Just figured I'd post some results of a little uncore bump. It's only a single bench, but I think it demonstrates the importance of uncore overclocking.
I ran some benches for Idontcare to see how my PhI and PhII stacked up against his Q6700 (I think that's what he has). As the cores of the PhII got pushed to higher speeds I think the 1.8GHz uncore can start to hold the cores back, not feed them so to speak.
IDC has a bench he runs that crunches some numbers, 28 jobs. The faster your CPU is the faster the bench finishes.
For this run I had four instances going at once (one for each core). The cores were at 3.7GHz, the uncore untouched. The benches finished from 8:47 - 8:50.
With this run I had the uncore at 2070MHz, but the core at just 3.4GHz. Even with the cores having a ~300MHz disadvantage the jobs finished in 8:17 - 8:19. I think as the cores got to higher MHz the uncore couldn't feed them, so the scaling gets pretty poor.
And here I ran the uncore at 2070MHz with the cores at just under 3.7GHz.
So, unless my math is off (very possible
) the moderate overclock of the uncore to 2070Mhz gained about a 10% speed advantage at the same core speed (477 seconds for the first finished job at 2070MHz/3.68Ghz vs. 527 seconds for the first finished job with the chip at 1.8GHz/3.7GHz.
I'm probably not telling you guys anything you didn't know, but a moderate overclock of the uncore delivered 10% better performance clock for clock. So, if you were just using your multiplier to overclock you PhII it's a good idea to play around with you HT speed as there is more oomph you're leaving on the table otherwise.
Nothing earth shattering here, but just wanted to share some results incase anyone was curious aobut what overclocking the uncore could provide in performance. Any thoughts pleae feel free to share.
I ran some benches for Idontcare to see how my PhI and PhII stacked up against his Q6700 (I think that's what he has). As the cores of the PhII got pushed to higher speeds I think the 1.8GHz uncore can start to hold the cores back, not feed them so to speak.
IDC has a bench he runs that crunches some numbers, 28 jobs. The faster your CPU is the faster the bench finishes.
For this run I had four instances going at once (one for each core). The cores were at 3.7GHz, the uncore untouched. The benches finished from 8:47 - 8:50.
With this run I had the uncore at 2070MHz, but the core at just 3.4GHz. Even with the cores having a ~300MHz disadvantage the jobs finished in 8:17 - 8:19. I think as the cores got to higher MHz the uncore couldn't feed them, so the scaling gets pretty poor.
And here I ran the uncore at 2070MHz with the cores at just under 3.7GHz.
So, unless my math is off (very possible
) the moderate overclock of the uncore to 2070Mhz gained about a 10% speed advantage at the same core speed (477 seconds for the first finished job at 2070MHz/3.68Ghz vs. 527 seconds for the first finished job with the chip at 1.8GHz/3.7GHz.I'm probably not telling you guys anything you didn't know, but a moderate overclock of the uncore delivered 10% better performance clock for clock. So, if you were just using your multiplier to overclock you PhII it's a good idea to play around with you HT speed as there is more oomph you're leaving on the table otherwise.
Nothing earth shattering here, but just wanted to share some results incase anyone was curious aobut what overclocking the uncore could provide in performance. Any thoughts pleae feel free to share.
Facebook
Twitter