Batman AC negative CFX scaling with hybrid Physx

[automaticman]

So I've been fooling around with my new build and Playing Arkham City. I'm using a pair of 6950 1gb in CFX and just swapped out a gt430 with a friend's 9800gt dedicated to hybrid PhysX.

At 1920x1200, DX11, Physx high, normal tesselation, fxaa high and all other settings at their highest I'm seeing around 50fps avg on the benchmark with crossfire disabled, but only 38fps with crossfire enabled. Max is 72 either way.

I have a second monitor showing the desktop with various monitoring tools, so I can see that I am getting usage on both GPUs (although only about 40%), as well as usage on the 9800gt for physx. Without crossfire, the single 6950 shows around 90% usage.

Unfortunately I can't compare to what I was getting using the gt430 for physx because I didn't realize hybrid physx wasn't working (turns out the newest physx sw 1107 doesn't work with 1.05ff mod).

I was running physx high on my 2500k without realizing it and only really losing a few fps, which odd enough, is pretty inline with other reviews.

So anyways, anyone else have a CFX with hybrid physx setup and seeing the same negative scaling?

[automaticman]

OK, so I ran a slew of benchmarks this morning and came up some interesting numbers. Also, for the record, I do have the dx11 patch installed for the game.

For these benchmarks the only thing I'm changing is DX11 on/off (with dx 11 on I have features set to MVSS and HBAO, and tessellation on normal), crossfire on/off (via modifying the game profile in 12.1), and physx off/high.

Resolution is 1920x1200, FXAA at high, everyhting else turned on or very high for detail.

So first round - Physx OFF, DX9

crossfire off
min 62
max 127
avg 103

Crossfire ON
min 71
max 170
avg 137

Decent scaling here, no stuttering

next round - Physx OFF, DX11

Crossfire off
min 20
max 83
avg 68

Crossfire ON
min 26
max 108
avg 70

Really pretty crap scaling here, seeing a little bit of the stuttering that the patch was supposed to fix. DX11 really kills any kind of crossfire scaling it seems.

No let's turn on Physx - Physx HIGH, DX9

Crossfire off
min 20
max 99
avg 53

Crossfire ON
min 20
max 96
avg 54

Again, no scaling. I guess I'm just maxing out the 9800gt here. I will say that with crossfire on the benchmark felt smoother and the min was at 30 up until right when the ice shard phase started (fps came right back up but there was a dip right at the beginning). With one card the min was down at 20 pretty much from the get-go. I think I'll be playing the game with these settings.

Last we have Physx HIGH, DX11

Crossfire off
min 18
max 72
avg 49

Crossfire ON
min 13
max 96
avg 41

Here we see the negative scaling, and an annoying amount of stuttering. And really, the dx11 effects just arent that noticeable. Yeah the tessellation is ok, but I'd rather have physx on.


Ideally I'd re-install the gt430 for physx and see how it compares to the 9800gt, but I already installed it into my wife's computer and had to fight for about 15-20 minutes to get the HDMI cable plugged in. I may still try it, but not today.


Really the most telling number for me was the dx11 crossfire scaling without physx. I mean with physx on, it's not hard to just imagine that the 9800gt is bottlenecking the fps, it can only go as fast as that card can process the physx. But with the physx turned off and to still see almost no scaling with dx11 was pretty disappointing.

I don't know if the issue is with the AMD driver or the games coding (not surprising given the issues this game has had).