Yep. Just a hypothesis based on the evidence that I have. It was enough for me to make a educated guess on the right GPU to purchase to act as a PPU - so I'm happy! :)
 
Perhaps if EVGA wants to sponsor me I can run PhysX tests in more and more ways. :) But seriously, I think with big games like Borderlands 2 having PhysX we're going to finally see more info come out to help endusers get good info on what choices to make...that's a very good thing.

sk3tch

EVGATech_DaveB

Excellent write up, shed some good light on things.  PhysX implementation in games is still a bit unpredictable some times results don't make a lot of sense, but at least we see where our bang for the buck is for PhysX cards.
Going forward, as you do have the hardware, you think you will run comparisons when other PhysX heavy games come out?

Thank you.

Although I would like to be able to do that, I mostly did this research so I could determine which card I should buy for dedicated PhysX purposes. I was going to send the GT 640 and GTX 650 back based on my research. Perhaps I will reconsider...but it's tough to invest $200 into cards just being used for testing purposes. :)

 
I wasn't aware you were cashing out on those after the tests.  Figured since you had them, had the rigs etc, that you would keep them, hence my suggestion.  Id never advise you buying cards to test for everyone  :)
 
Overall some good info though, thanks for putting in the time.  Just had a thought though, I wonder how much PhysX usage scales with surround, as it would be processing for a lot more real estate.

Dedaciai

Thanks for the info, Sk3tch!  This now gives me more to chew on since I have been considering getting a 3rd 570 so I have the option of going either 3-way SLI or 2-way SLI +PhysX. However, seeing the results of the 650 and 650Ti is making me rethink things. 

 
Hey, thank you. Glad it could be of some use to you.
 
I'd definitely get a 3rd GPU for tri-SLI if I were you - no matter what, performance will increase in more situations by going SLI or +1 SLI versus adding a dedicated PhysX GPU/PPU. Don't forget, that you can always set a single 570 as dedicated PhysX whenever you want when you have 3 of them. That's nice flexibility! Dedicated PhysX + SLI when you need it...tri-SLI when you don't! Plus 570s are priced nice in the used market now.  Good luck!

sk3tch

Yep. Just a hypothesis based on the evidence that I have. It was enough for me to make a educated guess on the right GPU to purchase to act as a PPU - so I'm happy! :)

Perhaps if EVGA wants to sponsor me I can run PhysX tests in more and more ways. :) But seriously, I think with big games like Borderlands 2 having PhysX we're going to finally see more info come out to help endusers get good info on what choices to make...that's a very good thing.

 
So thanks to your testing and very informative threads on PhysX cards I decided on getting a EVGA 650ti SSC 1Gb.  I dedicated the 650ti to PhysX in the control panel to aid my 680 FTW+ 4Gb.  I ran the Batman AC benchmarks as well as did some testing with BL2 like you did and after making some small adjustments saw a solid improvement on average FPS as well as minimum FPS. 
 
The big thing I want to put out there is BEFORE I overclocked the 650ti I was seeing notably LOWER minimum's but still slightly better average frame rates than just using my single 680.  So, I went ahead and ramped up the 650ti to the max it would go stably(1175mhz core) and then tested again.  Much to my delight my results improved dramatically, not only was I getting a solid 10-15fps better(than the single 680) on the minimum but I was also averaging 10+fps higher overall with the Batman AC benchmark.  I moved back to BL2 testing and found identical results there.
 
I'm not really sure why the stock clock speeds would be holding the 650ti back so heavily but it certainly appeared to be the case.  The plus side is that it only takes 10 seconds to apply the overclock and set the fan to 25% constant(lowest setting).  The card being used for dedicated PhysX never got over 39C @ 25% fixed fan speed and is as quiet as you could ever want it to be.  Fluid mark score for the standard 1080p bench was 5585 with both cards running together which from what I can tell tops the chart on their website.   Barring any sudden change of heart I think this one will be a keeper... can't beat it for 150 dollars including a free copy of Assassin's Creed 3.

staypuft

^^ Did you crank up the voltage or is that the max OC at stock volts?

 
stock voltage, looks like there is some over-volting headroom in there after all however with my applied OC's is almost at the max anyways
 
EDIT: http://www.anandtech.com/show/6359/the-nvidia-geforce-gtx-650-ti-review/20  <-- this seems to show the same max OC across several 650 Ti's and the operating voltage at those speeds to be almost max

I think the strange Borderlands 2 results are all problems of poor coding.  It probably won't be addressed any time soon as they seem to be focused on pushing out DLC.

^^ Yeah it won't help a whole lot, I got nearly the same results in Batman AC benchmark w/PhysX using a GTX650 non-Ti as PPU.  Will likely run the card stock anyway, just wanted to see how far it would go

Definitely a solid write-up, BR well earned!
 
The PhysX benchmarks are really something I wish more reviewers did (that, and Folding@Home)!

goldfoxgt

Hey guys, quick question. So I have a solid, although aging gtx 470 at about 730-800 core and 1700-1920 RAM depending on what game im playing and what performance I need at the time. I recently got a third monitor (found it at parents house, why not use it right?), so I picked up a GT 620 on newegg on sale for 29 bucks! I didn't even consider this being a good enough card to out-due my 470 in physx, but nvidia auto selected it, and after running the 1080p preset in fluidmark, I got 23 FPS with 470 alone, and a solid 25 with the 470 + 620 (470 was at 730 core and 1770 RAM for both tests). Is it actually possible the 620 is better than the 470? Is this 2 FPS actually indicative of how it might perform in game?

Ermmm...I'd do some game tests.  Those numbers don't really seem likely at all - the 620 is considerably less powerful than the 470 in...well, everything.
 
I mean, the 470 crushes even the 640.
 
I'd say...something odd is going on, there.  Certainly, test this in some games.  Unless there is something about the settings you are using for the test (like 32x super-sampled anti-aliasing??) that is crushing the 470 in render performance so badly that it has NOTHING left for the PhysX calculation, but...

Im actually not 100% sure my logic was correct actually. Obviously at the top end I gained some FPS, but over 60 ish FPS a few more FPS really doesn't mean anything. The lowest FPS is definitely where I want to see gains, I just need to discern what a "real world" physx scenario would be. Maybe some explosions would trigger as much physx calculations as all the glass did. Im installing metro 2033 right ow to test its benchmark as well.