Dedicated GPU Physx benchmark thread.

[Keysplayr]

UT3 would be a good place to start.

 

[Keysplayr]

Updated Rig3 GRAW2 benches.

 

[Keysplayr]

Updated Rig1 UT3 benches.

 

[Qbah]

Who cares if the performance drop is 30% as long as the games have smooth framerates. Perhaps for some minor discussion on how much power of the card PhysX zaps. But, in the end, as long as everything still runs smooth, there's really no point. keysplayr2003 still shows you min and max and you can see that it's pretty nice as in not too much of a hit.

Anyway, nice info here. Though I must admit that I don't see how it is of any benefit right now, as I really don't see any major games being a PhysX only title or the whole thing extremely changing gaming experience right now. In 2-3 years? You'll buy a new GPU anyway by then and perhaps ATI will have support for it too?

 

[Keysplayr]

GTX280 + 9800GTX+ benches on the way. Running them now. GTX280 being the rendering card, and the GTX+ running Physx.

 

[ZimZum]

Originally posted by: keysplayr2003

Originally posted by: Janooo

Originally posted by: keysplayr2003

Originally posted by: BFG10K
I think what also might be useful are results without PhysX so we can see the performance hit from enabling it.

I can save you, and more to the point, me, a lot of time and just say that results without Physx would be faster. A lot less to do for the graphics card. A lot less to see and experience for the user.

Update: Just added Warmonger results for Rig1 with forced AA in NVCP. It works fine. Jaggies are gone. Performance hit (at 1280x1024 at this point) is minimal. Check it out.

Still it would be nice to know how much faster.

Who cares if it's eleventy billion fps faster? There still would be no physx. It's sort of self defeating, no?

This statement begs the question. Do you understand the concept of benchmarking? Because it doesn't seem like you do.


"Hey, GPU MAKER X just released supercalifragalisticexpealidocious Anti Aliasing, its 17 different kinds of awesome.

Cool, whats the performance hit for enabling it.?

Why do you care? Nothing without supercalifragalisticexpealidocious AA is relevant anymore as it is our new lord and saviour. Bow before it or your blood will flow in the streets with the rest of the non believers!!!.

Ummm whatever.

 

[Keysplayr]

Well said ZimZum, well said indeed. That is all. LOL.

 

[BFG10K]

Some comparisons with and without PhysX:

http://www.computerbase.de/art...cuda_physx_ueberblick/

In most situations the GTX280 running hardware physics is reduced to a slideshow even with such low settings. If you run 1920xYYYY with high levels of AA you need not apply.

 

[chizow]

Originally posted by: BFG10K
Some comparisons with and without PhysX:

http://www.computerbase.de/art...cuda_physx_ueberblick/

In most situations the GTX280 running hardware physics is reduced to a slideshow even with such low settings. If you run 1920xYYYY with high levels of AA you need not apply.

Is that the correct link? I don't even see 1920 tested with most of the games, but the others that go up to 1680 with 1xAA vs. 4xAA show "free AA" at playable framerates (aka CPU bottlenecking).

 

[Tempered81]

Originally posted by: ZimZum

This statement begs the question. Do you understand the concept of benchmarking? Because it doesn't seem like you do.


"Hey, GPU MAKER X just released supercalifragalisticexpealidocious Anti Aliasing, its 17 different kinds of awesome.

Cool, whats the performance hit for enabling it.?

Why do you care? Nothing without supercalifragalisticexpealidocious AA is relevant anymore as it is our new lord and saviour. Bow before it or your blood will flow in the streets with the rest of the non believers!!!.

Ummm whatever.

lol. Havok = m$ + intel + Amd/ati
PhysX = Nvidia + Ageia - Ageia = Nvidia

 

[Keysplayr]

Originally posted by: BFG10K
Some comparisons with and without PhysX:

http://www.computerbase.de/art...cuda_physx_ueberblick/

In most situations the GTX280 running hardware physics is reduced to a slideshow even with such low settings. If you run 1920xYYYY with high levels of AA you need not apply.

I'm with Chizow here. Can you give more specific examples? Because that's not what I am seeing in that computerbase link. Are we missing it?

 

[thilanliyan]

I think Keys and chizow are right. The "GTX280+PhysX Karte" is using a PPU for physics. I'm not really sure what the difference between "GTX 280 + QX9770 @ 4 GHz " and "Nvidia GeForce 280 GTX" is supposed to be since they're all running with a 4GHz QX9770 aren't they?

 

[Keysplayr]

So, I was able to borrow an 8600GT to use as a PhysX processor.

Rig3: Q6600 @ 2.4. 9800GTX+ & 8600GT
1920x1200 HeatRayPhysX level

9800GTX+ Render & PhysX
Min: 24 Avg: 36 Max: 44

9800GTX+ & 8600GT
Min: 29 Avg: 39 Max: 51

 

[Fox5]

Hmm, it doesn't look like PhysX (not using the GPU) benefits from going to dual to quad core, yet physx with the gpu does.
Perhaps software physx isn't optimized for multicore processors? The gains seen generally would map pretty well to the gains seen from going to single threaded to multi threaded.

 

[Elcs]

One theory that was thrown around when I think it was FiringSquad did the GP PPU vs. PPU tests was that the PhysX PPU was possibly limited by PCI bandwidth whereas a GP PPU has PCI-E bandwidth to go at.

I think it would be interesting if it was possible, to reduce the bandwidth of the PCI-E based GP PPU slot to PCI speeds and compare the results. Perhaps this is outside of the remit of this review but I do think it would be interesting to see.

 

[SergeC]

Bump for major personal interest. I wonder what level of dedicated physx card is necessary for acceptable PhysX performance? Would, say an 8500GT work? Just throwing that out there.

 

[MarcVenice]

SergeC, just look at how much the 8600gt improves performance. A single 9800gtx+ that does all the work gives 24 minimum, 36avg and 44max framerates. Add in a 8600gt, and your min fps increase with THREE, avg increases with THREE, and max increases with 7. So no, I'd say a 8500gt is completely suckage, and not worth adding next to a 9800gtx+, at least not in that particular level of UT3.

9800GTX+ Render & PhysX
Min: 24 Avg: 36 Max: 44

9800GTX+ & 8600GT
Min: 29 Avg: 39 Max: 51

 

[BFG10K]

Originally posted by: thilan29

I think Keys and chizow are right. The "GTX280+PhysX Karte" is using a PPU for physics.

I could be mistaken but I thought that was with the GTX280 rendering physics. The reason I thought that is because test setup page doesn't list a PPU anywhere.

 

[Keysplayr]

Originally posted by: MarcVenice
SergeC, just look at how much the 8600gt improves performance. A single 9800gtx+ that does all the work gives 24 minimum, 36avg and 44max framerates. Add in a 8600gt, and your min fps increase with THREE, avg increases with THREE, and max increases with 7. So no, I'd say a 8500gt is completely suckage, and not worth adding next to a 9800gtx+, at least not in that particular level of UT3.

9800GTX+ Render & PhysX
Min: 24 Avg: 36 Max: 44

9800GTX+ & 8600GT
Min: 29 Avg: 39 Max: 51

Marc, you are assuming that PhysX power is dependent on the number of shaders in the dedicated PhysX GPU. We have not tested anything less than a 8600GT yet, so you cannot have any definitive opinion on anything less than a 8600GT like the 8500GT example you gave. This weekend, I will be trying out the IGP on the 780a board to run dedicated PhysX, and a 9800GTX+ to render. Why don't you reserve judgement until then. It has been my observation, that PhysX is NOT very intensive on GPU shaders, but more testing should confirm or disprove this idea. You were looking at individual fps differences. I'm looking at percentage differences in the most important area. Minimum fps.

 

[MarcVenice]

I'm not assuming anything keys, I'm simply deducing that the 8500gt, which is a inferior card to the 8600gt will show little to no performance gains when paired with a 9800gtx+. How do I deduce that? Because the 8600gt barely showed any improvement when paired with a 9800gtx+, so how is a inferior card suddenly going to change that? From your own benchmarks you can obviously not conclude anything other then that adding a 8600gt was almost useless, the extra powerconsumption and heat aren't worth 5 minimum fps. Let alone worth buying a 8600gt.

 

[Keysplayr]

Originally posted by: MarcVenice
I'm not assuming anything keys, I'm simply deducing that the 8500gt, which is a inferior card to the 8600gt will show little to no performance gains when paired with a 9800gtx+. How do I deduce that? Because the 8600gt barely showed any improvement when paired with a 9800gtx+, so how is a inferior card suddenly going to change that? From your own benchmarks you can obviously not conclude anything other then that adding a 8600gt was almost useless, the extra powerconsumption and heat aren't worth 5 minimum fps. Let alone worth buying a 8600gt.

The thing is, you don't know (nor do I) what proportion of shader power was utilized in the 8600GT which has 32 shader processors. Who is to say that even 8 sp's (which I think is what the IGP on the 780a has) is not sufficient? You can deduce all you want, but as of yet that has no foundation until tested. Now, when I test, i will happily share my findings with you, and only then should we make an educated decision on what is what. But alas, it will have to wait until the weekend, unless of course you can test it before I.

 

[MarcVenice]

Who cares what proportion was used in the 8600gt Keys, the performance increase SUCKED. It will not be worth it to buy a 8500gt or 8600gt, if you allready have a 9800gtx+, which gave 'satisfactory' FPS all by itself. The only interesting part could be the IGP, if it gives the same boost as a 8600gt, then it could be a nice boost for people who own mobo's with a IGP from nvidia. But I doubt many if any people will care, what enthusiast runs a AM2 mobo with a Phenom CPU and a high-end graphics card. That's besides the point though.

Because the 8600gt performance increase was not satisfactory, I deduced that the performance increase of any card inferior to a 8600gt will not be satisfactory either, and will thus not be worth spending any money on to complement a 9800gtx+. How hard is that to understand? You're benchmarking all by itself might be interesting, but only for people who either allready OWN a 8600gt or lesser card and don't want to put it in the trashcan, or have a IGP from nvidia. That's all I'm saying.

 

[Keysplayr]

Originally posted by: MarcVenice
Who cares what proportion was used in the 8600gt Keys, the performance increase SUCKED. It will not be worth it to buy a 8500gt or 8600gt, if you allready have a 9800gtx+, which gave 'satisfactory' FPS all by itself. The only interesting part could be the IGP, if it gives the same boost as a 8600gt, then it could be a nice boost for people who own mobo's with a IGP from nvidia. But I doubt many if any people will care, what enthusiast runs a AM2 mobo with a Phenom CPU and a high-end graphics card. That's besides the point though.

Because the 8600gt performance increase was not satisfactory, I deduced that the performance increase of any card inferior to a 8600gt will not be satisfactory either, and will thus not be worth spending any money on to complement a 9800gtx+. How hard is that to understand? You're benchmarking all by itself might be interesting, but only for people who either allready OWN a 8600gt or lesser card and don't want to put it in the trashcan, or have a IGP from nvidia. That's all I'm saying.

And yet, this was only one game, at one resolution. Who in their right mind (gamers and game enthusiasts and game reviewers) would base any final decision on that? This was just a quick bench.
Just relax, sit back, and see what unfolds. In the end you may be right, or you may not.

 

[Elfear]

Originally posted by: keysplayr2003

The whole point of this thread is too RUN Physx. What's the point of NOT running it? Of course it's going to be faster when you are not running it. As far as playability goes, so far, everything I have run is very playable as you can see by the benches run so far, even on the lower end system with AA and AF. So, what is it that you want to see? Or in this case, not see? As in, you won't see any physics, as is part of the point of this thread.
Just know, that if I ran the same benches without Physx, it would be faster. That's like comparing two runs of a bench with and without eye candy turned up. The one without the eye candy will be faster. And probably ugly.

I can't speak for the others asking for before and after benchmarks, but for myself it may be useful at determining the hit that Physx takes which would hopefully help at determining performance at higher resolutions. Seems like most guys running mid- to high-end video cards will be running at greater than 1280x1024. I'd like to know what kind of hit will occur at 1920x1200 or 2560x1600. Benching with and without Physx would give some indication of the fps penalty which MAY be able to be extrapolated for higher resolutions. Of course benching with Physx at higher resolutions would be better but I understand you may not have the equipment or time to do that.

Thanks for the testing you have done Keys. Physx looks intriguing.

 

[Keysplayr]

Thanks Elfear.

Marc, here is another quick one for you.

GRAW2
1920x1200
edge smoothing AA and 8xAF

9800GTX+ alone = Min: 27 Avg: 53 Max: 77

9800GTX+ & 8600GT = Min: 32 Avg: 60 Max: 86

I mean, it is only FIVE fps increase in minimum. But it is a 20% minimum framerate improvement.