SLI & Crossfire in PerformanceTest V8

[David (PassMark)]

The DX10 test is different from the DX11 test. So it is expected that the frame rate will be different.

[TheUpgiven]

We've just updated the drivers (to 13.12) used for the crossfire system in this test and updated the scores, it looks like the DirectX9 Complex test should be getting much better results now with crossfire enabled.

Is it me or is the framerate better for DX11 than DX10?

Why is that?

[Tim (PassMark)]

We've just updated the drivers (to 13.12) used for the crossfire system in this test and updated the scores, it looks like the DirectX9 Complex test should be getting much better results now with crossfire enabled.

[David (PassMark)]

Your results seem to line up with our Crossfire results in PT8.

For the DirectX 9 Simple test, results were about the same with Crossfire on and off.

For the DirectX 9 Complex test, Crossfire hurts the performance a bit.

For the DirectX 10, Crossfire helps a lot.

For the DirectX 11, Crossfire also helps a lot.

For the DirectCompute test Crossfire had no significant impact.

Of course we had a different CPU, video card, etc.. so the changes aren't absolutely identical. But broadly speaking, they line up.

[hawks]

crossfire

My results with crossfire off are #28863 on are This Computer this cant be right
3D Graphics Mark
#28863 - AMD Radeon HD 6950 3249
This Computer 3286

Graphics 3D - DirectX 9 Simple
#28863 - AMD Radeon HD 6950 215.5
This Computer 195.2

Graphics 3D - DirectX 9 Complex
#28863 - AMD Radeon HD 6950 94.2
This Computer 57.2

Graphics 3D - DirectX 10
#28863 - AMD Radeon HD 6950 55.0
This Computer 91.7

Graphics 3D - DirectX 11
#28863 - AMD Radeon HD 6950 50.9
This Computer 85.2

Graphics 3D - DirectCompute
#28863 - AMD Radeon HD 6950 1229
This Computer 1196

[Michael (Passmark)]

SLI

SLI using default settings actually performs poorly compared to a single card. Setting the mode to AFR2 (Alternate Frame Rendering 2) for PerformanceTest and related executables (all the executables in the install directory) appears to provide best results.

Result from 2 x GTS450 in SLI




CROSSFIRE
Crossfire fares better under default settings, and attempting to change to custom settings seems to have little effect. Crossfire does seem to cause some additional graphical disruption in the DirectX 9 Complex test, although the current AMD drivers seem to have a few visual quirks in many of the tests regardless of whether crossfire is used or not, this should not be adversely affecting the scores though.

Results from 2 x HD 5750 Crossfired

[David (PassMark)]

All the 3D tests will 'use' SLI if it is available.

What we observed during testing however is that SLI doesn't always increase the frame rate and can also generate some visual artifacts.

Overall however you should get a higher 3DMark score with SLI or CrossFire in PerformanceTest V8. The performance gain is dramatic in some cases like the DirectX 11 test, and non existent other cases, like Direct Compute.

So this is an improvement over the SLI situation in V7 but once again we found that the documentation and developer support for SLI from nVidia varied between pretty poor and non existent. My feeling is that they put all their support effort into a small number of AAA game titles (and customize their drivers to support these games). Every other 3D application doesn't get SLI support in their drivers. Note that most of the DirectX samples from Microsoft and nVidia themselves don't work with SLI either.

We post up some more details of our SLI & Cross testing next week.

[Dbo]

is the directx11 test suppose to use SLI?