PhysX Benchmark on a hybrid system

Phixsator · 11 Oct 2012 at 07:45

tommybhoy said:
Your not going to notice anything I imagine as PhysX doesn't use much of the core at all.

Actually during borderlands 2 playthrough some area my dedicated GTX 260 physX Card would spike through 60% usage only doing physX calculations..

On another note i think nvidia is trying to push some changes to how heavy the physX is implemented in some games. They want sales which is logically which is why we see in games like borderlands 2, Batman arkham city that a dusty old 8800/200 series Card wont be enough anymore. It could be driver manipulation on purpose to slow down performance or just the sheer number of calculations gone thrugh the roof. Neither would really surprise me. I've never had anything against nvidia ( i use to run 2 gtx 560 ti and love them dearly) but their physX business has always been shady *looks at an PPU Card from ageia and crys*

tommybhoy · 11 Oct 2012 at 09:24

Phixsator said:
Actually during borderlands 2 playthrough some area my dedicated GTX 260 physX Card would spike through 60% usage only doing physX calculations..

That's what I heard, any chance you can up a save anywhere so I can test it out myself and see if the 9800GT gets pushed any more?

I thought I saw a delivery van...

Phixsator · 11 Oct 2012 at 10:08

Well ive just done a quick bench of the GT 640 in fluidmark at 60k particals and so fare it doesnt look good

. Looks like its roughly 25% slower than the gtx 260.

layte · 11 Oct 2012 at 10:40

Here is what I had posted before it got nuked for large pics (sorry about that). I'm just cutting and pasting what I posted on another forum I visit with the imbedded pics turned into links.nalso I do have Metro, I'll run some numbers tonight.

The latest version of Fluidmark has much more intensive tests. Here is a simulation of 100,000 particles running on the CPU (this maxed out my 4Ghz hyperthreading enabled i7-920)
http://pcmr.t15.org/stuff/FluidMark 2012-10-09 23-59-04-01.png

and the same running on the GPU (with nominal CPU use)
http://pcmr.t15.org/stuff/FluidMark 2012-10-09 23-58-10-22.png

I then did a test with 200,000 particles running on the GPU, it it still gave double the framerate of the 100k CPU test.
http://pcmr.t15.org/stuff/FluidMark 2012-10-10 00-01-09-32.png

You can grab the latest version of fluidmark from here: http://www.ozone3d.net/benchmarks/physx-fluidmark/

layte · 11 Oct 2012 at 10:44

Phixsator said:
Well ive just done a quick bench of the GT 640 in fluidmark at 60k particals and so fare it doesnt look good . Looks like its roughly 25% slower than the gtx 260.

I score 2831 on the 1080p 60k pre baked bench. Compared to other cards running the display and PhysX that's pretty darned good. http://www.geeks3d.com/20120502/physx-test-gtx-680-vs-gtx-580-vs-gtx-480-in-fluidmark/

Rusty0611 · 11 Oct 2012 at 11:04

Come on Tommy - get on it as well

!

Phixsator · 11 Oct 2012 at 11:38

Well had a bad feeling about the last bench and decided to reinstall the entire system and start over on a proper platform (win 7) and its already looking better and makes more sense

shankly1985 · 11 Oct 2012 at 16:34

Think am going to pass installing this card from what I have found out its either a 9600+ or 9600GSO with
128-bit
Core Clock: 600 MHz
Memory Clock: 800 MHz
Direct X support: DX10
Stream Processors: 48
Shader Clock: 1500 MHz

Seems far to slow

layte · 11 Oct 2012 at 19:14

A few quick Metro numbers. To be honest the most this hit my 480 for was 12% usage, so not the best test really.

Normal physics on the CPU

Advanced (GPU) PhysX enabled

Couldn't see much of a difference between them, certainly not enough to warrant a dedicated card.

Phixsator · 11 Oct 2012 at 20:06

Finaly the first batch of tests are done. Lost 6 hours of work due to BSOD and a corrupted spreadsheet

but here it is as promised. The link is in the OP which has been updated.

shankly1985 · 11 Oct 2012 at 20:19

Thanks OP seems the GTX260 is the card to be using from your tests. Have you got Borderlands 2? If so how does it perform on that?

Phixsator · 11 Oct 2012 at 20:27

shankly1985 said:
Thanks OP seems the GTX260 is the card to be using from your tests. Have you got Borderlands 2? If so how does it perform on that?

Yes i have BL2. Problem so fare has been every test has been suffering from minor stuttering, small enough so the FPS counter wouldnt catch it but big enough so it annoys me. Borderlands 2 is even worse in this regard. Running some of the tests a second time straight away reduced the stuttering which would to be suggest i have some sort of caching problem somewhere in my system. Disabling hardware physX removed all stutter nomatter what test i ran.

Also found that using the dedicated PhysX card in a 8x PCIe 2.0 lane would reduce the performance with 20% roughly over a full speed 16x PCIe 2.0 lane. Same result on my older gigabyte x58-ud4p motherboard. I find that really strange.

Phixsator · 13 Oct 2012 at 08:35

A little update: The overall feel from actual gameplay was better on the GT 640 in both Batman and Borderlands 2. Going to do some proper comparison after i get home sunday.

Also you dont want to use any sockets for your physX card thats running off the soouthbrigde, Causes a good deal of slow down atleast on my rig.

And last regarding lag/stutter issues in Borderlands 2. Ive found that gearbox has done something to the physxcore.dll that they are using.. Using that, my CPU was able to go through bloodshot stronghold on medium setting almost all the way through(slow downs to 20 fps was seen in the big whirlpool area and a few other places). Now since im on a hybrid system i cannot use that physXcore.dll that comes with the game because it disables GPU acceleration and if the game uses ANY other physxcore.dll than what was shipped with the game the performance is tanking or stuttering regardless of being powered by the CPU or the GPU.

layte · 14 Oct 2012 at 00:18

I've not had any lag/stutter issues myself. I went through the whirlpool room earlier today, and with PhysX on high GPU use on my 480 hit 48% at one point. Perhaps the 640 just doesn't have the grunt required in places? I'd imagine 48% use on a 480 is still more than what a 640 can muster, especially as the most I've ever seen PhysX push the 480 is 68% on a Fluidmark run of 300k particles.

I've also made a few tweeks to the ini file, increasing GPU PhysX memory use from 128MB to 1GB and the mesh cache from 8 to 32

from

Code:

PhysXGpuHeapSize=128
PhysXMeshCacheSize=8

to

Code:

PhysXGpuHeapSize=1024
PhysXMeshCacheSize=32

Edit: If you compare compute performance of the slower 470 against the 650 (faster than the 640, but similar enough to give meaningful numbers), it's a curbstomp victory for the 470.

humbug · 14 Oct 2012 at 00:42

Great bit of work.

no_1_dave · 14 Oct 2012 at 01:53

layte said:
Edit: If you compare compute performance of the slower 470 against the 650 (faster than the 640, but similar enough to give meaningful numbers), it's a curbstomp victory for the 470.

Like I said, the 6xx range are not designed with compute in mind. The 4xx/5xx range are much better at compute tasks, especially folding etc. Not sure if Physx comes under that same category or not...

tommybhoy · 14 Oct 2012 at 03:35

@Phixsator,

Great info, just by starting the thread I have gained some very valuable info giving me a huge boost.

I asked earlier, and you never replied, is there any way to give me a save for the heavy PhysX areas in B2?

It will be some time before I get round to playing it and I just want to get some data for the thread.

layte said:
I've also made a few tweeks to the ini file, increasing GPU PhysX memory use from 128MB to 1GB and the mesh cache from 8 to 32

from

Code:

PhysXGpuHeapSize=128 PhysXMeshCacheSize=8

to

Code:

PhysXGpuHeapSize=1024 PhysXMeshCacheSize=32

Thanks for that tip, it has considerably boosted my PhysX performance!

I added it to BAC and the results are below, before I got a min of 9fps!

Rusty0611 said:
Come on Tommy - get on it as well !

Just for you Rusty.

BAC Everything@max inc PhysX:

Doubled the mins by doing layte's tip, I owe you a pint mate.

If I can find a 46/70 for £4/50, then I think I'm going to have some fun!

Rusty0611 · 14 Oct 2012 at 11:27

Keep checking the MM. They can go quite cheaply

.

Good info.

Phixsator · 15 Oct 2012 at 10:47

tommybhoy said:
@Phixsator,
I asked earlier, and you never replied, is there any way to give me a save for the heavy PhysX areas in B2?

So sorry about that. Must have overlooked it. Here you go

https://docs.google.com/open?id=0B0zUB4Ihwv5sLVVlQTJJSWg3RlU

I actually think the GT640 is more than plenty as a dedicated PhysX but i cant test it for sure since im not running a pure nvidia enviroment. The PhysXcore.dll shipped with borderlands 2 has been modified a lot and im sure of this because its the only PhysXcore.dll that will give you playable fps running PhysX on medium and on the CPU. My guess is still that if i were somehow able to get hardware acceleration using that dll on my hybrid system the GT640 would truly shine.

I've had a short look on the hybridiz patch that most seems to be using and while it works i find its methods a bit bad to be honest. The old PhysX Mod 1.05ff would patch your files which is what we want since we get to keep any optimizations done. The Hybridiz seems to just replace them with prefixed files which is why it always works cause the files are always the same. This IMHO is extremely poor done and not what we want to use since the software is changing all the time and 3rd party developers are using custom physX files for their games which is thrown out the window with this patch for those of us using a hybrid system.

layte · 15 Oct 2012 at 16:04

I'm not sure about that. The 640 does not have all that much power. As an example, Batman AC suggests a dedicated 460 for the best results with high PhysX. That is more power than what a 640 can bring to the table. Plus like I have said, B2 has hit 50% use of a 480 in certain areas. That is probably more than 50% use of the available compute power as the most use I can force out of Fluidmark is only 68% no matter how many particles I toss at it