New PhysX thread.

[Keysplayr]

PhysX Testing Results CLICK HERE


Platform:

Core2 Quad Q6600 @ 3.0 GHz
eVGA 790i Ultra SLI
6GB DDR3 (Will pick up another 2GB for this today 8/23.09.
WD 120GB for OS
Seagate 500GB for games
Samsung 2233RZ 22" native 1680x1050
Windows 7 64-bit
Forceware 190.62 WHQL


Primary Rendering card: GTX295 and GTX280
Dedicated PhysX GPUs: 8600GT, 8800GTS 512, GTX280


Games:
Cryostasis
Batman Arkam Asylum Demo
Mirrors Edge

Running this test to determine answers to frequently asked questions, which will be added to the 3rd post as a F.A.Q. like we did a while back. many things have changed since then. I thought it was time for an updated thread.

So ask any questions about the testing. I'll enter them in the 3rd post.

 

[Keysplayr]

PhysX Testing Results CLICK HERE

All tests being redone due to buggy Vista system. Redoing all benches with Win7 64 bit.

Interesting combination with a GTX280 and a 8600GT.
Check out the link.
Currently swapping out the 8600GT for the 8800GTS 512. Will update soon.

 

[Keysplayr]

Reserved for F.A.Q.

 

[apoppin]

PM sent; thanks for unblocking.

i think PhysX is becoming important and at the very least, it should be tested thoroughly. i am glad to help if possible.

 

[lavaheadache]

I suppose i could do some testing of my own. I have 2 9600gt SC's I could try in sli and then the 280 gtx setup as a phsyx card. what's sweet is that I don't even have to pull my machine apart to switch bettween main cards because of the quad monitor setup up. just switch which monitor is the primary in the control panel and the driving card will render it.


let me know if you want me to post some results cause i don't wanna hijack your thread

 

[Keysplayr]

Originally posted by: lavaheadache
I suppose i could do some testing of my own. I have 2 9600gt SC's I could try in sli and then the 280 gtx setup as a phsyx card. what's sweet is that I don't even have to pull my machine apart to switch bettween main cards because of the quad monitor setup up. just switch which monitor is the primary in the control panel and the driving card will render it.


let me know if you want me to post some results cause i don't wanna hijack your thread

Post away!!

I know you have a quad monitor setup, but wouldn't your gaming be a bit quicker with the GTX280 as your primary and one of those 9600GT's as the PhysX card?

You could still run quad monitors AFAIK.

 

[lavaheadache]

yeah it is way quicker with the GTX280 running primary. I was just stating for the sake of testing 9600gt with 9600gt as physx and then the gtx 280 as the physx card to see if there is a benefit from a faster physx card. I'll try to get them up some time in the next 24 hours.

 

[Keysplayr]

Ah, I gotcha.

 

[vj8usa]

Could you also benchmark with PhysX off and other settings the same, as a baseline? Also, is PhysX on normal or high in your Arkham Asylum tests?

 

[Keysplayr]

Originally posted by: vj8usa
Could you also benchmark with PhysX off and other settings the same, as a baseline? Also, is PhysX on normal or high in your Arkham Asylum tests?

I could do PhysX off. But what for? I'm trying to find a minimum all around sweet spot for a dedicated PhysX GPU.

And PhysX set to highest in Arkam Asylum.


UPDATE: Test rig now has 4GB DDR3. OP updated.

 

[cmdrdredd]

Is there a benchmark or a way to timedemo batman? How did you determine the fps numbers?

 

[apoppin]

Originally posted by: Keysplayr

Originally posted by: vj8usa
Could you also benchmark with PhysX off and other settings the same, as a baseline? Also, is PhysX on normal or high in your Arkham Asylum tests?

I could do PhysX off. But what for? I'm trying to find a minimum all around sweet spot for a dedicated PhysX GPU.

And PhysX set to highest in Arkam Asylum.


UPDATE: Test rig now has 4GB DDR3. OP updated.

i sent a PM with a few suggestions

test it without PhysX - disable it in the CP so we have benchmark on GTX295 + low-end 8400GS
- Perhaps without the 8400 at all

THEN test it 'enabled on the GTX 295' [just so you can keep it in the 2nd slot]

THEN test it 'enabled on the 8400'

THEN test it with your fastest GPU enabled in the PhysX slot

that way we see what is happening

 

[Keysplayr]

Originally posted by: cmdrdredd
Is there a benchmark or a way to timedemo batman? How did you determine the fps numbers?

Use Fraps. Just do walk through and fight the first 10 or so enemies. Do this three times, and get an average from your three runs.

 

[Keysplayr]

Originally posted by: apoppin

Originally posted by: Keysplayr

Originally posted by: vj8usa
Could you also benchmark with PhysX off and other settings the same, as a baseline? Also, is PhysX on normal or high in your Arkham Asylum tests?

I could do PhysX off. But what for? I'm trying to find a minimum all around sweet spot for a dedicated PhysX GPU.

And PhysX set to highest in Arkam Asylum.


UPDATE: Test rig now has 4GB DDR3. OP updated.

i sent a PM with a few suggestions

test it without PhysX - disable it in the CP so we have benchmark on GTX295 + low-end 8400GS
- Perhaps without the 8400 at all

THEN test it 'enabled on the GTX 295' [just so you can keep it in the 2nd slot]

THEN test it 'enabled on the 8400'

THEN test it with your fastest GPU enabled in the PhysX slot

that way we see what is happening

The last 3 make sense and I intend on doing exactly that, but running without PhysX on will make these games CRAWL. I'm not doing the "with or without" PhysX tests. Those were done ages ago. This is a "what is the best, or most sufficient, GPU for the dedicated PhysX job" test.
I'm not understanding the purpose of testing WITHOUT PhysX on/disabled. We all know what happens. Performance tanks to single digits sometimes.

 

[apoppin]

Originally posted by: Keysplayr

Originally posted by: apoppin

Originally posted by: Keysplayr

Originally posted by: vj8usa
Could you also benchmark with PhysX off and other settings the same, as a baseline? Also, is PhysX on normal or high in your Arkham Asylum tests?

I could do PhysX off. But what for? I'm trying to find a minimum all around sweet spot for a dedicated PhysX GPU.

And PhysX set to highest in Arkam Asylum.


UPDATE: Test rig now has 4GB DDR3. OP updated.

i sent a PM with a few suggestions

test it without PhysX - disable it in the CP so we have benchmark on GTX295 + low-end 8400GS
- Perhaps without the 8400 at all

THEN test it 'enabled on the GTX 295' [just so you can keep it in the 2nd slot]

THEN test it 'enabled on the 8400'

THEN test it with your fastest GPU enabled in the PhysX slot

that way we see what is happening

The last 3 make sense and I intend on doing exactly that, but running without PhysX on will make these games CRAWL. I'm not doing the "with or without" PhysX tests. Those were done ages ago. This is a "what is the best, or most sufficient, GPU for the dedicated PhysX job" test.
I'm not understanding the purpose of testing WITHOUT PhysX on/disabled. We all know what happens. Performance tanks to single digits sometimes.

i get 3 outta 4, huh?
- not bad .. 75%
:thumbsup:

- those were the main 3 -

1. 1. GTX 295 doing all the work
   2. GTX 295 + slow card
   3. GTX 295 + fast card

However, without physX on the GPU sets a 'standard' .. just to see if anything was updated since last year when this was mostly explored

Here is the question - were these old tests done by disabling PhysX or shifting it to the CPU?
- what i wannt to know is, will framerates tank if you disable PhysX altogether? - if you do not put it on on the GPU or CPU at all?
- completely disable HW physics as an option

 

[Qbah]

I would also like to see how much of a hit PhysX is performance-wise. For example we can run AAx8 and that usually cuts the framerates in half. We know that because we run the test with AAx0 first and then with AAx8 and compare Batman, Mirror's Edge and Cryostasis most likely have an option to disable the PhysX effects altogether - by running a test like that we will know how much performance PhysX uses. Also, any chance running Sacred 2?

My wild guess is the performance should be similar with PhysX on to running a better extra PhysX card (like a 9600GT+). Though it would be nice to have some numbers on it Thanks in advance!

 

[happy medium]

Finally something worth reading. Thanks Keys:thumbsup:

 

[SSChevy2001]

What AA settings are you using for batman?
Also can you redo batman with the bsmoothframerate off?

Here's the average I'm getting with 10 person kill running with the following ( 4 runs through )
4870 1GB 9.8 (810/975) + 8800GTS 512MB 186.16 (821/1890/1053)
Batman Arkam Asylum
1680x1050 max - ingame 4xAA - PhysX on high

31/111/63.613
30/111/64.687
32/111/65.186
27/111/65.228

30min / 111max / 65avg

 

[cmdrdredd]

Batman bench below for me. Everything high 1920x1200 2xAA. Rig in sig

GTX295 alone
26min/60max/42avg

GTX295+9600GT for physx
27min/63max/57avg

So it would seem the 9600GT helps a good deal. Cryostasis saw about 20fps increase.

edit: 4x AA loses 3fps average in each bench. Still perfectly playable.

 

[Sylvanas]

Finally some testing between PhysX standalone GPUs I've been searching forever for this! Thanks for the input Keys and others, it confirms my suspicions that for all games to run better than a standalone top end GPU we need something like a 8600GT as a minimum. Although of course with a lesser setup than a GTX295 (maybe a 275 or something) there would be less horsepower available for PhysX on the GPU alone and perhaps a 8400GS would make a positive difference here- be interesting if anyone can clear that up.

 

[toyota]

Originally posted by: cmdrdredd
Batman bench below for me. Everything high 1920x1200 2xAA. Rig in sig

GTX295 alone
26min/60max/42avg

GTX295+9600GT for physx
27min/63max/57avg

So it would seem the 9600GT helps a good deal. Cryostasis saw about 20fps increase.

edit: 4x AA loses 3fps average in each bench. Still perfectly playable.

what is wrong with the gtx295 in Batman? I have an older 192sp gtx260 I can nearly match your gtx295 numbers with the identical settings except 1920x1080. I did my runs during the entire first fight scene which is pretty demanding and my minimums even match your gtx295 and 9600gt results. even if sli isnt working you still have 48 more SP than my gtx260 so why is your setup that slow.


here are a couple runs on all very high graphics, physx on high, 2x AA 1920x1080. again this was pretty much all fighting so in other parts of the game my framerate is even higher.

192sp GTX260 alone. first run I looked at a wall for just a second so I guess thats why it hit 62fps. lol


Frames, Time (ms), Min, Max, Avg
2678, 69054, 29, 62, 38.781

Frames, Time (ms), Min, Max, Avg
2321, 59309, 26, 50, 39.134

 

[thilanliyan]

Originally posted by: Sylvanas
perhaps a 8400GS would make a positive difference here- be interesting if anyone can clear that up.

Minimum is 8600gt/9500gt since the 190 drivers according to what Keysplayr said.

 

[cmdrdredd]

Originally posted by: toyota

Originally posted by: cmdrdredd
Batman bench below for me. Everything high 1920x1200 2xAA. Rig in sig

GTX295 alone
26min/60max/42avg

GTX295+9600GT for physx
27min/63max/57avg

So it would seem the 9600GT helps a good deal. Cryostasis saw about 20fps increase.

edit: 4x AA loses 3fps average in each bench. Still perfectly playable.

what is wrong with the gtx295 in Batman? I have an older 192sp gtx260 I can nearly match your gtx295 numbers with the identical settings except 1920x1080. I did my runs during the entire first fight scene which is pretty demanding and my minimums even match your gtx295 and 9600gt results. even if sli isnt working you still have 48 more SP than my gtx260 so why is your setup that slow.


here are a couple runs on all very high graphics, physx on high, 2x AA 1920x1080. again this was pretty much all fighting so in other parts of the game my framerate is even higher.

192sp GTX260 alone. first run I looked at a wall for just a second so I guess thats why it hit 62fps. lol


Frames, Time (ms), Min, Max, Avg
2678, 69054, 29, 62, 38.781

Frames, Time (ms), Min, Max, Avg
2321, 59309, 26, 50, 39.134

Look at the averages. I'm way ahead of you, portions of this game simply run slow although not noticably so.

There is nothing "wrong" with my 295 at all. Also I do not use 190.xx drivers. They are total failures by Nvidia because when I do 3D in a window with Aero, the game becomes a slideshow. Until they fix this I will not move from 186.18

 

[SSChevy2001]

Here's what I'm getting at 1920x1080 max - ingame 4xAA - PhysX on high
4870 1GB 9.8 (810/975) + 8800GTS 512MB 186.16 (821/1890/1053)
Batman Arkam Asylum

28/99/58.295
31/100/58.099
26/101/60.08
32/99/58.677

29min / 100max / 59avg

 

[cmdrdredd]

Originally posted by: SSChevy2001
Here's what I'm getting at 1920x1080 max - ingame 4xAA - PhysX on high

28/99/58.295
31/100/58.099
26/101/60.08
32/99/58.677

29min / 100max / 59avg

Since you don't have an Nvidia card isn't physx off anyhow?