PCIe Speed Test v0.1 CPU-GPU/GPU-CPU bandwidth test

**Stoner133** · 06-04-2009, 10:24 AM

Mechromancer I did as you said and ran it this time directly from the Command Prompt and low and behold the screen doesn't close when the test completes as can be seen here:

By knuckles6 at 2009-06-04

The only thing this screen shot doesn't display correctly is GPU activity in Catalyst Control Center but thats only because the screen shot is taken after the test completes. That actual activity hits 98% during the test.

**Mechromancer** · 06-04-2009, 11:19 AM

You have some huge bandwidth on that rig. Way to go! I've decided it would be more than worth it to go with an X2 card the next time I upgrade. I'll have to see what these DX11 cards can do when they appear in a few months. I love my single GPU, but theres something to be said for single-slot Crossfire configurations. The performance rocks and you have the flexibility of using one or both GPUs as needed.

BTW, Beyond3D Update: http://forum.beyond3d.com/showthread.php?t=54321. They have the luxury of getting a bit more specific, technical information from people in-the-know about these things.

**Stoner133** · 06-04-2009, 12:00 PM

My next cards will still be singles, just because I can use the waterblocks I currently have on them.

**andinobi** · 06-11-2009, 08:41 PM

Try make BAT file to save result to notepad. (sory ,i dont read from page 1 to end) so i dont know if anyone also post

Another real world benchmark
http://graphics.stanford.edu/projects/gpubench/

(must make BAT too) but i like this program, because this can work on NVDIOA and ATI, and can find out your max GDDR dan Gigaflops.

Interisting when i tes 9600Gt on X58 and P35, bandwith x58 more bigger , so i think QPI bandwith is true working , with this program you can find out your max bandwith chipset.

remenber to make BAT file (or i uploaded to rapidshare gpu bench modifiend later)

sory if my english not good

**Cheffy** · 06-12-2009, 04:01 AM

just to report:
single 3870, boosting the htt/pcie clocks ~10% each resulted in very similar changes as aGeoM already graphed: cpu->gpu increased pretymuch 10% gpu->cpu didnt move atall (actual changes cpugpu +9% (from 5.197 to 5.628, gpucpu from 5.322 to 5.317. cpu/ram/nb held at 3250/1000/2250 htt/pcie from 2000/100 to 2250/110)
pcie 110 = 5.628/5.317
pcie 120 = 5.798/5.316
pcie 125 = 5.858/5.311(see below)
pcie 130 = 5.924/5.313
pcie 140 = 6.042/5.315
pcie 150 = 6.109/5.318
pcie 160 = 6.105/5.312
pcie 170 = 6.105/5.313
pcie 180 = 6.107/5.316
pcie 190 = 6.099/5.312
pcie 200 = 6.106/5.313 (is this a record for amd 770 nb pcie clock?)
so. it apears neither PCIE clock or HTT clock limits the GPU->cpu trafic. but Both, at the same time, have an effect on cpu->gpu traffic (perhaps during the buffered stage, in nb, that allows for unsynced clocks?) on my machine 125mhz pcie would be an exact ratio of htt clock, so...
nope. its in line.. next theory?

**badger620** · 07-16-2009, 01:21 PM

This is what I got with 2 3870 Toxic editions OC 860 both cores and 1265 mem on primary and 1230 mem on secondary. AMD Athlon 64 X2 6000+ OC to 3.4ghz with a multiplier of 13 and HTT Clock @ 260.9mhz and a HTT Speed 1304mhz. This was the first time that I ran the benchmark and got over 3 gb/s every other test was just under 3; I got above 3 when I raised the clock on the HTT Speed to 1304mhz which really didn't increase performance but everyso slightly but that is what it is about though getting the most that you can out of what you have right?

[img=http://img520.imageshack.us/img520/785/38382933.th.jpg]
[img=http://img520.imageshack.us/img520/633/30wlid.th.jpg]
[img=http://img520.imageshack.us/img520/9568/30w.th.jpg]