Originally Posted by Timeofdoom
Well, I ran cinebench when i had the 1 compute unit per module and it netted me 4.6 score, so 4,6/4 = 1.15.
I ran it Again with all cores enabled and I got 7.68/8= 0,96; so the trick still holds very true (1,15/0,96=~1,20 or 20 %).
I wish AMD would fix the FPU problem and give us another 8-core, even if on a new socket. (8x1,15=9,2)
But that's probably a pipedream - which reminds me: did Kaveri really fix the FPU problem?)
Run this test once more, but with all cores active in bios and then run a custom cinebench with only 4 threads. You will get almost the same score, in my case, it is slightly lower when disabling cores.
Also to reiterate my first point; that isn't a single threaded calculation you are getting by dividing your scores by how many cores you have active. Run a single threaded cinebench run and you will get near, if not the same single threaded score using 1c/1m and having all cores activated. It will then give you a scaling calculation as to how well the cores actually scale... 1c/1m scales very well (3.9ish/4), all cores activated should be around 6.9/8---> this is a guess as I can't remember the exact number I got.
My point is that by disabling the cores you don't gain anything. You will get the same score if you just leave them active. If a program only uses 4 threads, you will get the same performance with all your cores activated in the bios as you would going 1c/1m. Also, in my experience there is really no added overclocking headroom from disabling cores. So again, really no sense in disabling cores.Edited by Ultracarpet - 3/7/14 at 10:33pm