My HD7970 Crossfire setup is not working properly. HELP!

[vipernig]

Hi all.

It has been two days that I have been trying to crossfire my 2 7970 and it just doesn't work. Now I am at a point where I have completely uninstalled all AMD stuff, ran Ccleaner and Driver Seeeper. I then reinstalled Catalyst 12.4 and enabled crossfire. When I run in Skyrim, Alan Wake and Hard Reset I get half the FPS or WORSE than on each 7970 alone. I check FPS with FRAPS. I had GPUZ running ok with CAT 12.3 but now it craches my pc when I try to read info on second video card. I disabled ulps. I had 2 entries, set them to 0. For Skyrim on single card i get 45fps with eveything set to max. Alan Wake is the same thing. Also the same with Hard Reset. I read a Guru3d 7970 crossfire review and they show benchmarks that Skyrim and Hard Reset scale at almost 100%. All my tests are at 2560x1600.

I tried with one and 2 crossfire bridges connected and I get the same results. And I have the 8pin and 6 pin connectors on each GPU connected to the PSU. And I tried both PCIE Power Supply locations on my PSU when I did the single card tests so I am pretty confident that my PSU is running ok and should be able to drive 2 GPUs.

I made a little discovery later. I was reading on other crossfire problem threads and I did not try to to test each card on the second pci slot. So for the Asus P8Z68-V Pro GEN3, the top PCIe in blue is rated PCIe 3.0/2.0 x16 for single or or dual at x8/x8. The second gray PCIe is rated PCIe3.0/2.0 at x8. Now having a i7-2600k I can only use PCIe 2.0. So I procedeed to test both cards on the gray PCIe X8 mode and to my surprise they performed SIGNIFICANTLY better. In the scenes where I was averaging 45fps in Skyrim I now average 55fps. This is a MASSIVE difference. I went to benchmark in Hard Reset and I get an Average of 100fps. And Alanwake is also on average 10fps more. This is a HUGE difference and this leads me to believe that there is an issue with the PCIe interface.

Does this mean I have a faulty motherboard? I checked my asus drivers and they are up to date. What do you guys think? I am off to bed. Atleast now I am performing like those benchmarks I read for a single HD7970.

As for micro stuttering this is not what I am experiencing. I read up on it and looked at benchmarks and the worst FPS was that of a a single card and the best fps was double what a single card was delivering and then it fluctuated between both values which is not what I experience when I crossfire. I get a constant worse FPS than a single card. And I would say with the new FPS that I gathering from running on the second PCIe slot, they are literary half or WORSE the FPS that I get from my single card.

So back to the question. Faulty MB or PSU(doutfull it is a XFX PRO 1000W Platinum, which is the same as a the Seasonic 1000W Platinum. Or is there something I am missing?


Here are my specs:
CPU: I7 2600K @ Stock 3.4 Ghz (for now)
MB: ASUS P9Z68-V Pro Gen 3
GPU: 2 x Saphhire Refence Radeon HD 7970 3TB
RAM: 16GB 4x4GB Gskill PC-1600MHz
PSU: XFX Pro Series 1000W Platinum

[dandelioneater]

Do you have any games that you can test that are proven xfire friendly? BF3 maybe? I don't run multiple cards so I don't really know which games.

What about running a synthetic benchmark? 3dmark, unigen heaven just to see some fps numbers with xfire on and off? If you get almost double the fps with xfire enabled, then I think that would prove that your hardware is ok and it may be catalyst drivers.

7970 xfire drivers may still be immature or buggy.

[Adzsask]

I am guessing you can only CF X8/X*, or ya you have a faulty motherboard.

[vipernig]

It tried 3dMark11 and honestly I have 0 experience with it and it did kill my PC in crossfire and have erros with single card on the top PCI. I will now try unigen on both PCI slots and then crossfire. Again I keep reading about people having Alan Wake working in crossfire from day one. I know that Skyrim can be problematic that is why i test with Alan Wake and Hard Reset. Although for Hard Reset on the second PCI slot at all mask FXAA and vsync off on my Dell U3011 (2560x1600) the game never drops below 60fps and averages 100fps. So crossfiring is useless for it. But yes. Off to do unigen heaven test today.

[vipernig]

Yes when I crossfire they become PCIe2.0 x8/x8 but again I am getting way better performance on my second slot 2.0 x8 than on the first slot 2.0 x16. So PCI speeds should not be an issue. And a new developement now, GPUz crashes my PC and I have to do a hard reset when I try reading the second GPU when crossfire is enabled.

[vipernig]

Ok I just finished running Unigine Heaven Benchmarks. I tried 3dmark11 but that killed the PC in Crossfire and had errors in single card mode. Can't be sure because I don't know how to use it. Unigine was great and simple. I did multiple test and will detail those thourouly later.

My setting:
Binary:Windows 32bit Visual C++ 1600 Release Mar 7 2012
Operating system:Windows 7 (build 7601, Service Pack 1) 64bit
CPU model:Intel(R) Core(TM) i7-2600K CPU @ 3.40GHz
CPU flags: 3400MHz MMX SSE SSE2 SSE3 SSSE3 SSE41 SSE42 HTT
GPU model:AMD Radeon HD 7900 Series 8.961.0.0 3072Mb

Settings:
Render: direct3d11
Mode:2560x1600 8xAA fullscreen
Shaders:high
Textures:high
Filter:trilinear
Anisotropy:16x
Occlusion: enabled
Refraction:enabled
Volumetric:enabled
Tessellation:extreme

For the PCIe1 Slot that can run at x16 I get:
FPS:28.1
Scores:707
Min FPS:12.9
Max FPS:60.6

For the PCIe2 Slot that can runs at x8 I get:
FPS:28.6
Scores:722
Min FPS:17.0
Max FPS:62.2

For Crossfire atx8/x8 i get:
FPS: 51.4
Scores:1295
Min FPS:20.2
Max FPS:120.6

Now the first time I ran the crossfire test it was crap and half the performance of a single crad. I enabled Overdrive and left everything stock. This was what yielded the above crossfire result. When I disabled the overdrivre in CCC and then rebooted the PC and reran the test I got the same results above for the crossfire.

But Alan wake still runs like crap even with overdrive enabled and Skyrim killed my PC. So this leads me to believe that my PSU is fine. Now it is either a faulty PCIe hardware because the PCI1 rand a bit worse than the PCI2 in the benchamark and as stated above in games there is a significant difference. Or it could be fubared PCIe drivers that are not properly power managing the MB. The final real test is a fresh install but I so don't want to do that. And considering that Unigine worked in crossfire and scaled properly there has to be a way to solve this.

[NyteOwl]

Have you tried updating the motheboard BIOS?

[SKYMTL]

FYI, from my testing, Skyrim still isn't properly supported by Crossfire.

Can you give us your 3DMark11 score please?

[croy_00]

Originally Posted by SKYMTL

FYI, from my testing, Skyrim still isn't properly supported by Crossfire.

Can you give us your 3DMark11 score please?

This.

I have tested Xfire 6970's, SLI'd 560ti's, and SLI'd 570's in Skyrim. The 6970's were absolutely horrible on any settings higher then high (no-ultra), where as the Nvidia cards rocked it without breaking a sweat.

[vipernig]

OK People here is an update.

I used my older ssds that had a relatively fresh windows with ccc11.13 with the drivers from 12.1 only. So catalyst version still showed 11.13. I reran all the benchs on it and 3DMark11 did work on it so did GPUz. My 3DMARK scores on single card in each PCI slot was about 8000 and then 11700 on crossfire. Yeah not great. I reran Unigen and both PCI slots were very closely matched and crossfire scaled ok. The scores in Unigen were a bit better than on the main install with CAT12.4. And Skyrim was running okay with microstuterring from 60fps to 130 fps which is the best result I had so far. So this confirms that my PCI slots hardware wise are fine. Now I udated the drivers on that OLD intall to CAT12.4 by doing proper uninstall, driver sweeper, install driver. And the PCI slots were benching the same in Unigen. So this confirms a driver issue and or registry pollution in my main setup. But 3DMark11 was crashing the PC in the same manner as the main setup and so did GPUZ. Now i downgraded properly to CAT12.3 and 3DMark and GPUz all worked fine. My score now pushed up to 13500 can't remember exactlybut the graphics score was 14900. And again games ran like crap.

So I confident to say that it is not a MB bios issue and more likely issues with CAT12.4. And on the main install I have to admit that I installed AIsuite from ASUS, which is PURE CRAP btw, and it did not unistall properly and I had to use force uninstall with Revo Uninstaller and that just my explain my PCI discreptencies on the main install. Again I got the second 7970 for future roofing my system but it doesn't seem that will be the case because I have been reading now that 6970 still don't crossfire properly and this is really unacceptible. Although I have to admit that with the main install CAT12.4 on on 7970 I get awesome results. I will do furher research but I really feel duped by AMD and ASUS and all these reviews online that say my setup should work. So Yes as always it is the drivers that are the culprit and shame on you ATI and at the team in Toronto. What I did in 3 days should be done at ATI. As a firmware/electronics Engineer myself I never release anything before doing thorough and proper testing. I have a feeling the barely do real world application testing at ATI. And common in this day age having to do a big dance just to cleanly uninstall something is just unacceptable.

So I will go back on my main install and try to fix my driver issues and getting back the first PCI slot to cooperate. And I will see if I will foreward my findings to AMD or not. Again your consumer should never be your unknowngly your Beta testers. I think these cards might both go on Kijiji and I will keep reading on the 680 SLI and see what the feedback was on those. But hey feedback on 7970 seemed good too.