FiringSquad test Mirror's Edge Physx performance

[AzN]

http://www.firingsquad.com/har..._performance/page3.asp

Pretty interesting results. GTX260 barely edging out 9800gtx+ with PhysX with AA.

 

[Lonyo]

What would have been better would be to see numbers run with it off, on on the GPU and on but with either a PhysX card or a second graphics card in there doing the PhysX work instead of the main card, to see if the extra PhysX itself has any impact on performance (like somehow more CPU load even though it really shouldn't), or an impact on performance due to extra rendering of the on screen effects.

 

[error8]

Originally posted by: Azn
GTX260 barely edging out 9800gtx+ with PhysX with AA.

Why does that happen? It makes no sense. Maybe it's a driver bug or something?

 

[cusideabelincoln]

At 19x12, the performance drop for the GTX260 is 40%. For the 9800GTX, the performance drop is 32%. Maybe Pysx favors the much higher clock speeds of the 9800GTX+. And the 9800GT's performance drops 35% at 19x12, so its performance drop falls in between the 9800 and 260 much like its shader clock speed does. This is probably just a crazy theory, though, and the GTX260's performance is probably the result of (the lack of) driver optimizations.

 

[AzN]

Originally posted by: error8

Originally posted by: Azn
GTX260 barely edging out 9800gtx+ with PhysX with AA.

Why does that happen? It makes no sense. Maybe it's a driver bug or something?

I don't know either. They used a GTX 260 216 core and should be a bit faster than 9800gtx+. If it was regular GTX 260 it makes sense because shader performance isn't any faster than 9800gtx+.

 

[LOUISSSSS]

"As you can see, turning on GPU-based PhysX had a profound impact on performance. The GeForce GTX 260 saw its frame rate drop by nearly half at 1920x1200, falling 40%"

wow a 40% FPS drop? not me not now.
how can anyone truly say the physx is a factor in the mainstream video card market (<$250)?

 

[yh125d]

Originally posted by: LOUISSSSS
"As you can see, turning on GPU-based PhysX had a profound impact on performance. The GeForce GTX 260 saw its frame rate drop by nearly half at 1920x1200, falling 40%"

wow a 40% FPS drop? not me not now.
how can anyone truly say the physx is a factor in the mainstream video card market (<$250)?

because you can get a 9800GTX and a 9600GT for Physx for that money

 

[SlowSpyder]

Definitely a performance hit, but it still looks perfectly playable considering the lowest end card (9800GT) at 1920x1200 4xAA averages 30FPS and only dips to 26FPS at it's low point. Hardly silky smooth, but you could certainly play it.

Now, I have not played this game (with or without Physx) but from the videos I saw it looks like Physx just adds more stuff on screen at once... I don't know if the performance hit is worth more 'stuff' on screen or not.

 

[nRollo]

Originally posted by: LOUISSSSS
"As you can see, turning on GPU-based PhysX had a profound impact on performance. The GeForce GTX 260 saw its frame rate drop by nearly half at 1920x1200, falling 40%"

wow a 40% FPS drop? not me not now.
how can anyone truly say the physx is a factor in the mainstream video card market (<$250)?

Perhaps because the all important minimum fps was still at 38fps for the <$250 GTX260 C216, more than playable.

 

[chizow]

Looks like they finally found a use for all that wasted SP performance. If you run a lower end card and want to run PhysX, you'll need to reduce resolution or turn off AA. This would be similar to any trade-off depending on the title, regardless of PhysX.

 

[Keysplayr]

Originally posted by: LOUISSSSS
"As you can see, turning on GPU-based PhysX had a profound impact on performance. The GeForce GTX 260 saw its frame rate drop by nearly half at 1920x1200, falling 40%"

wow a 40% FPS drop? not me not now.
how can anyone truly say the physx is a factor in the mainstream video card market (<$250)?

Yeah, I was going to say minimum framerate is 38 at 19x12 4xAA for the GTX260, and 32fps at 19x12 4xAA for the 9800GTX+. That's pretty much what should be focused on.

 

[AzN]

It's still only 6fps difference. That's 1920x1200 with AA too. I would have though with AA GTX 260 should have walked all over the 9800gtx.

 

[toyota]

what actually matters on the video card for processing physx? is the shaders mainly?

 

[chizow]

Originally posted by: Azn
It's still only 6fps difference. That's 1920x1200 with AA too. I would have though with AA GTX 260 should have walked all over the 9800gtx.

It does walk all over the 9800GTX when SP aren't the limiting factor, as is the case with PhysX. Just look at the % drop with PhysX on vs. PhysX off for both cards. Again, they've found a use for all those wasted SP cycles to the point SP performance is the dominant bottleneck with PhysX.

 

[dflynchimp]

what I'd be interested in knowing is if it's physically and software-wise possible to use an Nvidia card for the physics paired with an ATI card for the graphics.

 

[chizow]

Originally posted by: toyota
what actually matters on the video card for processing physx? is the shaders mainly?

Yep, all the math calculations are done on the shaders, which again, is one of the main strengths of the G92 and its high SP clocks where overall single-precision FLOP performance actually approaches and sometimes eclipses the GTX 260.

 

[chizow]

Originally posted by: dflynchimp
what I'd be interested in knowing is if it's physically and software-wise possible to use an Nvidia card for the physics paired with an ATI card for the graphics.

Supposedly its possible with XP, since you can run two different video drivers. I think there were some early results confirming this was possible. Multiple video drivers are not supported in Vista under WDDM1.0. Rumor has it that it is supported under WDDM1.1 in Win7, although I haven't seen positive confirmation yet.

 

[nRollo]

Originally posted by: LOUISSSSS
"As you can see, turning on GPU-based PhysX had a profound impact on performance. The GeForce GTX 260 saw its frame rate drop by nearly half at 1920x1200, falling 40%"

wow a 40% FPS drop? not me not now.
how can anyone truly say the physx is a factor in the mainstream video card market (<$250)?

Advances in IQ have always resulted in large performance hits. For example, check out the drop from 49fps average at Oblivion at 16X12 HDR with a X1900XTX all the way down to 35fps just by adding 2XAA 8XAF.

That's a 29% drop in performance with only 2XAA used, and back then a lot of people thought HDR+AA was the only feature it mattered to have.

 

[AzN]

Originally posted by: chizow

Originally posted by: Azn
It's still only 6fps difference. That's 1920x1200 with AA too. I would have though with AA GTX 260 should have walked all over the 9800gtx.

It does walk all over the 9800GTX when SP aren't the limiting factor, as is the case with PhysX. Just look at the % drop with PhysX on vs. PhysX off for both cards. Again, they've found a use for all those wasted SP cycles to the point SP performance is the dominant bottleneck with PhysX.

Yeah I know without PhysX it does quite well @ 1920x1200 4xAA. But when you consider all the bandwidth advantage it has it would have beaten more with PhysX at least with AA. It seems to take the bottleneck away from bandwidth to SP performance. Firingsquad also tested GTX 260 216 core too. The difference should have been more pronounced.

 

[alcoholbob]

So what the hell is going on with the SLI numbers?

The SLI numbers are the same as GPU + physX add-on GPU, I wonder if SLI was even enabled?

To clarify:

SLI 260 - physX evenly distributed (SLI enabled + physX enabled in nvidia control panel)

or

SLI 260 - single card + 260 for PhysX (SLI disabled + physX enabled in nvidia control panel)

If it's the former, then why the hell are the numbers so close to single card + add-on card? It seems like it has to be the latter, or nvidia drivers default to PhysX over SLI.

 

[chizow]

Originally posted by: Azn
Yeah I know without PhysX it does quite well @ 1920x1200 4xAA. But when you consider all the bandwidth advantage it has it would have beaten more with PhysX at least with AA. It seems to take the bottleneck away from bandwidth to SP performance. Firingsquad also tested GTX 260 216 core too. The difference should have been more pronounced.

Bandwidth doesn't matter when the limiting factor is SP, and at 1920 4xAA + PhysX you can see the difference between 9800GTX+ and GTX 260 is ~13% which is the kind of scaling you'd expect given the difference in SP performance is also around ~14% (235 MFlops to 265 MFlops).

You can verify by using PhysX off and PhysX on as min and max references, then comparing to the PPU or SLI results once PhysX is offloaded from the rendering card's SPs. You'll see the GTX 260 again re-establishes dominance over the 9800GTX+ once it no longer has to handle PhysX calculations on its own SPs.

 

[Elfear]

Originally posted by: chizow

Originally posted by: dflynchimp
what I'd be interested in knowing is if it's physically and software-wise possible to use an Nvidia card for the physics paired with an ATI card for the graphics.

Supposedly its possible with XP, since you can run two different video drivers. I think there were some early results confirming this was possible. Multiple video drivers are not supported in Vista under WDDM1.0. Rumor has it that it is supported under WDDM1.1 in Win7, although I haven't seen positive confirmation yet.

If more games come out with some cool Physx effects (Mirror's Edge doesn't seem to revolutionary), I might have to give that a try.

 

[zagood]

this info is soooooo last week...

http://forums.anandtech.com/me...erthread=y&STARTPAGE=7

 

[AzN]

Originally posted by: chizow

Originally posted by: Azn
Yeah I know without PhysX it does quite well @ 1920x1200 4xAA. But when you consider all the bandwidth advantage it has it would have beaten more with PhysX at least with AA. It seems to take the bottleneck away from bandwidth to SP performance. Firingsquad also tested GTX 260 216 core too. The difference should have been more pronounced.

Bandwidth doesn't matter when the limiting factor is SP, and at 1920 4xAA + PhysX you can see the difference between 9800GTX+ and GTX 260 is ~13% which is the kind of scaling you'd expect given the difference in SP performance is also around ~14% (235 MFlops to 265 MFlops).

You can verify by using PhysX off and PhysX on as min and max references, then comparing to the PPU or SLI results once PhysX is offloaded from the rendering card's SPs. You'll see the GTX 260 again re-establishes dominance over the 9800GTX+ once it no longer has to handle PhysX calculations on its own SPs.

Well Firingsquad has those results.

PhysX off
1600x1200 4xAA
GTX 260 216
93.2fps

9800gtx+
76.3fps

GTX 260 216 is 22% faster with no PhysX


PhysX GPU
1600x1200 4xAA
GTX 260 216
52.8fps

9800gtx+
51.4fps

2.7% faster


PPU PhysX
1600x1200 4xAA
GTX 260 216
76.1fps

9800gtx+
67.8fps

GTX 260 216 is 12% faster

It doesn't quite add up @ 1600x1200. @ 1920x1200 the difference is 2% with no PhysX and PPU physx.

 

[chizow]

Originally posted by: Azn
It doesn't quite add up @ 1600x1200. @ 1920x1200 the difference is 2% with no PhysX and PPU physx.

Yep scaling is less evident at 1600x1200, most likely due to additional CPU overhead from PhysX and less GPU load at lower resolutions. Again, you can verify as frame rates with PhysX on are never higher than with a single GPU with PhysX off, even in SLI.

As for 2% difference at 1920, not sure what you're comparing to as there's a very clear difference between PhysX on/off, PPU and GPU acceleration with the 260 and 9800.