Quad Crossfire X question

[GeforcerFX]

Hey i haven't been on for awhile, but i have been working in my system for most of the summer. I started with a amd790GX chipset board which had integrated graphics and i was trying to run 2 3870 x2's on this board, with my amd phenom II 940 X4 BE. I later found out i couldn't do that and ended u getting an older MSI board Its a 790X no integrated so quad CFX worked but i am getting weird results. In single card with everything stock clocked i get 16,000-16,500 in 3dmark06, on AMD 9.12, 10.12, 11.3, 11.6, and 11.7 drivers, when i add the second card i get form 16,500-17,000 in 3dmark06, i would figure i would see at least 3 to 4,000 point difference maybe get into the 20's once i overclock my 940 again, but this seems messed up, I have the GPus running on 2 450 watt PSU boosters (5.25 bay PSU's made for video cards) one for each card, and the system is powered by a 400 watt PSU, the boosters have 44 amps on the 12 volt rails, i should have plenty of power here and i am just thinking somethings off, OS is windows 7 64 bit, 5gb of ddr2 @ 667. Any suggestions?

 

[red1776]

Hi Ge,
first the basics, do you have CF enabled in catalyst? and a CF bride or dongle as the case may be? Use 'Afterburner' to check if the second card is showing activity.
Does Catalyst show all the GPU's as 'enabled adapter" ?

 

[dividebyzero]

Probably a CPU bottleneck.
3dMark06 is (I think) exclusively single-threaded, so CPU core speed is going to be the significant player in the equation.

Don't run the benchmark personally.

This seems like a timewarp -3dMark06 and 3870's - get's me all nostalgic for 2008.

BTW: interesting if puzzling moniker you've got there considering your system.

@red
Glad to see you showed up. You think our OP can squeeze out a points jump at stock CPU clock ?

****EDIT***
Just did a little Googling. Maybe a bit of a bump with the second card added

 

[GeforcerFX]

red everything is setup correct correctly and enabled, i have built lots of computers for people and this is my first multi gpu system,a nd yes its wierd i have that name and using ati stuff i know, my other system is a phenom x3 and geforce gtx260, so it balances out lol. Yeah its bit old but u gotta admit that quad 3870s was overkill in 2008 they can finally be utilized today, and they were pretty cheap. also i use afterburner everyday on all my computer lol, i love that thing, all 4 gpus are running at a percentage with my master card doing the most work. but on my older board i remember getting 3.6 on my 940 and overclcoking the 3870 x2 in single card and overclocking that aswell and getting a 18,000 score. If it was single threaded a 3 ghz cpu would still go pretty good, and 06 is multi threaded 05 was single threaded if i remember correctly. Thanks for the responses guys.

 

[GeforcerFX]

some new scores i reinstalled 11.7 drivers, and overclocked from 3ghz to 3.4ghz using multiplyer only stock FSB. In single card i got 16730, in crossfire x i got 18,286, maybe the cpu was bottlenecking slightly, idk though @ 3.4ghz it scored 4600 on cpu score so thats pretty fast in my book, but it just seems like i should be getting more.

 

[dividebyzero]

...and 06 is multi threaded 05 was single threaded if i remember correctly. Thanks for the responses guys.

Just for clarification.
3dMark06 is multi-threaded on the CPU tests, single-threaded on the graphics tests afaik

Easy enough to check- run the graphics tests and check your core utilization/resource monitor. If the graphics tests are single core (as I suspect) then you'll be CPU bound to an extent.

 

[red1776]

@red
Glad to see you showed up. You think our OP can squeeze out a points jump at stock CPU clock ?

Hi Chef,
I'm a bit puzzled actually. should be a decent bump (%20 or so) if memory serves. it has been a while since I had the 3870's fired up though.

 

[red1776]

I found this that seem to jibe with my recollection of things

http://www.tweaktown.com/articles/1319/radeon_hd_3870_x2_in_crossfire_x_tested/index2.html

 

[dividebyzero]

Same link I posted back in #3

BTW: Anyone with a full version of '06 out there who can verify if the graphics tests are single-threaded?

 

[red1776]

Same link I posted back in #3

BTW: Anyone with a full version of '06 out there who can verify if the graphics tests are single-threaded?

Oh....well then...great minds and so forth :wave:

Here is what she looks like CPU wise

 

[dividebyzero]

Looks single-threaded to me taking into account Win7's attempts to get the other cores involved. The spike on the fifth core correlates with the dip on the second core.

This being the case then you'll have an awful lot of GPU computation waiting around on one cores worth of cycles.
This is one of the graphics tests I presume.

 

[red1776]

yup, Graphics test 1. they all look similar however.

Here is an odd looking graph I thought. Single threaded and CPU dependent. Same test (thats riding 22% usage on all four GPU's)

 

[dividebyzero]

It's about what you'd expect with all four GPU's waiting around (to some degree) in the queue waiting on available CPU core cycles. The CPU bottleneck is effectively rendering most of the GPU idle.

Just for Sh*** and giggles you could run the CPU at stock (or underclock) and re-run the bench. My guess is that you'll find the GPU usage drops accordingly.I'd also venture that if you run a graphics test without hardware physics (assuming there is such a test or physics can be disabled) then the CPU core usage on 4 and 6 will probably drop - I'm guessing that this is at least partially causing the activity on those CPU cores.

 

[red1776]

It's about what you'd expect with all four GPU's waiting around (to some degree) in the queue waiting on available CPU core cycles. The CPU bottleneck is effectively rendering most of the GPU idle.

Just for Sh*** and giggles you could run the CPU at stock (or underclock) and re-run the bench. My guess is that you'll find the GPU usage drops accordingly.

Thats what i thought, however, It seems to me that other single threaded games have loaded the GPU's much more substantially than this. shattered Horizon comes to mind.

 

[dividebyzero]

Umm, maybe that's why 3dMark06 is to graphics testing what the Trabant is automobile evolution.

3dMark06 is usually used exclusively as a CPU test. I certainly wouldn't use it for benchmarking graphics cards (thats what 3dMark11 is for!)

 

[red1776]

Umm, maybe that's why 3dMark06 is to graphics testing what the Trabant is automobile evolution.

3dMark06 is usually used exclusively as a CPU test. I certainly wouldn't use it for benchmarking graphics cards (thats what 3dMark11 is for!)

And not a bad CPU test either!

Thats better.

 

[GeforcerFX]

thats about what my graphs were lookin like, but when it got to the cpu tests all 4 core went waaayy up and all 4 cores were used, but graphics test yes i agree once core working the rest siting around my 4 gpus really never went over 30-50%, i think i need to get vantage, since if i remember correctly i can't run 3dmark11 on 3 series cards.