Crysis (Original) - Poor DX10 Radeon 6950 Crossfire Scaling

[DrBoss]

I am in the process of getting a new rig up and running. The system specs and build history can be read in this thread which i posted in the CPU/Overclocking Forum.

As the title suggest, my two MSI R6950 Twin Frozr III's are not scaling as well as many benchmarks/reviews suggest they should. Crysis is patched to Version 1.2.1 and i am running Window 7 x64

I only recently got the system up and running, but my initial benchmarking results are as follows:

I benchmarked using CrysisBenchmarkTool v1.05
AMD/ATI Catalyst 11.3
3 runs on the ASSAULT_HARBOR level
64Bit, DX10, all settings @ VERY HIGH, 1920x1200, 8xAA, VSYNC enabled

With Crossfire DISABLED (only 1 GPU)
Pass 1: 25.27 MIN - 45.47 MAX - 34.57 AVG
Pass 2: 25.27 MIN - 46.23 MAX - 35.95 AVG
Pass 3: 25.27 MIN - 46.23 MAX - 35.69 AVG

GPU1 temp at end of test, 54C

With Crossfire ENABLED
Pass 1: 23.33 MIN - 59.92 MAX - 44.95 AVG
Pass 2: 23.33 MIN - 59.92 MAX - 40.18 AVG
Pass 3: 23.33 MIN - 60.97 MAX - 45.36 AVG

GPU1 temp at end of test, 52C
GPU2 temp at end of test, 47C

As you can see, the minimum framerate is actually lower in crossfire, and my average framerate has only improved around 30%. I've seen reviews where Crossfire 6950's achieve 80-90% scaling in Crysis.

So here i am scratching my head. I've read about the r_multigpu command... but haven't attempted to change this from whatever the default is.

I am also a bit unclear on how to use CCC application profiles. I installed the driver for profiles but don't understand how to use it, or if i need to use it on Crysis.

Any insight or help is greatly appreciated
Cheers

 

[Qbah]

I am in the process of getting a new rig up and running. The system specs and build history can be read in this thread which i posted in the CPU/Overclocking Forum.

As the title suggest, my two MSI R6950 Twin Frozr III's are not scaling as well as many benchmarks/reviews suggest they should. Crysis is patched to Version 1.2.1 and i am running Window 7 x64

I only recently got the system up and running, but my initial benchmarking results are as follows:

I benchmarked using CrysisBenchmarkTool v1.05
AMD/ATI Catalyst 11.3
3 runs on the ASSAULT_HARBOR level
64Bit, DX10, all settings @ VERY HIGH, 1920x1200, 8xAA, VSYNC enabled

With Crossfire DISABLED (only 1 GPU)
Pass 1: 25.27 MIN - 45.47 MAX - 34.57 AVG
Pass 2: 25.27 MIN - 46.23 MAX - 35.95 AVG
Pass 3: 25.27 MIN - 46.23 MAX - 35.69 AVG

GPU1 temp at end of test, 54C

With Crossfire ENABLED
Pass 1: 23.33 MIN - 59.92 MAX - 44.95 AVG
Pass 2: 23.33 MIN - 59.92 MAX - 40.18 AVG
Pass 3: 23.33 MIN - 60.97 MAX - 45.36 AVG

GPU1 temp at end of test, 52C
GPU2 temp at end of test, 47C

As you can see, the minimum framerate is actually lower in crossfire, and my average framerate has only improved around 30%. I've seen reviews where Crossfire 6950's achieve 80-90% scaling in Crysis.

So here i am scratching my head. I've read about the r_multigpu command... but haven't attempted to change this from whatever the default is.

I am also a bit unclear on how to use CCC application profiles. I installed the driver for profiles but don't understand how to use it, or if i need to use it on Crysis.

Any insight or help is greatly appreciated
Cheers

You have v-sync on, which means the cards are capped at your monitor's refresh rate -> 60Hz from the results. Turn off v-sync and run the tests again.

 

[DrBoss]

You have v-sync on, which means the cards are capped at your monitor's refresh rate -> 60Hz from the results. Turn off v-sync and run the tests again.

makes sense, i will check this when i get home from work...

though; i ran a number of benchmarks last night (some with Vsync off) and while i don't have the numbers in front of me, i was not getting 80% scaling. Performance was actually all over the place during each run. It was weird... at times i would hit 100fps (with average fps in the mid 70's), but then at other times the performance would seem to bog down and fps would drop to around 25 for a sustained amount of time (perhaps 3 or 4 seconds). These drops in performance happened 3 or 4 times during each benchmark and severely impacted the average fps.

 

[Keysplayr]

You have v-sync on, which means the cards are capped at your monitor's refresh rate -> 60Hz from the results. Turn off v-sync and run the tests again.

This will affect his avg and max numbers, but shouldn't budge the minimums.

 

[Kenmitch]

Maybe it's a directX problem? Not 100% sure but it seems win7 doesn't load all of the directX stuff to me at least. If I remember correctly I think it was OCCT that pointed this out to me as some of the features won't work unless you load the latest build of directX. Might be worth loading the latest build to see if it does or doesn't help.

Just for the record where are you comparing your results to? Using same map and settings?

 

[penske2298977]

I had this same problem --- I have a couple of 5970's in x-fire and the scaling absolutely sucked and never came close to what I saw in review sites. Tried everything, even bought a new drive just to install a fresh copy of vista 64 bit and nothign else other than crysis and essential drivers etc...frustrating. Hope you have better luck.

 

[DrBoss]

Is there a gpu monitoring program i can run in the background (during fullscreen benchmarking) that will record gpu stress so that i can ensure both cards are running at peak efficiency? I think that the second card might now be running at full clocks.

 

[notty22]

You can configure msi afterburner to record all that and more.

 

[cusideabelincoln]

afterburner

notty is it REALLY necessary to insert a full 1080p image? alt+printscreen or just learn to crop.

 

[DrBoss]

You can configure msi afterburner to record all that and more.

Thanks

 

[rangda]

To give you a point of reference:

I just ran Crysis yesterday with the same video cards. Played a bit at the beach in the beginning where the sun rises and you disable the jammer. High settings 4xaa gave me 60-70fps @2560x1600, Very High 4xaa at same rez was 30-35 (usually closer to 30).

I didn't bench a single card as I'm having serious heat problems with the frozr's (video card slots are 2 slots apart so no spacing between the cards unlike your mobo) so I've been watching temps more than FPS.

 

[DrBoss]

To give you a point of reference:

I just ran Crysis yesterday with the same video cards. Played a bit at the beach in the beginning where the sun rises and you disable the jammer. High settings 4xaa gave me 60-70fps @2560x1600, Very High 4xaa at same rez was 30-35 (usually closer to 30).

I didn't bench a single card as I'm having serious heat problems with the frozr's (video card slots are 2 slots apart so no spacing between the cards unlike your mobo) so I've been watching temps more than FPS.

Thanks for the reference.

When i get home i am going to try a few things in hopes of improving crossfire scaling.
1) ensure multigpu is enabled in crossfire 'r_multigpu 1'
2) install latest DirectX content
3) ensure that everything in the Catalyst Control Center is set to "let application decide"
4) compare benchmarks with Catalyst AI on/off
5) ensure V sync is turned off to allow fps over 60 (my refresh rate)
6) install MSI Afterburner to ensure that both my cards are being fully utilized
7) potentially uninstall and reinstall various Catalyst drivers to gauge performance

Is there a recommended driver cleaning application for Windows 7 to ensure video drivers are completely removed?

 

[penske2298977]

Is there a gpu monitoring program i can run in the background (during fullscreen benchmarking) that will record gpu stress so that i can ensure both cards are running at peak efficiency? I think that the second card might now be running at full clocks.

yes GPUZ


http://www.techpowerup.com/gpuz/

 

[Kenmitch]

Is there a recommended driver cleaning application for Windows 7 to ensure video drivers are completely removed?

I use the one found on http://downloads.guru3d.com/ not sure what everybody else uses.

 

[DrBoss]

Appears the issue is DirectX 10 considering the substantial improvement i see running in DX9.

DX9 Results
Min - 46.46
Max - 100.77
Avg - 76.93

 

[DrBoss]

I spent the night troubleshooting Crysis and Crossfire. Turns out the problem is quite simple. Crossfire and DX10 do not play nice with Crysis in a 64bit environment.

BENCHMARKS
crysis | 1920x1200 | 8xAA | vsync off | all setting 'very high'

DX10 64bit
min:23.22 - max:75.74 - avg:52.15
DX10 32bit
min:49.73 - max:79.69 - avg:68.66
DX9 64bit
min:46.27 - max:102.43 - avg:76.67
DX9 32bit
min:47.07 - max:107.93 - avg:79.16

DX9 isn't as effected by running the 64bit version of the game, but DX10 obviously takes a big hit. In the 32bit environment, i am pretty damn happy with 68.66 average frames per second with all settings maxed. For reference, my DX10 32bit average frame rate using only 1 of the GPU's was 35.49fps. So in 32bit, Crossfire is scaling around 193% :]

The problem appears to be Crossfire GPU utilization in 64bit DX10. During that test, GPU 1 Load = 98%, and GPU 2 Load = 72%. Running in 32bit DX10, both GPU Loads were measured at 100%.

Also of note, the highest measured GPU temp during my benchmarks was 68.0C, but i assume that will rise with extended play.


Bedtime. Hope this information helps others.
Does it run Crysis? Yes

 

[badb0y]

Nice

 

[Qbah]

Thanks for the extra effort on pinpointing the reason for the low min FPS and poor scaling I guess reviewers are running the 32bit executable in their tests then?

 

[DrBoss]

Thanks for the extra effort on pinpointing the reason for the low min FPS and poor scaling I guess reviewers are running the 32bit executable in their tests then?

i can only assume

 

[DrBoss]

There is some additional Crysis CPU scaling information i've posted in my overclocking thread.

All future performance information will be posted in there.

 

[rangda]

I spent the night troubleshooting Crysis and Crossfire. Turns out the problem is quite simple. Crossfire and DX10 do not play nice with Crysis in a 64bit environment.

DrBoss, that's a good catch. When I ran it I was running in 32 bit as afterburner only displays the overlay (which I was using to monitor temps in-game) in 32 bit.

 

[DrBoss]

DrBoss, that's a good catch. When I ran it I was running in 32 bit as afterburner only displays the overlay (which I was using to monitor temps in-game) in 32 bit.

My next step is to install the CCC extreme quality configs and bench Crysis in DX9. With the right tweaks i believe DX9 can be made to look better than DX10 (with similar, or better performance).

 

[12andy]

IIRC, object blur is only applicable for DX10.

With that said, a tweaked-out DX9 is a sight to behold, still.

 

[at80eighty]

I spent the night troubleshooting Crysis and Crossfire. Turns out the problem is quite simple. Crossfire and DX10 do not play nice with Crysis in a 64bit environment.

[..]

Bedtime. Hope this information helps others.
Does it run Crysis? Yes

Interesting. Ive become utterly frustrated with Crysis crashing on me (DX10 / W7 64) . When it runs, its absolutely gorgeous and smooth. but it crashes regularly on saves and on random points as well - just not sure what to do anymore. preliminary 32bit play crashed after a bit as well - let me give it a few more tries and see