sli 2 graphic cards and dedicate one to physx

[HappyBuffalo]

Hello, I currently have a system that has 3 gtx 980s. I was thinking about upgrading. I would like to keep a graphic card dedicated to PhysX but I don't wanna use a super high end one to do it. Is it possible to sli two new graphic cards and keep a 980 (not sli) just to dedicate to physx/processor? 3 cards, the two sli ones will be the same but the one dedicated to PhysX will be different. Just wanting to know if this is possible. Thanks!

 

[_JakeDavies]

I'm 99.99% sure that this is possible.

 

[iyzik]

Yes, it is. I used to run crossfire 7970's with a GTX 550Ti dedicated to PhysX. If I could run crossfire AMD's with a dedicated NVIDIA for PhysX...I'm sure you could do it with SLI NVIDIA's.

 

[photonboy]

1) Doing so can actually cause an FPS drop. It depends on how much PhysX processing is done relative to the main game load.

(if the main thread waits for the PhysX to finish then the PhysX card is being a bottleneck)

You can TEST this by finding a repeatable scenario and force it on to the dedicated GTX980 then leave it set to the SLI 1080's or whatever you have and compare FPS.

2) Even if it helps the benefit may not warrant the extra heat and noise put into the system, especially when there may not be many that benefit (vs 2xGTX1080 for example).

3) For games with no SLI support that does have PhysX (like Batman AK) you'd want PhysX on the second GTX1080 (if you have two).

If you gained 5% FPS it also may not matter much if you have a high FPS or already exceed your FPS cap.

4) Avoid 3xSLI. In fact, NVidia doesn't support beyond 2xSLI for the GTX1000 series. I agree with this. It's not really their fault as the game engines have been changing and don't always support SLI very well.

Once DX12 and game engines like Unreal 4 Engine mature we'll get SFR (Split Frame Rendering) but that's a long ways off aside to be common.

5) Flat out avoid this for future DX12 games that have PhysX likely. I doubt asynch compute (dynamic load balancing for GTX1000) would work properly. It could cause added stutter (lack of async compute hardware + software is a source of game stuttering if you weren't aware... game devs have to choose the processing length which if too long means stopped processing of code but if too short is wasted GPU cycles.. asynch methods can minimize or negate this issue. again dynamic load balance but also preemption etc).