PSU limiting 4890's in crossfire?

[Dougx1317]

I borrowed a friend's 4890 to put in crossfire with my own to test it. It worked, and crossfire was enabled. Some of my benchmarks actually went down, and the one's that went up didn't get close to reviewers benchmarks. I was wondering if my PSU wasn't strong enough. Would a 700W FSP PSU limit an i7 set up with two 4890's in crossfire?

 

[paperfox]

Specs for the 4890 say 600w psu so i think you are good there.

Specifically which benchmarks? Maybe overclocking your i7 will help make up the difference in some as many people that benchmark overclock. Maybe double check that the settings where exactly the same.

 

[Dougx1317]

The settings were the same. OCing my i7 to 4Ghz helps a bit. CS:S went down. Crysis only went from 16FPS to 26FPS which is quite low for 1680x1050, Very High, 8x AA. Windows experience Index went down, but I don't know if that means anything.

What's concerning me is that this shows one 4890 getting 26FPS (vs 16), and 4870x2 getting 36FPS (vs 26 with my 4890's)
http://www.tomshardware.com/reviews/radeon-hd-5970,2474-8.html

I also know that my motherboard (GA-P55A-UD3) has a 8x/4x crossfire. But it still seems like low FPS. Benchmarks I ran with one 4890 switch between slots only showed a 15% decrease.

"I tested the XFX 4890 in both the PCI-express 16x and 4x slot. Here are the results. This was with the GPU benchmark in the Crysis demo with all settings to "Very High" at 1440x900. All numbers are the average of three runs. The stock settings were 850/975 and the OC settings were 925/1075.

16x.....= 28.17
4x......= 24.21
16x OC= 30.86
4x OC.= 25.93 "

 

[paperfox]

I see your problem now, its your motherboard / PCIE slots.

PCI Express And CrossFire Scaling: Is P55 Good Enough? : The So-Called Mainstream Solution?
http://www.tomshardware.co.uk/p55-pci-express-scaling,review-31780-7.html

All though this article uses only 5870s you will notice there is a slight difference between (LGA1366)x16/x16 and (LGA1156)x16/x8 crossfire. On top of that your motherboard only offers x16/x4 crossfire(x8/x8 would have been much better), in addition, if you turn to page 3 you will see that a single card in an LGA1156 x4 slot suffers alot compared to a LGA1156 x16 and LGA1366 x16 slot (I think somewhere in that article it says that the x4 slot is not realy ment to handle gaming).

This I can safly say is the reason why your crossfire FPS is low, however I cant realy find a reason why your single card performance is under.

 

[mactronix]

Also dont forget dual cards can be a pain although much better than even a couple of years ago they still have the odd issue with drivers and game support.
Also some games will play nice with an i7 and hyper threading and some wont. Try disabeling the hyperthreading and see if it makes a differance or not.

Mactronix

 

[Dougx1317]

According to this review, PCI-e 4x was only 5% slower than 16x. And this was with a 5870. My tests showed only a 15% difference. Would the 4x slow down the entire crossfire setup, or only limit the helpfulness of a second card?

I guess I won't get crossfire, but I'm still upset that my single 4890 is so much slower than it should be. The only difference in setups is that they have 2GB more of RAM, a faster harddrive, and a i7 965 (vs my 860). I'll turn off HT and see if that helps.

 

[megamanx00]

Yup, those uneven PCI-E slots are the problem. Crossfire doesn't really like slots having a different number of lanes.

 

[JofaMang]

In Crossfire, both cars will be limited by the lowest common denominator of limitations. Thus, your CF setup will run like a 4x/4x setup, as one card cannot be called upon to do more work to compensate for the other. I was never even able to get my computer to enable crossfire with my x16/x4 mobo and a borrowed 4870, but that may have been due to my OS, which was xp64 at the time.

 

[erdinger]

You tested crysis with 8x AA I dont think that a 4890 alone reaches 26 frames there!

I'm quite sure that if you test Crysis with lower AA you will come near to the results from the benchmarks of the others...


16 fps normal to 26 is over a 50% increase so the system probably works as it should.

 

[Dougx1317]

Erdinger, I got my expected FPS from the link above.

Overclocking to 4Ghz with HT off and turbo off did not change the FPS.

I just thought of a new potential problem. I'm using the demo of crysis since Warhead doesn't have a built-in benchmark. Could this be an issue? Perhaps I'll look for a benchmark hack for Warhead.

 

[JofaMang]

Erdinger, I got my expected FPS from the link above.

Overclocking to 4Ghz with HT off and turbo off did not change the FPS.

I just thought of a new potential problem. I'm using the demo of crysis since Warhead doesn't have a built-in benchmark. Could this be an issue? Perhaps I'll look for a benchmark hack for Warhead.

You won't find any more performance anywhere else. The fact is, in CF, both cards HAVE to do the same amount of work, and if one of the cards is limited in anyway, the other card will only perform to that specific limitation. If you were to underclock one card, you would find that both cards will run at the lower clocks, or if you pair two cards that are identical except one is a 1gb card, and the other is a 512mb card, they will both act like 512mb cards. You are effectively running a 4x/4x crossfire setup. I researched this topic heavily when looking at a CF setup, including borrowing a card as mentioned above. x16/x4 may have been passable in the 3xxx days, but newer cards saturate the PCI lanes to a MUCH greater degree then ever before, and x4 is highly restrictive, as compared to 8x or 16x.

 

[mactronix]

^
Its true you know AFR means it cant run any faster than the lowest denominator, this is the sort of thing that the hydra chip is meant to sort out.

Mactronix

 

[Dougx1317]

So, OCing the card in the 4x slot higher will not make a difference? (And the extra cash for an 8x/8x board may have been worth it if I wanted to crossfire?) Will the 4x/8x problem matter at all if I only have one card?

 

[paperfox]

If you only have one card in you motherboard it should run at x16 and be able to get the full performance out of your 4890. (if you only have one card in and you see that its only running at x8, that could be the source of all your low FPS problems with a single card)

Overclocking your cards on a x4 slot will not increase performance as you are already bottlenecked. Yes getting a x8/x8 motherboard for crossfire would have been worth it, but only if you accually do/will use it.

 

[Dougx1317]

How can I tell how fast my single card is going? (4x, 8x, 16x)

 

[paperfox]

Go download GPUZ, in the middle of the app window it will tell you.

 

[Dougx1317]

GPU-z says "Bus Interface: PCI-E 2.0 x16 @ x16" so I think I'm good. I'll just stick with one card, and hope that I figure out why performance is down.

 

[JofaMang]

Seriously, you are ignoring your 4x slot as the culprit? Ok then, just try not to blow too much money trying to fix it, unless you want to get a real CF mobo.

 

[daedalus685]

Seriously, you are ignoring your 4x slot as the culprit? Ok then, just try not to blow too much money trying to fix it, unless you want to get a real CF mobo.

No he isn't. He agreed that it is the issue and is sticking with one card for now, as he doesn't want to buy a new MB :|


Though the OP still feels he gets low fps with just a single card at x16.. thus more things to work on.