#why stockfish 16.1 only gives me around 12,000 kn/s on i914900k but give me 30,000 kn/s on i913900k?

33 messages · Page 1 of 1 (latest)

dusk gorge
#

are they the same version of stockfish?

tender lion
#

yes, 16.1 Avx2

dusk gorge
#

it says "Stockfish 16" on the top right corner there, are you sure that it's the latest version?

dawn hornet
#

try with a benchmark like cinebench to see if 13900 and 14900 have similar score.

Also 13900 and 14900 have 32 logical core (24physical), so the best would be to use 32cores in my opinion.

tender lion
#

thanks for replies though

unborn oasis
#

your system is not stable

#

what cooler are you using and are you overclocking

#

try running some stress test programs

#

or try a different gui

tender lion
#

I'm not overclocking.

#

Yes. I suspect that it's about system as well. I will try test it.

tacit surge
#

16x1
Two sticks is ideal for dual channel* mode. Does the other machine have two DDR5 sticks?

tender lion
#

yes

#

32gb ram

tacit surge
#

That's likely the problem (or at least one problem), the older system has double the ram bandwidth.

tender lion
#

Thanks for pointing that. I suspect that too.
So you mean ddr4 could be better? why is that. what is the possible fix?

#
  • The papar is wrong. I have 2 ddr5 sticks
tacit surge
#

No, I mean running two phyical sticks of ram allows doubling your RAM bandwidth on most CPUs.

#

You have two physical sticks on the 14900 system?

tender lion
#

Yes

#

it crash with 24 cpus. sometimes even crash with 14 cpu 😦

tacit surge
#

Have you done any other benchmarking? Or temperature monitoring?
Also chessbase probably shouldn't crash like this, not exactly sure what that means.

tacit surge
#

Well you have liground set to small hash, so if it is ram bandwidth related it would help hide it. Try increasing it. Though maybe not impossible CB is somehow screwing something up.
What slots are your ram sticks in?
And I meant benchmarks not directly related to SF. Like something that measures CPU perfomance and something that measures RAM bandwidth.
Can you view temperature and power limit while running too?

dawn hornet
#

maybe your chessbase is using a corrupted tablebase like gaviota or syzygy tablebase

tender lion
tender lion
tacit surge
#

Does it run the right speed in liground?

tender lion
#

yes

tender lion
#

not sure where to look at cpu limit

tender lion
#

Sorry this is more relate to Chessbase than stockfish, still I want to resolve it.

tacit surge
#

Are you sure you are using the same version of SF on all machines or are you using an included version from a GUI if available?
Also your liground recording had 32 threads instead of the 24 you are claiming you used on everything else. Please make sure things like threads and hash match. Also confirm that liground doesn't also perform better on your 13900k, since it only got around 24M in the recording with small hash.
Also hwinfo >>> hwmonitor imo. And can you confirm the memory mode on the main screen? Also the mem frequency. You may want to enable XMP if it isn't already.
Also fwiw BMI2 is probably slightly faster for both, and for 14 further gain with AVX512 AFAIK. Although you should still be faster on the 14 even with both using the same version.

tender lion
#

how to delete entire post?

dusk gorge
#

You can do in the dropdown menu after clicking on the three dots on the top right corner