What are everyone's dual core opteron SuperPi 32M/1M times?

trodas said:

My sub 30 result on 2800Mhz on Sapphire Pure Innovation "Grouper" A9RX480. I can do it way better, as I reached 3Ghz too ( http://valid.x86-secret.com/show_oc.php?id=148241 ) but unfortnately I killed my opty by wrong voltage selection in bios - 2.0Vcore.
Fanless watercooling, not naked die. Could run faster for sure, but I was just... stupid

Click to expand...

fabulouscoops said:

Someone else with too much time.

Here is a study of Super_pi 1M times on my single core 3700+ (first rig in sig) using different RAM dividers. We all know Super_Pi is related to CPU frequency but this is fixed at 264 x 11 = 2.904 GHz. RAM timings are 3-4-4-8-1T. Times were measured at different RAM dividers. The graph is the calculated RAM frequency (HTT x RAM frequency) vs 1M times in sec.



Ratio, Divider, Frequency,1M time
1:1, 200 MHz, 264 MHz,29.7 sec
9:10, 180, 238,30.5
5:6, 166, 219, 31.1
3:4, 150, 198, 31.7
2:3, 133, 176, 32.5
1:2, 100, 132, 34.5

Super PI Ver1.1e (modded to display millisec) Numbers are rounded to nearest tenth of a sec. Frequency rounded to nearest MHz

Click to expand...

malvindo said:

using divider should bottleneck memory to fsb, but if running dual channel i think the different should be minimal
if fsb 200 n memory 200 (single channel)so cpu bandwith is 3.2gb & memory bandwith is 3.2gb/s
if fsb is 300 using divider n memory(single channel) is 200 so cpu bandwith is 4.8gb/s & memory bandwith is still 3.2gb/s
if fsb 300 using divider n memory is 200 (dual channel) so cpu bandwith is 4.8gb/s and memory bandwith is 2x 3.2gb/s = 6.4gb/s

pls correct me if i am wrong

PS:fabulouscoops, u using single channel or dual channel when do the test?

Click to expand...

fabulouscoops said:

This was done with dual channel. The only parameter that was changed between runs was the memory frequency.
I am not understanding what you are saying about dual channel vs single. If all the measurements were taken using single channel how would that change the interpretation of the results? Slower memory = higher super_pi times. Bandwidth was not changed from one run to the next.

Click to expand...

fabulouscoops said:

This was done with dual channel. The only parameter that was changed between runs was the memory frequency.
I am not understanding what you are saying about dual channel vs single. If all the measurements were taken using single channel how would that change the interpretation of the results? Slower memory = higher super_pi times. Bandwidth was not changed from one run to the next.

Click to expand...

fabulouscoops said:

I see what you are saying. I may have been confusing terms. These were all done with dual channel but I could try with single and report back. I think that would be an interesting study and put to bed the question of dual vs single channel in AMD systems.

Unfortunately, I no longer have the 3700+ I used for the above tests since I upgraded to an opteron dual core. So I may need to repeat everything.

Click to expand...