GTX 780 just for 3D, GT 640 for PhysX?

M

Mr. Stryker

Supreme [H]ardness
Joined
Mar 15, 2005
Messages
5,221
As the thread title says, can I do this? My backup card is a Galaxy GT 640 that I'm currently using right now while the 780 gets here.
 
yes, you'll be able to set the gt 640 as physics only.

make sure to run tests to see if the 780 even needs the gt 640 though, sometimes you'll see the greater card getting reduced performance by using an underpowered card for physics.
 
I'm using a GT 640 for PhysX with my GTX 680 Lightning. They're working great together. The 640 will do just as well with a 780, if its even necessary.
 
Yeah, that's one of the reasons why I made this thread, I was hearing things like a slower card dedicated to just PhysX would be hindering some of the performance on the card doing just the 3D. It didn't make sense to me...
 
I've got a GTX 650 for PhysX and a GTX 780 on the way for my primary card. I can run some tests and see how much the 650 helps or hinders performance to give you an idea of what to expect.

What are the best PhysX benchmarks to run?
 
What games that support Physx do you plan on playing? I can tell you from experience that a game like Batman AA & AC will work smoothly with a single GTX 660, but a game like Planetside 2 (most stressful game for Physx that i know of) lagged like crazy for me with Physx on. I tried adding a GT 640 to do handle Physx to relieve the stress on the GTX 660 but it still lagged for me. So i just purchased another GTX 660 to use SLI on my system and that smoothed out my performance issues with that game specifically (my other games benefit greatly as well from SLI Tomb Raider, Arma 3 Alpha Borderlands 2 and most recently released Metro Last Light). Anyway give the GT 640 a try with different Physx titles and use the GTX 780 by itself to see the results you get. Warframe and Planetside 2 support Physx and its F2P. I wouldn't trust the results you get from benchmarks as to me they don't represent actual gameplay experience.
 
gladesmellzgood said:
What's the point of having a dedicated PhysX card? And how recent would it have to be?
Click to expand...
Wondering the same. Don't play Planetside 2 but with a single GTX 660 on Borderlands 2 I cranked up PhysX to high and all the detail up and everything's been fine.
 
having a dedicated physx card takes the load off the primary card. however if the dedicated physx card can't keep up with the primary 3d card, the primary 3d card will be stuck waiting for the physx to "catch up".
 
bigted said:
having a dedicated physx card takes the load off the primary card. however if the dedicated physx card can't keep up with the primary 3d card, the primary 3d card will be stuck waiting for the physx to "catch up".
Click to expand...

That sounds way too complicated lol. Why not just go sli? Wouldn't that be better for performance overall?
 
Oh ok, let's just spend another $650 to have good PhysX performance if I'm fine with the 3D performance of 1 card.

I just want to see if it was reasonable to use the GT 640 that I already have, for PhysX. But I'm doing that already, and it seems to be OK so far, haven't seen any insane slowdowns yet.
 
You should really test with the 640 both enabled and disabled. Seems the 640 is less than 20% the performance capability of the 780, so depending on how much PhysX actually takes out of the 780 the 640 could very well be making things slower than it would without. Maybe not insane slow downs but slower than you could get with a single card (and that much less power draw)
 
Its a waste of space, the gtx780 alone is fine. Physx is under utilized anyway.
 
I would be looking at a 650 or a 650ti

640 is okay for a 680

but the 680sli guys use 650/650ti

780 should be between a 680 and a 680sli.
 
Do you own Metro: Last Light? I am running a 650 Ti with my GTX 780, and at 2560 x 1600 I gain an average 5 FPS with the dedicated card using the benchmark the game provides. The highest utilization I see in that benchmark is 30% (with a few spikes to 38%). Since a 640 has half the cores the 650 Ti does, it should still stay under 100% utilization in the worst of scenarios. I'm not sure what impact the memory bandwidth difference would have.

I ran each benchmark twice, so here are the averages of the two runs:

EVGA GTX 780 SC w/ 650 Ti SSC:
•Average Framerate: 45.00
•Max. Framerate: 95.28
•Min. Framerate: 19.46

EVGA GTX 780 SC
•Average Framerate: 40.00
•Max. Framerate: 94.12
•Min. Framerate: 18.89

The max and minimums don't appear to change at all, but the frames seem to come in a more linear fashion. You can see it in the image below.

metrollphysx.jpg
 
Wow awesome. Now I feel much better lol. Thank you very much for doing this...
 
The 650Ti is a much much more powerful card than the GT640, you really need to test the results with the 640 active and deactive
 
Yeah, true.. *sighs*

I'm not good at doing these benchmarks like what he did.
 
Buy a copy of Metro: LL for $20 bucks in the FS forum here. The benchmark will automatically make the graphs for you! All you have to do is select which card does PhysX in the Nvidia control panel between runs.
 
jolli said:
*snip*
Click to expand...

I see you have a fairly similar setup to mine. Unfortunately, I don't have Metro: LL. Any chance you could fully disable your PhysX card so you can run your GTX 780 at PCI-E x16 instead of x8 to see how much of a difference that makes? I'm curious to see if cards are starting to get to performance levels that are saturating PCI-E 2.0 x16.
 
Sure. I have a backplate coming in tomorrow for the 780 so I have to take it apart then anyways. Should have something by tomorrow evening. Utilization in the benchmark is nearly always 99%, so I am not expecting much.
 
bigted said:
having a dedicated physx card takes the load off the primary card. however if the dedicated physx card can't keep up with the primary 3d card, the primary 3d card will be stuck waiting for the physx to "catch up".
Click to expand...

Not necessarily. If the game has separate threads for physx the rest of the game will not get slowed down by your physx card running slow. Your physx can be running at 10fps while the rest of the game is running at 60 fps.

I think most games handle physx this way. So if you run physx on a separate card you should see about the same performance as if you had physx turned off. Typically there are some extra graphics displayed when you have physx on though so it wouldn't be exactly the same.

For example in Planetside 2. The gravity lifts use physx. I have a GTX 460 running the graphics and a 9800 GT running Physx. I can tell the physx is running at a lower frame rate by looking at the particles in the gravity lifts. They move a choppier than everything else, probably around 20 fps, while everything else in the game is running pretty smooth at 60 fps. With Physx off I would maybe get a couple extra fps because instead of displaying the thousands of particles in the lift it would just display the much simpler tube graphic.
 
Then it may have something to do with how the game was designed to utilize PhysX. I already beat Metro LL on a GTX 660 Ti and it ran great at 2560x1440 on Med/High settings and never saw any serious slowdowns. I'm re-playing the game in Ranger mode, but not playing it much now (back to my daily BF3/WoW routine after beating two singleplayer games, one of them being Bioshock: Infinitey [wow great game as well]) but I can play Planetside 2 to see the difference. Do I need to download FRAPS? I do have a licensed copy that I bought years ago, hopefully it's still there.
 
Mr. Stryker said:
Then it may have something to do with how the game was designed to utilize PhysX. I already beat Metro LL on a GTX 660 Ti and it ran great at 2560x1440 on Med/High settings and never saw any serious slowdowns. I'm re-playing the game in Ranger mode, but not playing it much now (back to my daily BF3/WoW routine after beating two singleplayer games, one of them being Bioshock: Infinitey [wow great game as well]) but I can play Planetside 2 to see the difference. Do I need to download FRAPS? I do have a licensed copy that I bought years ago, hopefully it's still there.
Click to expand...

Planetside 2 has an option to display the frame rate in game, you just press alt-f. It will also tell you if you are CPU or GPU bottlenecked which is nice.
 
Parja said:
I see you have a fairly similar setup to mine. Unfortunately, I don't have Metro: LL. Any chance you could fully disable your PhysX card so you can run your GTX 780 at PCI-E x16 instead of x8 to see how much of a difference that makes? I'm curious to see if cards are starting to get to performance levels that are saturating PCI-E 2.0 x16.
Click to expand...

Finally had a chance to run the card in x8 and x16 (2.0 of course). There is indeed a performance benefit to be had by running at x16. Makes me wish I had a 3.0 capable processor and motherboard.

Here is the test setup I used:

Computer in signature, GTX 780 (1137MHz/6010MHz), GTX 650 Ti (1070Mhz/5400MHz), Metro: LL benchmark (Options: Resolution: 2560 x 1600; DirectX: DirectX 11; Quality: Very High; Texture filtering: AF 16X; Advanced PhysX: Enabled; Tesselation: Very High; Motion Blur: Normal; SSAA: OFF; )

GTX 780 @ x8 2.0:
•Average Framerate: 39.33
•Max. Framerate: 94.02
•Min. Framerate: 18.52

GTX 780 @ x16 2.0:
•Average Framerate: 40.33
•Max. Framerate: 110.56
•Min. Framerate: 18.93

GTX 780 + 650 Ti @ x8,x8 2.0:
•Average Framerate: 45.00
•Max. Framerate: 103.15
•Min. Framerate: 19.21

I ran each test three times in a row to get the averages. I lowered the overclock on my GTX 780 so that it throttles to 1137 instead of the 1167 I had before to add a bit of stability before I ran any of the tests.
 
Thanks for the results. Interesting to note that the low end framerate doesn't seem to be affected that much, so I'd guess it wouldn't really affect the observed "smoothness". It might cap the high end framerate a bit, but anything over 60 FPS is irrelevant for a 60Hz refresh anyway. 1 FPS (2.5%) average framerate drop for going from x16 to x8 isn't too bad.
 
You must log in or register to reply here.
Back
Top