PhysX F.A.Q. Compilation of questions I am finding throughout the forum.

[imported_D3v]

Keys what method are you using? Fraps? The 1.3 benchmark tool? That hardspell article is mighty, ah, questionable. I'll have the 3870 in my box in about 15 minutes, will see if I can do anything useful with it and my 8800 paired up

 

[MarcVenice]

Keys, here's something that has been bugging me. How do you actually dis/enable the physx being run on the GPU. Or do you have to install an 'old' set of drivers, or can you uninstall the physx driver? It's prolly a dumb question, but since your benching it, you must know

 

[Keysplayr]

D3v: I'm using Fraps to capture min/max/avg fps.

Marc: You can uninstall the PhysX Driver in Windows XP Control Panel > Add Remove Programs, or Vista Control Panel > Programs and Features.
Uninstall Nvidia PhysX v8.08.18

 

[Nemesis 1]

Looks like NV opened a case of whoop ass on intel . With a real demo of what NV is truely capable of when it comes to ARt of chip making. Maybe this will stay the hands of NVs competion.

http://www.youtube.com/watch?v=fKK933KK6Gg

 

[Nemesis 1]

Keys have ya done test using differant platforms? 1 core cpu / 2 core cpu / 4 core cpu . Than report the results . CPU usage seems way high for the gpu to be doing all the physics. What I want to know is does differant cpus result in higher lower scores? IF so why?

 

[Keysplayr]

Originally posted by: Nemesis 1
Keys have ya done test using differant platforms? 1 core cpu / 2 core cpu / 4 core cpu . Than report the results . CPU usage seems way high for the gpu to be doing all the physics. What I want to know is does differant cpus result in higher lower scores? IF so why?

As compared to what?

 

[will889]

Originally posted by: Nemesis 1
Looks like NV opened a case of whoop ass on intel . With a real demo of what NV is truely capable of when it comes to ARt of chip making. Maybe this will stay the hands of NVs competion.

http://www.youtube.com/watch?v=fKK933KK6Gg

In Theory and maybe in future practice, but not anything usable today in meaningful games (as of today). On the same token ATI is actually handing NV's its collective ass on a plate with performance vs. pricing with the 4850-4870-4870x2. Right now.

 

[Keysplayr]

This is a PhysX FAQ thread folks. Keep it on topic would ya?

 

[Nemesis 1]

Originally posted by: keysplayr2003

Originally posted by: Nemesis 1
Keys have ya done test using differant platforms? 1 core cpu / 2 core cpu / 4 core cpu . Than report the results . CPU usage seems way high for the gpu to be doing all the physics. What I want to know is does differant cpus result in higher lower scores? IF so why?

As compared to what?

Well the simplest test would be to run A game without using PX. Than turn PX on . and report Cpu usage with and without . Pretty simple and easy to reproduce the same results.

 

[SSChevy2001]

Originally posted by: Nemesis 1
Well the simplest test would be to run A game without using PX. Than turn PX on . and report Cpu usage with and without . Pretty simple and easy to reproduce the same results.

You would need a older PPU PhysX card to compare against if your comparing a game like UT3, but we can use badaboom and the physx demos to get a good picture of the CPU usage.

3.45ghz C2Q

Badaboom CPU Usage with CUDA ( PhysX on only works )
25-30% performance hit depending on the frames it's encoding

PhysX Fluid Demo CPU Usage
PhysX on - 30-34%

Taken from the processes section of the task manager, not total CPU usage.

So there's still plenty of work being done on the CPU, and I hate to see the usage on a C2D.

 

[Keysplayr]

Originally posted by: Nemesis 1

Originally posted by: keysplayr2003

Originally posted by: Nemesis 1
Keys have ya done test using differant platforms? 1 core cpu / 2 core cpu / 4 core cpu . Than report the results . CPU usage seems way high for the gpu to be doing all the physics. What I want to know is does differant cpus result in higher lower scores? IF so why?

As compared to what?

Well the simplest test would be to run A game without using PX. Than turn PX on . and report Cpu usage with and without . Pretty simple and easy to reproduce the same results.

Could you run that test? I'm a little backed up right now and time is a little tight to run another round of benches at the moment. But will do first chance I get. In the mean time, would you be able to run them and post here?

 

[Keysplayr]

Yeah, I think it has been shown in another review that more CPU is used when running PhysX on a GPU over the PPU.
And you also get more performance out of the GPU over PPU. So, it seems Nvidia utilizes the resources it has and that includes more CPU cycles. If you get better performance as a result, I see benefit. Todays CPU's are more than enough for most games anyway and these games tend to end up more GPU limited than CPU limited. So there would be CPU cycles to spare as a result. PhysX on GPU uses it's resources a bit better than the CPU/PPU solution.

 

[imported_D3v]

fluidmark is not multithreading yet and therefore not a valid test for multicore

x2 @ 2.5 w/physx on 8800gt = core1 avg 60%, core2 avg ~1%
x2 @ 2.5 w/software physx = core1 avg 100%, core2 avg ~1%

 

[Narynan]

Did we ever figure out weather a 8400 would be a good choice for a Physix card? becasue thats the one I would be looking at for when the games start hitting.

 

[Keysplayr]

Originally posted by: Narynan
Did we ever figure out weather a 8400 would be a good choice for a Physix card? becasue thats the one I would be looking at for when the games start hitting.

I would wait til the games start to hit. There will be more powerful secondary cards for less money. I am unable to test an 8400GS at the moment. I was wondering if anyone else had one to try alongside another NV card.

 

[Sylvanas]

Originally posted by: Narynan
Did we ever figure out weather a 8400 would be a good choice for a Physix card? becasue thats the one I would be looking at for when the games start hitting.

I have been searching for a definitive answer for this as well. There is a clearance sale on near me and they have 8400GS's selling for cheap, tempted to go pick one up but I don't want to if later on down the track it turns out that if I spent a little more on a better secondary PhysX card I would be getting better performance. I am sure there must be a sweet spot in terms of the hardware required to run dedicated PhysX and beyond a certain point there is no benefit- I am interested in where this is. Perhaps this is an idea for an AT article, I know many of us out there are looking for such an article.

I would wait til the games start to hit. There will be more powerful secondary cards for less money. I am unable to test an 8400GS at the moment. I was wondering if anyone else had one to try alongside another NV card.

Keys, you say wait for the games to hit but how would that be any different to testing them today on say the UT3 maps? Will PhysX be more/less demanding depending on games? This relates to what I said above about where the sweet spot is. I may venture to answer my own question here but I may be wrong. Say a developer decides to use PhysX and when designing an object like a car they decide that the car will break into 1000 fragments as opposed to 100 when exploded. In that instance I would think PhysX performance must depend on how the developer implements it, correct?

 

[Keysplayr]

Originally posted by: Sylvanas

Originally posted by: Narynan
Did we ever figure out weather a 8400 would be a good choice for a Physix card? becasue thats the one I would be looking at for when the games start hitting.

I have been searching for a definitive answer for this as well. There is a clearance sale on near me and they have 8400GS's selling for cheap, tempted to go pick one up but I don't want to if later on down the track it turns out that if I spent a little more on a better secondary PhysX card I would be getting better performance. I am sure there must be a sweet spot in terms of the hardware required to run dedicated PhysX and beyond a certain point there is no benefit- I am interested in where this is. Perhaps this is an idea for an AT article, I know many of us out there are looking for such an article.

I would wait til the games start to hit. There will be more powerful secondary cards for less money. I am unable to test an 8400GS at the moment. I was wondering if anyone else had one to try alongside another NV card.

Keys, you say wait for the games to hit but how would that be any different to testing them today on say the UT3 maps? Will PhysX be more/less demanding depending on games? This relates to what I said above about where the sweet spot is. I may venture to answer my own question here but I may be wrong. Say a developer decides to use PhysX and when designing an object like a car they decide that the car will break into 1000 fragments as opposed to 100 when exploded. In that instance I would think PhysX performance must depend on how the developer implements it, correct?

While it might be fine, I don't want to give an answer until I have tested it. The reason I am apprehensive about it is because of my test with the 780a using the onboard GPU for PhysX reduced performance rather than let the 9800GTX+ I was testing it with do all the work itself. Now, I can attribute the IGP's lower performance (which reports 16 shaders if GPUz is accurate) to it having to utilize the system memory, in fact, it's pretty much a given, but until I have an 8400GS tested, I can just say "Sure, go get it.".
And, there seems to be 8400GS' that have 8 stream processors, and some with 16 stream processors. The 780a I tested with had 16. And that didn't do so hot. But like I said, that most likely could be contributed to the system ram.

The 8600GT I tested worked nicely, bringing up minimum framerates 20 to 30% in UT3 and GRAW2. That is a 32 stream processor card.

"In that instance I would think PhysX performance must depend on how the developer implements it, correct?"

I would agree. The way any game performs depends on how the dev implements the code.

I said to wait, because I didn't want anyone doing anything that "could" make their PhysX game perform worse (going by my 780a test). But I don't have enough data yet to actually substantiate this.

 

[MarcVenice]

Keys, is it possible to disable the dedicated physx videocard. If I'm sitting in my desktop I'd hate to have my main videocard sucking power, and my old nvidia card as well. I don't think it is btw, but I might have missed something. Maybe that's nvidia to work on?

 

[Keysplayr]

Chances are, if you are at an enthusiasts level enough to want a second video card in your system, be it for SLI, Crossfire, or PhysX/CUDA, then I'm not so sure power consumption is on their minds. I don't see folks trying to disable their second card in their SLI or Crossfire setups. Same principal. Desktop apps take no advantage with multicard, (but CUDA seems to be changing all that) unless you are running that sweet new CUDA enabled Adobe . But, you get the idea.

However......

If you set your system up in a savvy way, you could have a nice powerful system that only sips juice when not being used for anything but desktop stuff. Hybrid power. Everybody is trying like mad to go "green" lately.

 

[Narynan]

Yeah, from what I am looking at and what I have been able to tell, I am thinking that an older model ?600 - ?800 model is what were are going to be looking at for that extra kick from the PhysX engine. I have been seeing some 8600gt's on my local craigslist for about 60 bucks. And that seems about right for me for the preformance differnce I will be seeing on the games available.

 

[Keysplayr]

Originally posted by: Narynan
Yeah, from what I am looking at and what I have been able to tell, I am thinking that an older model ?600 - ?800 model is what were are going to be looking at for that extra kick from the PhysX engine. I have been seeing some 8600gt's on my local craigslist for about 60 bucks. And that seems about right for me for the preformance differnce I will be seeing on the games available.

You could be right, but that is not yet confirmed. I am working on getting an 8400 series card. I've noticed them listed with 8 stream processors, and with 16. I'll only test the 16 stream processor card for a few reasons.

1. Conflicting stream processor count leads me to believe that online retailers really don't know how many sp's the 8400GS has.

2. EVGA's web site shows 16 sp's for the cheapest 8400GS they have. (I'll go with them)

3. The IGP on the 780a reported 16 sp's, so I'd like to have a direct comparison with that.

4. Online retailers prices seem to be the same for the 8 or 16 sp cards. So why not go with the 16 (which is what I believe all 8400GS have).

 

[BAMAVOO]

All I can say is I tried it and I like it. Warmonger is pretty cool, but I can't wait to give that football game a try.

 

[shaolin95]

How does the PCIe slot affects the gpu physx card? My mobo runs either 16x main slot and 2x second slot or 8x both. I dont to run my GTX280 at 16x but I am wondering if trying my 8800GT KO in the 2x slot is even worth it....any thoughts?

 

[PrinceofWands]

City of Heroes/Villains has had Physx support for some time now.

 

[Keysplayr]

Some updates to the OP. And thanks for stickifying the thread!!