should we add best affinity settings for each game?

[handlingcfg 0]

hi, this topic is about the affinity settings on game performance. if fps limited by cpu, it can give more fps with only setting different affinity.

example far cry 3, with ryzen 5 3600

when if open game directly (6 core+6 smt), i get 110-111 fps, if i set affinity on only 3-4 real cores, i get 142 fps.  we would most likely to experience same kind differences with  future (ie 128 real core) cpu's on todays games.

i wrote sloppy page about topic. https://www.pcgamingwiki.com/wiki/Multi-Threading_on_game_performance

so im thinking maybe we need to add affinity settings glossary to each game's page. what do you think about that, do you think its logical and benefical? or not?

 

[Patrxgt 7]

Good idea, but I think it should be added only to games, that are having problems like that and don't have any other way to fix it.

[evpointmaster 0]

Good idea in theory, but I think it will be a huge amount of work. You will need to test all configuration for each game across different CPUs to make sure there is a consistent pattern for set affinities.

There are also many variable besides HT/SMT, like P&E cores on Intel or multiple dies on AMD, which usually lower performance for games, but might increase performance for games that utilize many threads. And then, like you mentioned, there are differences between setting affinites and changing settings in BIOS, which for example on Intel 12th gen CPUs makes a big difference, because it results in considerably higher cache frequencies when E-cores are disabled.

[handlingcfg 0]

10 hours ago, evpointmaster said:

Good idea in theory, but I think it will be a huge amount of work. You will need to test all configuration for each game across different CPUs to make sure there is a consistent pattern for set affinities.

There are also many variable besides HT/SMT, like P&E cores on Intel or multiple dies on AMD, which usually lower performance for games, but might increase performance for games that utilize many threads. And then, like you mentioned, there are differences between setting affinites and changing settings in BIOS, which for example on Intel 12th gen CPUs makes a big difference, because it results in considerably higher cache frequencies when E-cores are disabled.

i'm more practical man, i dont think in needs %100 accurate and people needs to give huge amount of work, my testing takes at most 5 min on far cry 3.

far cry 3 benefits 3 threads and you have 3 or more cores, than you need to assign 3 real cores, This is the most important part, with this only i get   111 fps to 142 fps

if someone wants to get the scientificially best way possible, than the other things like multiple dies etc needs to be calculated, but this part is not necessary and not logical to testing hours for 3-5 fps difference.

alternatively, someone with huge amount of cores, 16 core or even 64 cores, could test few games and he/she can write this game works best on xx threads. if you have less than xx threads, than you dont gain anything.

 

and last part of problem: windows

game uses 3 cores, it makes calculations and makes their cache, then windows decides its best to move calculations on another thread and its generally random. so you loose  cache for a while and loose performance. even limiting windows gets users to gain.