7950X vs 13900k for chessprogramming

Discussion of anything and everything relating to chess playing software and machines.

Moderator: Ras

dangi12012
Posts: 1062
Joined: Tue Apr 28, 2020 10:03 pm
Full name: Daniel Infuehr

7950X vs 13900k for chessprogramming

Post by dangi12012 »

I just realized that 13900k supports DDR4 AND DDR5 making a platform upgrade much much cheaper than expected.
Running costs will be higher with 13900k. Performance seems to be about equal.

Which CPU would you prefer for chessprogramming?
Worlds-fastest-Bitboard-Chess-Movegenerator
Daniel Inführ - Software Developer
User avatar
Fabio Gobbato
Posts: 219
Joined: Fri Apr 11, 2014 10:45 am
Full name: Fabio Gobbato

Re: 7950X vs 13900k for chessprogramming

Post by Fabio Gobbato »

But the intel cpus with performance and efficient cores can add more noise to the tests?
If you are testing a patch and a version of the engine runs on a performance core and another on an efficient core does they affect the results of the tests?
I haven't an intel cpu but I would like to know how they work with engine testing.
Raphexon
Posts: 476
Joined: Sun Mar 17, 2019 12:00 pm
Full name: Henk Drost

Re: 7950X vs 13900k for chessprogramming

Post by Raphexon »

Fabio Gobbato wrote: Fri Oct 21, 2022 1:38 pm But the intel cpus with performance and efficient cores can add more noise to the tests?
If you are testing a patch and a version of the engine runs on a performance core and another on an efficient core does they affect the results of the tests?
I haven't an intel cpu but I would like to know how they work with engine testing.
If that was such a big issue people should turn off SMT too.
It doesn't really matter.

But for chess you're trading somewhat less upfront cost for lower efficiency and slightly worse performance.
Assuming they're both running full throttle the 13900K will cost €1.5 to €2 more per day.
So that lower up front cost is quickly eaten up with a higher bill.
Provided you're running tests non-stop.
dangi12012
Posts: 1062
Joined: Tue Apr 28, 2020 10:03 pm
Full name: Daniel Infuehr

Re: 7950X vs 13900k for chessprogramming

Post by dangi12012 »

Raphexon wrote: Fri Oct 21, 2022 2:13 pm
Fabio Gobbato wrote: Fri Oct 21, 2022 1:38 pm But the intel cpus with performance and efficient cores can add more noise to the tests?
If you are testing a patch and a version of the engine runs on a performance core and another on an efficient core does they affect the results of the tests?
I haven't an intel cpu but I would like to know how they work with engine testing.
If that was such a big issue people should turn off SMT too.
It doesn't really matter.

But for chess you're trading somewhat less upfront cost for lower efficiency and slightly worse performance.
Assuming they're both running full throttle the 13900K will cost €1.5 to €2 more per day.
So that lower up front cost is quickly eaten up with a higher bill.
Provided you're running tests non-stop.
Oh it does matter it's not like hyperthreading at all. People should not run cutechess on all cores!
Actually it's a huge issue. If running multiple engines and some get pcores and some weak cores the results will be wrong. 1v1 with ponder off is okay.
13900k and 1.5€/day will be 150€/year with non 24/7 use case.

Only thing that bugs me with Ryzen is higher Mainboard cost and forced ddr5 upgrade instantly.
Worlds-fastest-Bitboard-Chess-Movegenerator
Daniel Inführ - Software Developer
Modern Times
Posts: 3782
Joined: Thu Jun 07, 2012 11:02 pm

Re: 7950X vs 13900k for chessprogramming

Post by Modern Times »

dangi12012 wrote: Fri Oct 21, 2022 9:00 pm Only thing that bugs me with Ryzen is higher Mainboard cost and forced ddr5 upgrade instantly.
Yes but at least there is some chance of you being able to carry that DDR5 over to your next upgrade.

I believe you can disable the eCores on the Intels, I could be wrong. If chess testing was the use of the machine 24/7, it could make it a viable option.
Werewolf
Posts: 2058
Joined: Thu Sep 18, 2008 10:24 pm

Re: 7950X vs 13900k for chessprogramming

Post by Werewolf »

dangi12012 wrote: Fri Oct 21, 2022 9:00 pm
Raphexon wrote: Fri Oct 21, 2022 2:13 pm
Fabio Gobbato wrote: Fri Oct 21, 2022 1:38 pm But the intel cpus with performance and efficient cores can add more noise to the tests?
If you are testing a patch and a version of the engine runs on a performance core and another on an efficient core does they affect the results of the tests?
I haven't an intel cpu but I would like to know how they work with engine testing.
If that was such a big issue people should turn off SMT too.
It doesn't really matter.

But for chess you're trading somewhat less upfront cost for lower efficiency and slightly worse performance.
Assuming they're both running full throttle the 13900K will cost €1.5 to €2 more per day.
So that lower up front cost is quickly eaten up with a higher bill.
Provided you're running tests non-stop.
Oh it does matter it's not like hyperthreading at all. People should not run cutechess on all cores!
Actually it's a huge issue. If running multiple engines and some get pcores and some weak cores the results will be wrong.
For similar reasons I don't use Intel with Aquarium. How would I know which positions get a P core and which ones would get an E core? Nightmare.
Modern Times
Posts: 3782
Joined: Thu Jun 07, 2012 11:02 pm

Re: 7950X vs 13900k for chessprogramming

Post by Modern Times »

Werewolf wrote: Fri Oct 21, 2022 10:39 pm For similar reasons I don't use Intel with Aquarium. How would I know which positions get a P core and which ones would get an E core? Nightmare.
Depending on the motherboard manufacturer I think it may be possible to disable E cores in the BIOS.
smatovic
Posts: 3472
Joined: Wed Mar 10, 2010 10:18 pm
Location: Hamburg, Germany
Full name: Srdja Matovic

Re: 7950X vs 13900k for chessprogramming

Post by smatovic »

dangi12012 wrote: Fri Oct 21, 2022 9:00 pm ...
Only thing that bugs me with Ryzen is higher Mainboard cost and forced ddr5 upgrade instantly.
You will get the new AM5 socket, and AMD is keen to keep their sockets as long as possible alive, Intel not that much, AM4 and AM5 coolers are also compatible for example.

I really do not like the division into performance and efficieny cores in these new Intel/Apple CPUs, makes sense in a SmartPhone IMO, not in my desktop/workstation.

--
Srdja
User avatar
Ras
Posts: 2720
Joined: Tue Aug 30, 2016 8:19 pm
Full name: Rasmus Althoff

Re: 7950X vs 13900k for chessprogramming

Post by Ras »

The 13900K and the 7950X can both be configured to more reasonable TDP settings without losing much performance. The 230W / 300W issue boils down to silly default bios settings so that any argument to electricity cost is baseless. At 90W power, both deliver similar overall performance. Also, the efficiency is comparable to Apple silicon. The 7950X consumes 62% less power than at 230W while only losing 18% of performance - and it's still 22% faster than the 5950X at 140W.

However, the 13900K gets its multithreaded performance as presented e.g. in Blender benchmarks thanks to its E cores - and I agree with the other posters that mixing P and E cores for engine matches would skew the results. So you can try to lock the engines under test to certain cores which would solve that issue, but then you also need to remember adjusting the time controls relative to the core sort used. Otherwise, you'd mix up the results for different time controls.

Or you can disable the E cores in the bios entirely, but then you also remove most of the 13900K's multithreaded performance, making it a stupidly overpriced octacore. At that point, you could as well buy a 7700X.

While Intel currently offers better value in general thanks to the lower platform cost, it's not a good offer for chess programming specifically. Ryzen 7000 is a commercial flop so far, due to the platform cost and particularly the overpriced mobos, but it's an actually useful proposition for chess programming. If you mind the platform cost, the 5950X would be a better alternative to the 13900K, at the price of no further upgrade path ofc, but since Intel's 13th gen is already the refresh, upgrading to 14th gen will probably also require a new mobo, given that Intel changes sockets every other generation.

Or you just wait a few weeks or months for the Ryzen 7000 platform prices to go down and also have a few bios updates to deal with the usual initial problems.
Rasmus Althoff
https://www.ct800.net
CornfedForever
Posts: 650
Joined: Mon Jun 20, 2022 4:08 am
Full name: Brian D. Smith

Re: 7950X vs 13900k for chessprogramming

Post by CornfedForever »

Ras wrote: Sat Oct 22, 2022 1:10 pm

Or you just wait a few weeks or months for the Ryzen 7000 platform prices to go down and also have a few bios updates to deal with the usual initial problems.
What is frustrating to me....as someone looking for a NEW system and not wanting to go thru the hassle of building it...is that I can't find any pre-built 7000 platform builds anywhere.
Well, think I saw some paired with the most expensive gaming GPU's...but I really only want a 1060 ti at most...