Intel 12900 series for chess engine testing

[lkaufman]

I noticed that the new series of Intel 12th generation processors, including the latest 12900 (H or K or HK), combines two different types of cores running at two different speeds. My question is whether this would make them unsuitable for chess engine testing where you might run 20 games at once on 20 cores (for example), each game on one core. Normally this is no problem on any computer prior to this series, as all cores run at about the same speed, but wouldn't such testing be highly distorted on these 12th generation processors if two very different cores are used? Am I missing something or is there some work-around?

[adnoh]

One possible work around is to disable the e-cores via the BIOS. Not sure whether all motherboards / BIOS supports this but when the 12th gen chips first became available, some game copy protection schemes were confused by the 2 types of cores and refused to run so the manufacturer's solution was to allow complete disabling of the e-cores. I don't follow this topic for the current situation as it is not important to me.

[Branko Radovanovic]

Or, alternatively: if for, say, engine testing one doesn't run more processes than the number of available performance cores (which is 8 for 12900K), is then one more or less guaranteed to actually use only the performance cores for that sort of workload? That seems plausible (since one would expect processes that use 100% of the CPU to "stick" to performance cores) and, if true, would be somewhat convenient as it would make it possible to get consistent performance even without disabling the efficiency cores.

[Jouni]

With hyperthreading on (default?) I quess you have 24 threads running at quite equal speed. Note ,that even 12600K has asmfish benchmark 36.187.026 for 16 threads and 300$ .

[yurikvelo]

To run many simultaneous games (1-thread or low-count thread) - need to rewrtite testing environment (like fishtest), so that all processes get affinity to the same core/core(s) and not shuffling during run.

Time control multipler calculation also should be tested separately on P and E cores and cutechess should be forked with different TC for E and P cores

[Jouni]

If performance cores run 5 Ghz that means 2,5 Ghz for hyperthreading. So Efficient-cores run faster at 3,2 Ghz!?