GTX460 768mb = Total PhysX Overkill

[Zarathustra[H]]

Hey all,

Just thought I'd share my experience in case anyone is considering doing something similar.

When I decommissioned an older rig it had a 768Mb GTX460 in it, so I figured I'd pop it in my desktop alongside my GTX680 to use for dedicated physX.

Well, I can inform you that it is complete and total overkill. If you go the dedicated PhysX route, go with something much lower end.

I don't have many PhysX games, but I have tested it with Red Orchestra II and Metro 2033.

In both games, hardware PhysX is enabled and working, but the load on the GTX560 768mb is so low that its rounded down to 0% according to Rivatuner based charts (like those in MSI Afterburner and EVGA PrecisionX)

Then I tried FluidMark. With everything PhysX related maxed, and vsync disabled, when I started hitting a few hundred FPS on the GTX680, I finally started registering a few percent of load from the PhysX on the GTX460.

So to summarize. Unless the state of PhysX changes, pretty much any semi-modern GPU will be WAY overkill for PhysX.

If you just have one kicking around like I did, there's no reason not to use it (except maybe power consumption) but if you are buying one expressly for the task, go as cheap as you can, as otherwise you'll be wasting your money.

 

[sk3tch]

Good info. I came to a similar conclusion with a GTX 580 SC (which I knew would be total overkill) - the GT 640 I bought did just as well as it did. More info here.

 

[jolli]

I would encourage you to try Borderlands 2 and Batman: AC to see that a significant load may indeed be placed on a dedicated card.

I'm going to steal a quote from user jamsomito in the PhysX in Borderlands 2 thread to illustrate what I mean:

Ok, update. So the 9600 GSO as dedicated PhysX is definitely better than the CPU - very noticeable difference. However, as I was playing with some buddies tonight, I learned that what I considered "intense firefights" were not exactly what I had imagined flying solo. We were in Bloodshot Stronghold, the area with the swirling water vortex in the middle of the floor... there were water physics, tons of particles from everyone shooting everywhere, the siren's phase lock sucking in particles, singularity grenades sucking in more particles and swirling them around, and the blood and guts physics too.

Bottom line, my FPS hit 11 for a good chunk of time... definitely not playable in my book. *sigh* these are the times you need good performance the most, so to me, it's not worth keeping this card in. It does look very pretty though, so I envy you who can run medium or high PhysX. Guess it's back to low for me

Now I have to admit that the 9600 GSO is far slower than any GTX 460. However, I also had a GTX 460 as a dedicated PhysX card next to my GTX 680, and I experienced the same sort of massive FPS drop (prolonged periods at 15-20 fps) that the above quote describes.

It seems that a dedicated PhysX card is great until suddenly it is not. Particularly in coop BL2, you run into situations where the dedicated card hits a 'wall' and then the main graphics card is waiting on it to finish calculations. For me this manifested as ~75% useage on the dedicated card, and sub ~40% useage on my main card at the forementioned 15 fps. In that situation, I would have likely been far better off with no dedicated PhysX card dragging my main GPU down.

I say this because of the following oversimplified math:

GTX 460 (that I have) is: 336 cores X 1450MHz (shader clock) = 487200
GTX 680 (that I have) at 40% load is: (1536 cores * 0.6) x 1172MHz = 1080115.2

Again in my simplified example, when my GTX 680 was bottlenecked to 40% by my PhysX card, I had essentially 921 cores doing nothing. This is far more powerful an amount than any GTX 460.

Does my resolution of 2560 x 1600 factor into it? Maybe.

It just seems to me that many of the tests that are run are under best case conditions (short fights, solo) where the dedicated card isn't being pushed to the 'wall' that so seemingly causes ruinous performance. Try it with some buddies using elemental weapons in an unfavorable environment.

TLDR: I don't think going cheap as possible is a good idea when you have a $500 dollar main card.

 

[sk3tch]

TLDR: I don't think going cheap as possible is a good idea when you have a $500 dollar main card.

I don't think you can come to the conclusion. I did the work - the GT 640 is $100 and works fantastic as a dedicated PhysX card. Those same massive drops occurred with a GTX 580 SC being used as a dedicated PhysX processor. There is no way a GTX 580 SC is bottlenecking a GTX 680 when it's only functioning as a PhysX processor.

What's more likely is it is a software/driver/general performance issue - not unique to what PhysX card you have.

 

[jolli]

I don't think you can come to the conclusion. I did the work - the GT 640 is $100 and works fantastic as a dedicated PhysX card. Those same massive drops occurred with a GTX 580 SC being used as a dedicated PhysX processor. There is no way a GTX 580 SC is bottlenecking a GTX 680 when it's only functioning as a PhysX processor.

What's more likely is it is a software/driver/general performance issue - not unique to what PhysX card you have.

I would agree that it could be a software issue that is responsible for some degree of that slowdown we see with PhysX. I completely respect all the work you put into testing the various cards. I myself need to log some results when I get the chance so I can contribute to the discussion more.

My only critique is that you only tested by yourself. I've never run into a situation playing solo that pushed my dedicated card (at the time a GTX 460) to a point where it became a bottleneck. I did push my card in coop to a point where it became the limiting factor.

Environmental PhysX in BL2 is not the most demanding aspect of the technology. You can see that by standing next to a water pipe or an area with lots of cloth. The greater loads are a result of player action such as explosions, particles generated from shooting, liquid piles from corrosive weapons, etc.

When you are playing alone, your ability to generate these effects is extremely limited. Add another person and you double it. The third triples your potential, and so on. I have found that the dedicated card works great until it reaches a point where its potential is reached, then the rest of the system waits for it. With a GTX 640 or GTX 460 you just can't reach that potential when playing alone, and that is why it makes sense a GTX 580 didn't give you better performance. The GTX 640 had never encountered a situation where it ran out of power.

You could very well be right in that the implementation of PhysX is software limited to X amount of effects and a hundred dollar card meets that requirement. However, until we have tested under the most demanding of scenarios, not the lightest, I can't agree.

 

[sk3tch]

I think you're definitely onto something - but whether it is what you think it is or not, I'm not so sure. What helps me recognize that there is significant benefit is that my FPS went up 20%. That's no slouch number. So obviously the card dedicated to PhysX is doing something - and it's enough to take a 1250 MHz GTX 680 up 20% in frames - well, that's pretty damn good.