using?
I assume a 13900k should be the AVX2 version rather than the standard 64bit one. An AMD 7900 can get 20M n/s with the standard 64bit version, or 36M with AVX2, and peaks as high as 40M with AVX512.
I assume a 13900k should be the AVX2 version rather than the standard 64bit one. An AMD 7900 can get 20M n/s with the standard 64bit version, or 36M with AVX2, and peaks as high as 40M with AVX512.
Last edited: