list games where crossfire / SLI do not scale

[DerekWilson]

while we are always evaluating some game or other for inclusion in our benchmarks, we can't test everything.

one of the most important things going forward with crossfire and sli becoming more important to the strategy of NVIDIA and AMD (especially with single card multigpu soluitons on AMD's end) is broad compatibility.

we've seen a trend that when we test games that don't have good multigpu support, for some reason a week or two later we'll see a hotfix driver pop up with profiles for the games we tested.

so let's see if we can start doing something really interesting -- lets find all the games that currently, with the latest 4000 series hotfix driver from amd and 177.39 (yes .39 was built after .41 and is "more recent" even though it was released a couple days earlier), that do not scale either with crossfire, with SLI or with both. then we'll test them all and keep testing them until there's a fix.

the only thing i don't want to see are games where there is a hard framecap that can't be circumvented and a single card already hits it (or comes damn close). all other games that have scaling issues i want to see posted in this thread.

we really really really appreciate the help of the community on this one.

Thanks!!

:: EDIT :: now with more glue ...

 

[MegaWorks]

This is excellent idea! :beer:

 

[MegaWorks]

Mass Effect doesn't scale at all.

Radeon HD 3870 in CrossFire using Catalyst 8.6

 

[bunnyfubbles]

Sticky?

 

[Sylvanas]

Originally posted by: MegaWorks
Mass Effect doesn't scale at all.

Radeon HD 3870 in CrossFire using Catalyst 8.6

It does for me besides that Mass Effect has a framecap in the form of 'Framerate smoothing'- nothing higher than 62fps.

 

[MegaWorks]

Originally posted by: Sylvanas

Originally posted by: MegaWorks
Mass Effect doesn't scale at all.

Radeon HD 3870 in CrossFire using Catalyst 8.6

It does for me besides that Mass Effect has a framecap in the form of 'Framerate smoothing'- nothing higher than 62fps.

That's what I get around 60fps at 1680X1050

* Noise Filter on
* Textures: Extreme high
* Filter: Anisotropic
* Everything maxed out

 

[DaveSimmons]

Great idea. The 9800 GX2 dominated the ATI 48xx review because AT was testing games with good SLU support, and it would be nice to contrast that with the games where it runs more like a single 8800GT.

(Sorry that I have no games to add to the list, but I've always had single-GPU systems.)

 

[AmberClad]

Originally posted by: Sylvanas
It does for me besides that Mass Effect has a framecap in the form of 'Framerate smoothing'- nothing higher than 62fps.

Let me guess - an Unreal 3 engine game?

 

[MegaWorks]

Ok I removed the framcap and I got Avg: 64.345

After reading guru3d CrossfireX review it does scale well. :laugh:

DerekWilson remove Mass effect from the list.

Thanks Sylvanas!

 

[nitromullet]

Nice of you guys to ask for input from the forums.

I noticed that you guys left QW:ET off the GTX 280 SLI comparison. http://www.anandtech.com/video/showdoc.aspx?i=3334&p=19 was there an issue?

A few things I'd like to mention with regards to a multi-gpu scaling review that I'd like to see regardless of which games you test:

1) min fps
2) To address the concerns over micro-stutter, maybe show the maximum interval between any two frames. Alternately, show how often the interval is larger than the average interval necessary to obtain an avg. of 30fps (33.33ms).
3) High levels of transparency AA. For ATI: 8x AAA (Super sampling); for NVIDIA 8x Super Sampling transparency AA. Obviously, this would only be for games that allow these levels of AA.

 

[aka1nas]

Originally posted by: AmberClad

Originally posted by: Sylvanas
It does for me besides that Mass Effect has a framecap in the form of 'Framerate smoothing'- nothing higher than 62fps.

Let me guess - an Unreal 3 engine game?

Yep.

The min frame rate is much better for me with SLI enabled.

 

[BFG10K]

Last time I checked CF doesn't scale properly in Jericho and won't run Stalker at all. Also it won't run AA in Bioshock when single cards will.

As for SLI, there are a few case I've seen where 9800 GX2 configurations (single or dual) fail outright to run some tests at 2560x1600 with high levels of AA (i.e. the score is zero because they crash).

 

[Sylvanas]

Originally posted by: BFG10K
Last time I checked CF doesn't scale properly in Jericho and won't run Stalker at all. Also it won't run AA in Bioshock when single cards will.

As for SLI, there are a few case I've seen where 9800 GX2 configurations (single or dual) fail outright to run some tests at 2560x1600 with high levels of AA (i.e. the score is zero because they crash).

Stalker is fine in Crossfire and scales very well, I can attest to this it has worked fine for me for a long time. As for Bioshock it is my understanding that AA is available in DX9 mode only see here

 

[BFG10K]

Originally posted by: Sylvanas

Stalker is fine in Crossfire and scales very well, I can attest to this it has worked fine for me for a long time.

Stalker failed in CF 4870 in this test:

http://www.computerbase.de/art.../14/#abschnitt_stalker

However it is possible the hot fix resolved that issue.

As for Bioshock it is my understanding that AA is available in DX9 mode only see here

Yes, but the problem is AA won't work under DX9 mode when using CF. It'll work on single ATi cards though.

Also nVidia allows driver level AA in DX10 Bioshock and also in Stalker which ATi does not.

 

[CP5670]

It's worth mentioning Far Cry with SLI, although someone should verify if this is still a problem. I found that it normally ran okay but dropped to a single card's level whenever the nightvision was turned on, something that you didn't see in any benchmarks. This was a few years ago, however, so it may well have been fixed since then.

Note that this game is still not entirely smooth on modern cards if you crank up both the resolution and AA, so it would benefit from multiple GPUs despite being four years old. On a single GTX 280, I am finding that the framerate in some scenes (a few places in the River map, for example) can drop into the high 20s at 2048x1536 with HDR and 4x AA.

 

[DerekWilson]

CP5670 -- that's interesting ... we do often only get a small slice of games in our benchmarks. sometimes when a game is initially released we'll look at a few different scenarios, but we still never benchmark an entire game ...

... so if you guys know of any sections of games that have issues with mulit-GPU -- even ones we currently test -- please let us know that as well.

thanks for the feedback thus far.

 

[chizow]

I'd be interested in seeing FRAPs dumps graphed, like HOCP does. I'd also like to see it more clearly expressed when games are frame capped or CPU limited. It would also be great if AT could cover some of the hot button topics with SLI/CF, like micro-stuttering, vsync/triple buffering being broken?, x8 vs x16 bus performance, scaling and CPU bottlenecking, etc. etc.

 

[BFG10K]

...and AFR input lag too.

 

[DerekWilson]

we're looking at tools to address those issues, but we will not be doing any instantaneous frame rate versus time graphs, nor will we be doing any tests which are not repeatable or from which reliable data cannot be obtained.

we are looking at really diving deep into micro-stutter and other multi-GPU issues when R700 is launched. i'm hoping we'll have everything we need by then to really cover these issues.

i'm not sure how we'll test input lag though ... suggestions?

 

[ViRGE]

IIRC, didn't someone make a D3D app to test your response time, in effect measuring the input lag?

 

[Jax Omen]

AT using their review muscle to try to convince driver teams to fix problematic games? I LOVE IT!

Kudos to you, Derek!



Unfortunately, I haven't hopped on the multi-GPU bandwagon (yet) so I can't help V_V

 

[Hugh H]

Just fired up World in Conflict on my new rig (see signature below), and I'm getting 1 card performance with SLI enabled. I'm a little pissed, will conduct further testing and report back. I'm using drivers 175.16.

Edit: Got it to work, now I'm benching 34 FPS average with SLI (all settings as high as they go) and 19 FPS with single card. Forced AFR2 on rendering mode.

 

[DerekWilson]

Originally posted by: Hugh H
Just fired up World in Conflict on my new rig (see signature below), and I'm getting 1 card performance with SLI enabled. I'm a little pissed, will conduct further testing and report back. I'm using drivers 175.16.

Edit: Got it to work, now I'm benching 34 FPS average with SLI (all settings as high as they go) and 19 FPS with single card. Forced AFR2 on rendering mode.

does it not work on the defaults? this is important to us as well.

 

[LittleNemoNES]

IMO, while well intentioned, this thread is going to be a mess since different drivers, beta or WHQL can make or break multi-GPU

 

[Sunrise089]

Originally posted by: gersson
IMO, while well intentioned, this thread is going to be a mess since different drivers, beta or WHQL can make or break multi-GPU

This is probably why Derek specified one specific set of drivers in the OP.