Dedicated GPU Physx benchmark thread.

[Keysplayr]

Ok, so Physx is just about here for the masses. I have lots of news here.
I'll list the games I have been playing with for the past 2 weeks, and I'll also run them on various HW setups and post some benches.

Anyone have any ideas for input, testing, please cough it up.

I have a couple of rigs to use for the benches.

There will be available, on August 12th, a new Physx Pack, as well as a driver that supports Physx across the entire 8, 9 and GT200 series.

Games I have been messing with thus far:

Warmonger (UT3 engine)
GRAW2
UT3

I'll have some benches shortly, and I also have some videos to share.

I'd also ask any Video moderator to keep an eye on this thread and keep it in check. - Thanks.

 

[Keysplayr]

Hardware/Software configurations:
Nothing overclocked in any situation. For now.
I feel that many members would like to see mainstream systems as well as higher end
systems. So 3 rigs will be used for these benches.
Now, there is going to be a LOT of benches to fill in here since I'm running them on 3 separate systems and across a range of resolutions, so please be patient.

1st Rig: AMD AthlonX2 4600+
MSI K9N4-Ultra
2GB DDR2
9800GTX+ (single)
WinXP SP2
Forceware 179.79 beta
Physx Pack

2nd Rig: Intel C2D E6800
ASUS P5N-E SLI
2GB DDR2
9800GTX+ (single)
WinXP SP2
Forceware 179.79 beta
Physx Pack

3rd Rig: Intel Q6600
eVGA 790i Ultra
2GB DDR3
GTX280 (single)
Vista 32
Forceware 179.79 beta
Physx Pack

Games Tested:
UT3
Warmonger
GRAW2

 

[Keysplayr]

Rig1: AthlonX2 4600+ 2GB DDR2 9800GTX+

Warmonger
1280x1024 (higher resolutions will follow shortly)
Using Fraps.

Using software Physx calculations (on CPU) No AA
Min: 4 Avg: 8.13 Max: 21

Using software Physx calculations (on CPU) 4x AA forced in NVCP (works fine. jaggies gone.)
Min: 2 Avg: 6.53 Max: 24

Using hardware Physx calculations (on GPU)
Min: 28 Avg: 39.70 Max: 63

Using hardware Physx calculations (on GPU) 4x AA forced in NVCP (works fine. jaggies gone.)
Min: 29 Avg: 37.83 Max: 59

-------------------------------------------------------

GRAW2
1280x1024 (higher res will follow shortly) edge smoothing AA and 8xAF
Using Fraps

Using software Physx calculations (on CPU)
Min: 2 Avg: 8.37 Max: 25

Using hardware Physx calculations (on GPU)
Min: 6 Avg: 30.05 Max: 81

---------------------------------------------------------

Unreal Tournament 3
1280x1024 (Heat Ray Physx)
Using Fraps

Using software Physx calculations (on CPU) No AA
Min: 6 Avg: 8.31 Max: 10

Using software Physx calculations (onGPU) No AA
Min: 20 Avg: 28.07 Max: 39

xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

Rig2: 3GHz C2D 2GB DDR2 9800GTX+

Warmonger
1680x1050
Using Fraps

Using software Physx calculations (on CPU)
Min: 7 Avg: 13.01 Max: 27

Using software Physx calculations (on CPU) 4x AA forced in NVCP (works fine. jaggies gone.)
Min: 6 Avg: 12.87 Max: 33

Using hardware Physx calculations (on GPU)
Min: 41 Avg: 55.56 Max: 72

Using hardware Physx calculations (on GPU) 4x AA forced in NVCP (works fine. jaggies gone.)
Min: 32 Avg: 37.1 Max: 53

-------------------------------------------------------

GRAW2
1680x1050 edge smoothing AA and 8xAF
Using Fraps

Using software Physx calculations (on CPU)
Min: 6 Avg: 9.32 Max: 19

Using hardware Physx calculations (on GPU)
Min: 24 Avg: 39.73 Max: 67

Pretty amazing what a better CPU can do in these games. Even at higher resolution, the C2D severely outperformed the AthlonX2 in Rig1.

xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

Rig3: Q6600 2GB DDR3 GTX280 (coming next)

Warmonger
1920x1200
Using Fraps

Using software Physx calculations (on CPU)
Min: 6 Avg: 14.18 Max: 37

Using hardware Physx calculations (on GPU)
Min: 48 Avg: 56.61 Max: 78

----------------------------------------------------------------
GRAW2
1920x1200 edge smoothing AA and 8xAF
Using Fraps

Using software Physx calculations (on CPU)
Min: Avg: Max:

Using hardware Physx calculations (on GPU)
Min: 33 Avg: 62.70 Max: 87

xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

Rig3 (second configuration)
3 different methods for Physx use tested.
1. Single 9800GTX+ rendering graphics +Physx
2. 9800GTX+ SLI rendering graphics + Physx
3. 9800GTX+ rendering graphics & second 9800GTX+ running Physx

Notes: I also have an 8800GTS 512, and an 8600GT to utilize for Physx in upcoming tests.
Should be interesting.

 

[dug777]

I await benchmarks with keen interest :beer:

 

[SunnyD]

How about useful numbers - like including a benchmark of Software PhysX versus Hardware PhysX (GPU) versus Hardware PhysX (Actual PhysX hardware)?

Then we can see if NVIDIA is trying to hide anything.

Also, I wouldn't mind seeing results with "Pre-GPU-PhysX" drivers versus newer drivers. And for good measure, benchmarked with comparative ATI cards as well (yes, I know they don't do PhysX - yet).

 

[Keysplayr]

Originally posted by: SunnyD
How about useful numbers - like including a benchmark of Software PhysX versus Hardware PhysX (GPU) versus Hardware PhysX (Actual PhysX hardware)?

Then we can see if NVIDIA is trying to hide anything.

Also, I wouldn't mind seeing results with "Pre-GPU-PhysX" drivers versus newer drivers. And for good measure, benchmarked with comparative ATI cards as well (yes, I know they don't do PhysX - yet).

I don't have an actual Ageia Physx PPU. Sorry. All I can do is a direct comparison between running Physx on a CPU and a GPU.

If you want to see how comparative ATI cards run Physx, see all the CPU scores for your answers.

 

[SunnyD]

Originally posted by: keysplayr2003

Originally posted by: SunnyD
How about useful numbers - like including a benchmark of Software PhysX versus Hardware PhysX (GPU) versus Hardware PhysX (Actual PhysX hardware)?

Then we can see if NVIDIA is trying to hide anything.

Also, I wouldn't mind seeing results with "Pre-GPU-PhysX" drivers versus newer drivers. And for good measure, benchmarked with comparative ATI cards as well (yes, I know they don't do PhysX - yet).

I don't have an actual Ageia Physx PPU. Sorry. All I can do is a direct comparison between running Physx on a CPU and a GPU.

If you want to see how comparative ATI cards run Physx, see all the CPU scores for your answers.

Member of Nvidia Focus Group

You shouldn't have any problems getting one... (Unless of course NVIDIA wants to hide something)

 

[Keysplayr]

Originally posted by: SunnyD

Originally posted by: keysplayr2003

Originally posted by: SunnyD
How about useful numbers - like including a benchmark of Software PhysX versus Hardware PhysX (GPU) versus Hardware PhysX (Actual PhysX hardware)?

Then we can see if NVIDIA is trying to hide anything.

Also, I wouldn't mind seeing results with "Pre-GPU-PhysX" drivers versus newer drivers. And for good measure, benchmarked with comparative ATI cards as well (yes, I know they don't do PhysX - yet).

I don't have an actual Ageia Physx PPU. Sorry. All I can do is a direct comparison between running Physx on a CPU and a GPU.

If you want to see how comparative ATI cards run Physx, see all the CPU scores for your answers.

Member of Nvidia Focus Group

You shouldn't have any problems getting one... (Unless of course NVIDIA wants to hide something)

No reason to get what we don't need. And I suggest you leave this thread immediately if this is how you intend to carry on throughout.

 

[SunnyD]

Originally posted by: keysplayr2003

Originally posted by: SunnyD

Originally posted by: keysplayr2003

Originally posted by: SunnyD
How about useful numbers - like including a benchmark of Software PhysX versus Hardware PhysX (GPU) versus Hardware PhysX (Actual PhysX hardware)?

Then we can see if NVIDIA is trying to hide anything.

Also, I wouldn't mind seeing results with "Pre-GPU-PhysX" drivers versus newer drivers. And for good measure, benchmarked with comparative ATI cards as well (yes, I know they don't do PhysX - yet).

I don't have an actual Ageia Physx PPU. Sorry. All I can do is a direct comparison between running Physx on a CPU and a GPU.

If you want to see how comparative ATI cards run Physx, see all the CPU scores for your answers.

Member of Nvidia Focus Group

You shouldn't have any problems getting one...

No reason to get what we don't need. And I suggest you leave this thread immediately if this is how you intend to carry on throughout.

No need to get in a tissy keys. You asked for requests, I gave you requests.

I don't understand your comment though - I for one would like an apples to apples comparison between discrete PhysX hardware and NVIDIA's GPU solution. I don't get how you can say "No reason to get what we don't need." with a straight face unless there's something to it.

And as I alluded to, as a Focus Group Member and a high profile person on these forums, NVIDIA should have no problems providing you with PhysX hardware. If nothing, it will at least prove or disprove how genuine their PhysX solution is.

 

[Keysplayr]

Look at the topic title Sunny. GPU Physx. Done. Should be the end of this discussion now, yes? Unless you would care to continue to derail? Let us know. No further responses to you. Sorry. Could have been different.

And I can say "No need to get what we don't need" very easily. I already have a Physx processor incorporated into my GPU. Why would I want a discrete PPU taking up another PCI-e slot when I could add a second 9800GTX+ for SLI performance and Physx? Which will come in later benches btw.

I'm sure you'll try to find your conspiracy theories about that one too. Go ahead, but not in this thread. I hope you enjoy the data.

 

[SunnyD]

Originally posted by: keysplayr2003
Look at the topic title Sunny. GPU Physx. Done. Should be the end of this discussion now, yes? Unless you would care to continue to derail? Let us know.

K, you want to play that game that's fine with me. Shouldn't be anything interesting here then. As I said, I'd prefer to see GPU versus PPU benchmarks, because they're more meaningful for a technology comparison, and it would be a good way to show any possible deficiencies with the various technologies. Have fun!

 

[ViRGE]

Oy, cool it down boys. The man doesn't have a PPU so there's not much he can do about it. PPU v. GPU is a valid request, but you're probably going to have to wait until someone else can benchmark it next week unless Keys can get his hands on a PPU. In the mean time please don't start anything, chewing out Keys won't make things any better.

-ViRGE

 

[ViRGE]

BTW Keys, what exactly are you benchmarking in GRAW2? Is it one of the regular levels, or is it the PhysX Island?

 

[chizow]

Originally posted by: SunnyD
You shouldn't have any problems getting one... (Unless of course NVIDIA wants to hide something)

Is this a joke? There is 100% no doubt a GPGPU runs physics better than a CPU and similar to a PPU when available shader cycles permit. Nvidia doesn't have anything to hide on that matter....its clear as day. Its like saying GPU companies were trying to hide something when they said hardware T&L was faster than software T&L.

There's actually a 9800GTX+ vs Ageia PPU vs software comparison floating around, I'll see if I can find it but it shows the GTX+ keeping up with the PPU at lower resolutions but falling behind in GPU intensive resolutions.

Edit: Techgage PhysX Comparison. There's nothing to hide, just like there's nothing to hide when a single GT200 blows away entire server clusters in Folding@Home.

 

[Keysplayr]

Originally posted by: ViRGE
BTW Keys, what exactly are you benchmarking in GRAW2? Is it one of the regular levels, or is it the PhysX Island?

Hey Virge,

It is the Ageia Island level. Search and Destroy.
Warmonger appears to be entirely done with Physx. Wait for me to confirm though.

I'm off to work now, but I'll be back tonight with much more data.

 

[Janooo]

I am looking forward to see higher resolutions. I guess there is going to be limit when running PhysX will slow down graphics. I'd like to know where is the edge.

 

[Keysplayr]

Originally posted by: Janooo
I am looking forward to see higher resolutions. I guess there is going to be limit when running PhysX will slow down graphics. I'd like to know where is the edge.

Last post before I run out the door. Higher resolutions will be posted here later tonight.
See you all later.

 

[EvilComputer92]

As far as UT3 goes, the PhysX version of heatray seems to have about a 20 fps hit in comparison with the regular version. This is with GPU physx enabled. I would like to see a benchmark between the two versions to see what impact all the snowballs falling have.

Also from what I've seen PhysX has zero performance improvement in regular UT3 maps. You only see a gain in the three maps that came with the mod pack. I'd like to see a benchmark to verify this also.

 

[ViRGE]

Originally posted by: EvilComputer92
As far as UT3 goes, the PhysX version of heatray seems to have about a 20 fps hit in comparison with the regular version. This is with GPU physx enabled. I would like to see a benchmark between the two versions to see what impact all the snowballs falling have.

Also from what I've seen PhysX has zero performance improvement in regular UT3 maps. You only see a gain in the three maps that came with the mod pack. I'd like to see a benchmark to verify this also.

AT found ever so slight the increase in one or two normal maps. Apparently this surprised them a bit.

 

[Keysplayr]

Working on it. Results soon.

 

[dadach]

im gonna ask here sorry...if i buy dedicated physx card can i get the full functionality along with my amd 4870? do the current support drivers exist? will the solo card support all the future games?

thanks

 

[dadach]

double :S

 

[Janooo]

Keys, can you, please, update the results with the AA/AF settings used?
Thanks...

 

[MarcVenice]

Come on keys, throw in some AA on those higher resolutions, 4x aa should suffice, with 16xAF ? I wanna see if we once again get crappy minimum framerates or not when running physx on the GPU...

 

[Keysplayr]

Originally posted by: dadach
im gonna ask here sorry...if i buy dedicated physx card can i get the full functionality along with my amd 4870? do the current support drivers exist? will the solo card support all the future games?

thanks

Unknown, at least to me, at the moment. If Nvidia is supporting all GPU's 8 series and up with Physx, I don't see how they would want to continue supporting the PPU card. But that is just the way I see it. I just looked on Nvidia's web site, and I see no drivers for the PPU. I may be looking in the wrong spot, but at first look, I saw none.

You probably have two options at this point. Either wait until Havok Physics is implemented, or if you really want physics capability right now, sell your 4850 (I know, from your cold dead hands right? LOL) and get the equivalent 9 series card that meets your needs.

I don't think purchasing a PPU is a very good idea. There doesn't seem to be much support for it, and no telling for how long it will continue to be supported.

My 2 cents.