Comparison of Running PhysX w/a CPU, the PPU, and the GPU

[GoHack]

Here's a comparison of running PhysX w/a CPU, the PPU, and the GPU.

http://techgage.com/article/nvidias_physx_performance_and_status_report/

From the benchmarks in the article, it doesn't look like those PhysX cards that some of us purchased are obsolete. In fact the cards help in the processing of the graphics, helping to take the load off of the CPU and GPU's. In the Unreal Tournament III Performance, the PPU helped add another 10 FPS.

.

 

[Atech]

Nice find...so NVIDIA's 9800GTX can STILL not beat old hardware from 2006...with a TDP of 28 Watt...dedicated hardware is still the champ.

 

[Dual]

I noticed that too. The author seemed to downplay the fact that the hardware PPU was still the best. The new physx mod for ut3 can be found on guru3d.

 

[Atech]

I noticed that too. The author seemed to downplay the fact that the hardware PPU was still the best. The new physx mod for ut3 can be found on guru3d.

It actually shows more than that.
It's show that the PPU would have flabbergasted any 6800/X800 series cards in physics and stomped all over any 7x00/X1x00 series in physcis too.

That gives an interresting spin on the claims by NVIDIA/ATI..and the pro-GPU posters in this forum...they where all bunked, bogus or a pipedream...go figure

 

[cybereality]

Yeah, good article. What I'm wondering is if Nvidia will support the hardware card too. Lets suppose this PhysX GPU thing takes off and Nvidia pushes more developers to support it. Would the PPU solution see the benefits of this? Because it still looks like a separate card is a better option in terms of overall FPS.

 

[GoHack]

One thing that this article doen't talk about, and tests, is running PhysX in SLI, or on a GX2 card. Only one GPU is used in this article.

In the article, the single GPU is overloaded, doing not only it's normal job of processing graphics, but running the PhysX engine as well.

.

 

[Atech]

One thing that this article doen't talk about, and tests, is running PhysX in SLI, or on a GX2 card. Only one GPU is used in this article.

In the article, the single GPU is overloaded, doing not only it's normal job of processing graphics, but running the PhysX engine as well.

.

NVIDIA said they could do GPU-Physics, they couldn't, they had to buy AGEIA.
If they want people to go SLI, compared to a 28 Watt TDP PPU, they are moronic...

 

[GoHack]

NVIDIA said they could do GPU-Physics, they couldn't, they had to buy AGEIA.
If they want people to go SLI, compared to a 28 Watt TDP PPU, they are moronic...

No they're not. They can make a lot more money off of selling a second video card vs. selling just a single video card, or even just a 28 watt TDP PPU.

.

 

[Atech]

No they're not. They can make a lot more money off of selling a second video card vs. selling just a single video card, or even just a 28 watt TDP PPU.

.

Only if they sucker people in...but then again...the is *bling-bling*GPU-physics*bling-bling*...way more cooler than PPU-physics
So who cares that it cost more and uses more power...it GPU-physics

Will be funny to see people that bashed PhysX suddenly a like it...watch this subforum. the comming hypocracy will give pains...of laughter.

 

[GoHack]

Only if they sucker people in...but then again...the is *bling-bling*GPU-physics*bling-bling*...way more cooler than PPU-physics
So who cares that it cost more and uses more power...it GPU-physics

Will be funny to see people that bashed PhysX suddenly a like it...watch this subforum. the comming hypocracy will give pains...of laughter.

You got suckered into buying a PhysX card didn't you?

.

 

[Balthazor]

I got it fully realizing it won't pan out. But I was glad when nvidia purchased Aegia. A lot more developers will back a physx model if the majority of systems out there can run physx on their GPUs.

If having an extra PPU will give me a few FPS of a performance, then great. If not, well it wasn't that expensive.

 

[SuperGee]

What i expected good old PPU is powerfull.
A lightweight 130nm still beats a modern 45nm Quadcore. in a quadtheaded game engine.
cPU is stil caching up.

GPU must be more powerfull by its raw power out of transistor count high clocks of the shaders possible with a modern Procede for GPU.

Also if your on a budged a GPU wil not do that good in rendering and PhysX at the same time. Both are computing power hungry.
3D vantage give you a wrong synthatic impression.
The cheat is. It show 10x Physics performace but in real game it could be nothing to 50%
That a lot less then up to 1000% increase. Wile as the PPU is dadicated it performance count. Ofcourse if you keep in mind it is syntatic. Analise it. No harm done. What it show is what a dedicated G-card could do. That impresive as expensive. SLI could ease the $ pain.

What I don't understand is. The nV shader rop TMU etc ratio. Is de GT200 design coming out far before nV GPGPU PhysX plans? I would expect a far more shader heavy design.

For Physics the aTI shader ratio make more sense. that is still more shader heavy.
nV is more conservative with its choice. As if they forgot to put some extra for physX.
The HD4850 beats GT200 in theoretical shader power in Gflops.
The HD4870 with 1.2TFlops
The HD4870X2 ith around 2,4TFlops run's by it a factor 2,5.

It's bad that iNtel killed HavokFX and nV took over PhysX.
Havok FX would be aTI/nV neutral. Unless PhysX is more full fledge like gameplay PhysX out of the box.

It,s possible ATI would rule with GPU Physics. They have more shader avaible to do something else with them.

So the next review would be bechin ATI vs nV vs PPU with the hacked drivers for ATI.

Also CPU scaling vs GPU scaling vs PPU vs dedicate nV PhysX.

And of course there is yet agian benched pure on FPS.
I'am more interrested of what brings this lot more PhysX to the game.
UT3 PhysX maps where design with what a PPU could handle.

Like the said. Enhancing games with heavy PhysX at the cost of a bit FPS.

So I wonder to see a UT3 PhysX mod putting for a change a second GT280 to full physX use. And later a dedicated HD4870X2 with the hacked PhysX drivers.

 

[Dual]

In case anyone is interested I installed my GTX280 last night with drivers that support physx on the GPU.

I played all 3 physx maps on full resolution on my 30" monitor 2560X1600.

I didn't measure the frame rates but all of the maps ran very smoothly.

My system is Asus ROG Rampage with a q6600 oc'd to 3ghz.

 

[cybereality]

In case anyone is interested I installed my GTX280 last night with drivers that support physx on the GPU.

I played all 3 physx maps on full resolution on my 30" monitor 2560X1600.

I didn't measure the frame rates but all of the maps ran very smoothly.

That is interesting. So I guess PhysX on the GTX280 is probably better than what the PPU can do (assuming the game is not maxing out the GPU).

I'm barely getting 60FPS on the PhysX maps w/ an 8800GTS and a PPU. Very interesting indeed, but we would need more data to be sure.

 

[kumquatsrus]

In case anyone is interested I installed my GTX280 last night with drivers that support physx on the GPU.

I played all 3 physx maps on full resolution on my 30" monitor 2560X1600.

I didn't measure the frame rates but all of the maps ran very smoothly.

My system is Asus ROG Rampage with a q6600 oc'd to 3ghz.

yeah, can you run fraps and let us know your average fps on each of those maps?

That is interesting. So I guess PhysX on the GTX280 is probably better than what the PPU can do (assuming the game is not maxing out the GPU).

I'm barely getting 60FPS on the PhysX maps w/ an 8800GTS and a PPU. Very interesting indeed, but we would need more data to be sure.

what resolution are you running at?

 

[cybereality]

what resolution are you running at?

Running at 1360 x 768 with my sig-rig, Vista 32-bit, PhysX driver 2.8.1.

On regular UT3 levels I am getting upwards of 100FPS so my PC clearly has some horse-power to spare. On the PhysX maps I am getting much lowered speeds. This is expected, but much slower than I thought. For example, on Heat-Ray I can get around 60FPS avg but on Lighthouse I will be lucky to get a smooth 30FPS (it varies wildly from around 15FPS - 90FPS). Tornado is playable, but at around 45FPS avg. Judging from some of the published benchmarks I was expecting better performance. Honestly I am not all that impressed with those levels to begin with, nothing I've seen that can't be done better in software. So I'm not sure what to think. But the PPU certainly helps with the regular levels so I am happy about that much.

I think this might also just be a matter of badly optimized level design. I've seen demos which seem to showcase the PPU, it does have some power, but those UT3 PhysX levels are not the best example IMO. I'm going to run some benchmarks to verify this, but I was waiting for Nvidia to officially support GPU PhysX on the 8800 GTS before I do so.

 

[Dual]

Here's an update. I loaded fraps and tested 2 configurations. 8800 GTX with a BFG physx card and
the GTX280 usiing the physx drivers.

On a 24" monitor the bfg physx card was getting 50-60 FPS on the tornado map. The GTX 280 was getting 35-40. So even though the frame rate seemed ok before now running fraps gives a better picture.

Also this GTX 280 has a speed stepping feature to save noise and heat. But twice now it has locked into the lower speed of 300 mhz playing ut3 making the game unplayable. Frame rates of about 10-12. And there's no way to turn off the speed stepping so I had to reboot the computer.

 

[vicaphit]

I still personally think PPUs are worthless. Think about this. Every game out there had NO sound what-so-ever, then sound cards are released, but only 3 games allowed you to play sounds. You still don't get any sound out of the other games, so why buy a sound card? Why would I buy a PPU for games that I am not going to play anyway?

 

[kumquatsrus]

Here's an update. I loaded fraps and tested 2 configurations. 8800 GTX with a BFG physx card and
the GTX280 usiing the physx drivers.

On a 24" monitor the bfg physx card was getting 50-60 FPS on the tornado map. The GTX 280 was getting 35-40. So even though the frame rate seemed ok before now running fraps gives a better picture.

Also this GTX 280 has a speed stepping feature to save noise and heat. But twice now it has locked into the lower speed of 300 mhz playing ut3 making the game unplayable. Frame rates of about 10-12. And there's no way to turn off the speed stepping so I had to reboot the computer.

dual, those are some interesting results. seems comparable to techgage's running physx on their 9800gtx vs. ppu, except they ran the test on heatray w/physx. which means that hardware physx in ut3 seems to be quite taxing to the 280 at higher resolutions. so now i'm curious as to how a second 280 in sli might compare to a ppu with the same test, without it having to deal with graphics rendering. guess i'll have to wait awhile on that, i imagine.

I still personally think PPUs are worthless. Think about this. Every game out there had NO sound what-so-ever, then sound cards are released, but only 3 games allowed you to play sounds. You still don't get any sound out of the other games, so why buy a sound card? Why would I buy a PPU for games that I am not going to play anyway?

?????

are you for real?

 

[DogChainX]

I still personally think PPUs are worthless. Think about this. Every game out there had NO sound what-so-ever, then sound cards are released, but only 3 games allowed you to play sounds. You still don't get any sound out of the other games, so why buy a sound card? Why would I buy a PPU for games that I am not going to play anyway?

Logically, you make nonsense.

 

[Bdbtoys]

I still personally think PPUs are worthless. Think about this. Every game out there had NO sound what-so-ever, then sound cards are released, but only 3 games allowed you to play sounds. You still don't get any sound out of the other games, so why buy a sound card? Why would I buy a PPU for games that I am not going to play anyway?

So you're saying you would perfer to play games without sound... right?

[sarcasim]
Wow!

And to think there was a time when computers had no sound. I wonder how come we even needed this thing called sound on a computer to begin with.
[/sarcasim]


Actually, you're analogy to a sound card is suiting. We went from no sound, to beeps and blips from the basic of speakers, then voice (insert war-games joke), then basic hi-fi sound, to accellerated sound processing / surround sound. Physic proccessing is similar (starting at software, then hardware, now its just improving the hardware).

Each new technology has to start somewhere, some take off...some don't. I hope this one does... just like sound.

 

[Scali2]

This article shows different results:
http://www.firingsquad.com/hardware/physx_performance_update/

Their PPU never gets over 30 fps, and their CPU is not even half that.
I wonder why there's such a big discrepancy.
They both use the Heat Ray map, and the resolutions are pretty close (1680x1050 vs 1600x1200).
The PPU speed shouldn't be different, they both use a 9800GTX card, and both use a QX9650 CPU.
All I can think of is that Firingsquad ran the game at much higher detail settings or something.

 

[Balthazor]

Personally I tested the Nurien demo using 2 x 8800 GTX in SLI doing the physx, and using the 8800s for graphics only and an Aegia PPU doing the physx.

The 8800 GTX SLI doing the physx was hands down better - FPS was 20-30 in the demo with it. The Aegia PPU doing the physx limited the demo to 10-20 FPS.

 

[kumquatsrus]

Personally I tested the Nurien demo using 2 x 8800 GTX in SLI doing the physx, and using the 8800s for graphics only and an Aegia PPU doing the physx.

The 8800 GTX SLI doing the physx was hands down better - FPS was 20-30 in the demo with it. The Aegia PPU doing the physx limited the demo to 10-20 FPS.

i must concur with balthazor's results. i have an 8800gt sli setup and an asus p1 card that i have been testing over the past week. overall, the ppu can't quite keep up with a single 8800gt doing physx. the powerpack stuff seems kinda buggy and unstable, but my results are still pretty comprehensive since I ran the games/ tests numerous times. just the fluid demo itself, which anyone could hardly consider to be remotely graphic intensive, had the 8800gt beating the ppu by roughly 50%. that's not to downplay the value of the ppu to those who don't have access to gpu physx, or don't have the option due to motherboard limitations. it still does a fairly good job in actual physx games currently available. it's just not gonna best a 9600gt/8800gt for physx, though in certain games, i'm sure it would run on par with those. however, i imagine it wouldn't have too much difficulty beating an 8600gt. of course, with my testing, the cpu doesn't even stand a chance. so basically, if you have a ppu with an older non-cuda nvidia gpu, slow cuda-capable nvidia gpu, or an ati gpu, then it still is viable for hardware physx for awhile longer. anyone with a 9600gt or higher need not be concerned with the ppu, unless perhaps you are playing at 4mp resolutions where the gpu may be taxed too severly to handle heavy graphics loads along with physx. that's probably where something like a gtx280 might come in handy. i will try and post my results for the powerpack and other games (between a ppu vs. a single 8800gt vs. 8800gt sli physx vs. 2nd 8800gt as ppu) when i am finished with all of my testing if it is requested here.

 

[Balthazor]

Kumquat: I didn't test non-SLI, just SLI physx vs. PPU. If you are also testing with SLI disabled, I'd be curious to see your overall results.

Nvidia says that SLI physx should just distribute the load evenly between the graphics cards, but I wonder what the real world effect is.

 

[kumquatsrus]

Kumquat: I didn't test non-SLI, just SLI physx vs. PPU. If you are also testing with SLI disabled, I'd be curious to see your overall results.

Nvidia says that SLI physx should just distribute the load evenly between the graphics cards, but I wonder what the real world effect is.

balthazor,

i will try to post results later tonight when i have a chance. been trying to solve a bsod problem i've been having for awhile. just to give an idea though, it does appear that having a 2nd gpu for physx in non-sli performs better than sli with a split physx load, but the performance difference varies quite a bit from a nominal amount to a significant one. i think it may have to do with a combination of heavier graphics loads along with physx that may tax the sli setup moreso than the non-sli setup, but i'm not certain. it could also could be that the drivers are not quite optimized for sli physx, at least not to the point where it matches a non-sli setup with a dedicated gpu for physx. again, i will try to post my numbers later.

 

[BoudahXL]

Kumquat: I didn't test non-SLI, just SLI physx vs. PPU. If you are also testing with SLI disabled, I'd be curious to see your overall results.

Nvidia says that SLI physx should just distribute the load evenly between the graphics cards, but I wonder what the real world effect is.

Nurien Demo 8800GTX SLI and non-SLi

8800GTX+8800GTX(Physx)
45fps
8800GTX(SLi)(PhysX)
43fps
8800GTX(SLi)+PPU(Physx)
38fps
8800GTX+PPU(PhysX)
39fps

Ermmm.... I think there are just too many variable in the equation, doe sthe game(demo) support SLi, settings, bus, motherboard, OS, etc...

 

[Balthazor]

Nurien Demo 8800GTX SLI and non-SLi

8800GTX+8800GTX(Physx)
45fps
8800GTX(SLi)(PhysX)
43fps
8800GTX(SLi)+PPU(Physx)
38fps
8800GTX+PPU(PhysX)
39fps

Ermmm.... I think there are just too many variable in the equation, doe sthe game(demo) support SLi, settings, bus, motherboard, OS, etc...

Thanks for posting your results. You have the same graphics card as me, but are seeing greatly improved framerates... I'm guessing my CPU is bottlenecking me. Nvidia claims to use the CPU in addition to the selected physx card for physx, so this makes sense.

Still your results are showing that the PPU is bottlenecking your performance, which is interesting... and an argument for just leaving the damn thing out of my system.

 

[Scali2]

Still your results are showing that the PPU is bottlenecking your performance, which is interesting... and an argument for just leaving the damn thing out of my system.

That's not too much of a surprise really...
The physics test in 3DMark Vantage is a pretty good synthetic PhysX performance test.
If you compare the scores of a quadcore CPU, PPU and GPU, you'll see something like the following:
CPU - 8-16 ops
PPU - ~28 ops
GPU - 80-170 ops

So basically the PPU is only about twice as fast as current quadcore CPUs. And we already know that quadcore CPUs limit performance in games with heavy physics (they have trouble playing the UT3 PhysX levels for example).
So the PPU is not *that* far off of the CPU-limit. It was to be expected that you'd run into its limits quickly.

GPUs however are much, MUCH faster at processing PhysX. Currently it seems to be massive overkill to use a high-end or even mid-end (9800GTX-ish) GPU for PhysX alone.

The thing is, if you have a card that does about 170 ops, that means that about 10% of its performance equals a quadcore CPU in terms of PhysX. So theoretically, using that single GPU to perform the physics load of a CPU will give about 10% drop in framerate.
The PPU is less than 20% of the GPU, so you get less than 20% drop in framerate.
The flipside is that once you run into the limits of a CPU or PPU, the framerate will drop VERY quickly. Therefore GPU PhysX scales better than either CPU or PPU, even when using only a single GPU.

A single 8800GTX might be *just* below the 39 fps mark of the PPU when you use it for both graphics and PhysX.
But a single GTX260 or GTX280 is probably faster with GPU PhysX than with a PPU, just like the dual 8800GTX are completely PPU-limited, and are better off processing the PhysX workload themselves.

So indeed, why put a PPU in your system?

 

[Balthazor]

It would appear that there really isn't a reason to consider PPUs at this point; its better to just get another video card.

If that weren't possible due to lack of PCI-E slots than a PCI PPU could be considered, but otherwise it doesn't make sense.

 

[Scali2]

If that weren't possible due to lack of PCI-E slots than a PCI PPU could be considered, but otherwise it doesn't make sense.

Even that is doubtful.
If you already have a fast GPU, the PPU might actually slow you down.
Otherwise, perhaps it's better to invest the money in upgrading your GPU rather than spending it on a PPU.

 

[BoudahXL]

Well imho it's still kinda unfair to compare GPU vs PPU, there are so many factors, like I said.Maybe in 2-3 months once the drivers have matured and we get 2-3 solid games, then we could produce valuable benchies...

 

[Bdbtoys]

Well imho it's still kinda unfair to compare GPU vs PPU, there are so many factors, like I said.Maybe in 2-3 months once the drivers have matured and we get 2-3 solid games, then we could produce valuable benchies...

I think its safe...

We have the GPU running just as fast if not faster than the "mature" PPU on games that have been using the PPU for some time.

I did some testing on my 8800GTX with and without the PhysX card enabled and the 8800GTX seems to do a better job by itself. There were only a few cases where it appeared the PPU was starting to pull ahead barely (you can choke the 8800 by putting your video settings way up), but in general the single 8800GTX running GPU+PPU was very impressive. So much so that I have to eat my words on a previous postings when I speculated it couldn't be done (GPU+PPU) one one card and having come close to the performance of having a dedicated PPU card.

 

[Scali2]

In essence it all comes down to the games produced. For each game, the developer has to decide on the level of complexity they want to use for the physics in the game.
If they adjust the complexity so that the PPU won't get limited, PPUs would still be fine. If they go over that threshold, then only GPUs will be able to handle the game gracefully.

My guess is that future games will come with 2 or 3 levels of physics:
1) Simple CPU physics
2) High-end quadcore CPU physics, or simple GPU/PPU
3) High-end GPU physics (which would currently be GeForce-only).