TAGG Posted June 21, 2017 Share Posted June 21, 2017 I'm currently trying to figure out, why my efficiency is so bad when benching singlecore chips: TAGG`s SuperPi - 1M score: 16sec 656ms with a Celeron 420 With Dualcore CPUs the setup keeps up with submissions that have more or less the same core and RAM clocks/ timings, with this Celeron however the efficiency is terrible. Tweaks done: Maxmem 600 PCI lock Performance for Background tasks RAM as Systemcache (large system cache) Pagefile 512 MB Realtime Priority Green Theme Low resolution and color depth during bench Superpi files on individual partition kill explorer no wazza I'm sure I' missing some major tweak for singlecore chips here, since there is allmost a second difference... Hope someone is willing to help me Quote Link to comment Share on other sites More sharing options...
TASOS Posted June 22, 2017 Share Posted June 22, 2017 What i could think of , is Memory sub-timings (not the primary ones) ... and bios version. Quote Link to comment Share on other sites More sharing options...
Massman Posted June 22, 2017 Share Posted June 22, 2017 If I remember correctly, for Wolfdale processors SuperPI 1M would have a good boost if you use a BIOS that has a microcode that does not support the Wolfdale processors. Quote Link to comment Share on other sites More sharing options...
TAGG Posted June 22, 2017 Author Share Posted June 22, 2017 @Massman I know about the wolfdale bug, it only works on 45nm though, this Celeron is a 65nm Chip, will try different BIOS versions anyhow, once i have time (maybe try removing the microcode for this Conroe-L from the BIOS would work?) @TASOS Subs are not increddibly loose, and as mentioned the same setup is way better with same timings and RAM clock when using a dualcore CPU... Hence me thinking it's an OS/BIOS problem... Quote Link to comment Share on other sites More sharing options...
TaPaKaH Posted June 22, 2017 Share Posted June 22, 2017 (edited) If the computation size (1M) is bigger than your L2 cache (512K), you need copywaza. Edited June 22, 2017 by TaPaKaH Quote Link to comment Share on other sites More sharing options...
TAGG Posted June 23, 2017 Author Share Posted June 23, 2017 Thanks a lot, once I'm done with Sandy i'll fire the REX up again and try using wazza even for 1M Quote Link to comment Share on other sites More sharing options...
elmor Posted June 24, 2017 Share Posted June 24, 2017 If the computation size (1M) is bigger than your L2 cache (512K), you need copywaza. 1M decimals vs 512K bytes? Quote Link to comment Share on other sites More sharing options...
Recommended Posts
Join the conversation
You can post now and register later. If you have an account, sign in now to post with your account.