2x 5970 3dmark11 issues

I recently picked up another 5970 off of Newegg for $399! Right off the bat, I reinstalled all of my display drivers. I tested it out on Battlefield Bad Company 2 and I gained a hefty performance boost, but felt there was something wrong. When I zoom in on a scope it lags a bit and is not very fluid. It's like it takes a while to get everything into the memory of the video cards. I did not have this issue before. When I put Vsync on, the game runs perfectly, except the lag on zoom. This lag only happens once and then all is well. Also when joining a server when it is loading the map, my mouse is extremely unresponsive, the circle of bars that is indicating that the map is loading stalls for 10 seconds then chops to completion. -- This is all on the 11.10 preview drivers for BF3 - I even tried 11.9 and still the same problem.

I also did a complete driver sweeper to remove ALL AMD/ATI drivers and did a fresh install of 11.9 and the 11.8 ver.4 cap drivers too. Same problem.

Now for the real issue - 3dMark11
Stock 1 5970 and my processor at 3.8GHz I get P8136
http://3dmark.com/3dm11/849584
Overclocked 5970 @970/1200 and CPU at 4.4GHz I get P9085
http://3dmark.com/3dm11/1229168
Now - 2 5970 at stock with CPU at 3.8GHz I get P9334
http://3dmark.com/3dm11/1941827

The combined score drops to almost 50% of the original stock score. Not to mention that the entire benchmark is completely choppy regardless of the FPS I am getting. The FPS monitor on the benchmark keeps jittering 10FPS. Its like my crossfire is running AFR (Alternate Frame Rendering) and 2 of the 4 processors are slower than the other so I get 34fps for 2 frames then I get 24fps for the other two. You can see the frame rate jump back and forth. Fraps shows a constant FPS that varies expectedly. Here is a link to a YouTube video that I uploaded of the scenario. This is only the first test and this one doesn't jitter as bad as the other tests. Watch the framerate.

My system specs are as follows:
Asus P6X58 Premium
Intel Core i7-970
2x ATI Radeon HD5970
Intel X-25M 160GB SSD
6GB Corsair Dominator @1503MHz 8-8-8-20
one 5970 and the processor are watercooled. Another watercooler is on the way for the other 5970.
Thank you for your help!

Update your sig... Are you using that 750W PSU for 2 5970's?

I was in the process of updating my sig as you posted... I didn't realize that I had my old system on there. Sorry

See new sig below.

The vid looks like microstutter to me... that happens in SLi/Crossfire.. add 2 Crossfired cards and maybe it gets worse?

Although micro stutter is annoying, it should not reduce my combined score when I add 2 more GPUs to the system. Also, I mentioned that the first benchmark is not as bad as the other tests. In the 2nd benchmark the frames vary from 23-33 - you can see the 2 turn into a 3 but the second digit doesn't change much. That is a little extreme for microstutter, I would think.

To push those cards properly, you also need more CPU...3d11 loves CPU!

I dont know, something is amiss, thats for sure. How many monitors is are those beasts powering?

Let me go clock my CPU to 4.4 and try again - see if it makes a difference.

One monitor - 1920x1080.

Ok at 4.28GHz I gained a little bit in combined score, but nothing compared to my combined score with only 1 card installed.
http://3dmark.com/3dm11/1943136

I am running 6 cores 12 thread at 4.3GHz, I would think that is enough to power these 2 cards.

Well in games, you are wholly bottlenecked by that CPU at that res. Thats massive MASSIVE overkill (and maybe a hint of jealousy) for that setup. One 6950/570 will play anything with AA at highest settings, none the less 2 of those dual GPU monsters.

Im thinking you are just experiecing a bottleneck and significant microstutter.

The reason I got the second one is for battlefield 3. I am currently playing the beta and with one card I was only getting 50-70 fps on max settings and could not use vsync. When I play games, I enjoy smooth play with great visuals. With the second card I ran BF3 at ultra with 110-150 fps with my hardware at stock. But that is only on the Operation Metro map - bigger maps will require more power. I know that I am fine for games and will not have an issue at max settings at this point - but I was only concerned that there was an issue. I looked at other 3dMark11 scores with the same setup (or with the i7-980x) and at the same frequencies they are getting 9700 combined score on Performance setting - I am only getting 4300.... Same cards, Same CPU except mine isn't "unlocked", and same clock speeds.

BF3 plays very nice with vsync on, but has the same issue that Bad Company 2 has - randomly spikes to below 20 fps for a few seconds - but not very often.

Found the reason why the combined score is low. Seems to be driver issues with nVidia AND AMD. Oh well, I guess - as long as my games play smooth I am fine with the minimal issues.

Thanks for your help.

It is a driver issue.

FM is aware that the combined test and configurations with more than 2 GPUs in SLI/Crossfire does not always work as it should, but at this time it appears to be a problem with the video drivers.

The problem is...

Combined test spawns extra copies of the physics simulation for each GPU in the system. The theory is that for optimized performance, the work has to be duplicated to avoid GPUs having to share physics calculation data between cards over the slow PCIE bus. It is expected that any future game that choses to do complex physics using DirectCompute has to do this for proper multi-GPU scaling.

However, the drivers seem to be failing with certain SLI/Crossfire configurations, ending up running multiple copies of the physics calculations on the wrong GPU(s), affecting the performance. This varies from driver version to driver version, which leads us to believe that this is a driver issue. We have notified the GPU vendors but have no final word on the issue at this time.

SI disabled-scenario (or manually forcing GPU count to 1) runs only one copy of the DirectCompute physics simulation. This does not reflect a "real world" scenario where the simulation would have to be done simultaneously on each GPU to keep the performance optimal. Such a score is not an official valid score (for competition / high score purposes).

Click to expand...

Glad u found the issue! And I learned something else today!

Just to give you another look I'am running two 5970's 3DMARK11 score P9385 runs very smooth about 78% max on benchmark. i7 975 ex -1.1Kw
Heaven 2.5 is 2873 at 114 FPS avg. Over all I'am pretty happy.