Thanks for nice information. I have a unsolved question though
(regarding this image)
Attachment 121329
Right after see this table, I guess there's some bug in a chip regarding some operation clock speed which sticks together with HTT, L2 speed for instance.
So I've tried the same test with a hope of making another excuse for Bulldozer's low IPC at default HTT.
My comparison settings are:
1. 3.6GHz (200x18) / NB/HT 2.4GHz (200x12) / DRAM 800MHz (200x4)
2. 3.6GHz (240x15) / NB/HT 2.4GHz (240x10) / DRAM 800MHz (240x3.33)
3. 3.6GHz (266x13.5) / NB/HT 2.4GHz (266x9) / DRAM 888MHz (266x3.33)
4. 3.6GHz (266x13.5) / NB/HT 2.4GHz (266x9) / DRAM 712MHz (266x2.66)
Note that for 3 & 4, dram speeds don't stick exactly with 800MHz, so I tested twice for FSB266 under two different dram speeds both distant samely from around 800MHz.
Anyway, my results are totally indifferent - means that higher FSB doesn't make sense at all
(I'm so disappointed that I cannot even make a graph......)
I tested Cache & Memory benchmark, CPU Queen, CPU AES, FPU Julia, FPU Mandel in AIDA64 and Cinebench R10 / R11.5.
Do you think you have any other variable which I (wprobably) missed? - I've only controlled CPU multiplier, NB/HT multipler and DRAM multiplier along with HTT clock.
Bookmarks