Another Dedicated PhysX thread

[alexisvx]

Hi all,
I have read quite a few threads on this but I thought I'd ask again as my questions are slightly different and some of the benchmarking tests done are for older cards.

I own both the GTX970 and GTX650Ti Boost (2GB). Is it worth using the 650 as a dedicated PhysX card? I have read both that it does create a significant boost for PhysX enabled games but also that the 970 is enough to handle modern PhysX games. So my first questions is, does it make a significant difference for modern games (the ones that use PhysX) and games to be introduced in the near future?

My second questions is how do I benchmark myself to a) make sure that adding the card won't worsen my performance and b) to properly test performance change? My understanding is that to test the PhysX I should run the GeForce PLA. But are there any other tests I need to make to test overall performance like 3DMark/Unigine, should I also test the CPU as I have read that using an additional card uses one of the cores?
Also do I need to test on actual games? I own all Batman games expect A.Knight. I would rather not test on games as I would need to download them. I only have Witcher 3 and Far Cry4 installed at the moment.

In summary my questions are:
1. Does using the 650Ti boost as dedicated PhysX with the 970 give a significant improvement for PhysX enabled games?
2. What benchmarking tests should I make to see overall performance change?


Thank you!


My Rig:
Windows 7
Motherboard: Asrock Z97 Extreme 4
CPU: i5 4690k 3.5Ghz O/C to 4.6Ghz
GPU: GeForce GTX 970 SSC
CPU cooler: Noctua NH-D14
PSU: EVGA 750W ATX12V
RAM: G.Skill - Ripjaws X - 2x4GB (DDR3 2133)
Case: Fractal Design Define R4

 

[17seconds]

This should be similar to your setup:

Dedicated PhyX PPU / GPU Shoot Out: GT 640, GTX 650, GTX 650 Ti, and GTX 680 – paired with a GTX 690.
http://1pcent.com/?p=169

 

[alexisvx]

Thanks for the link, I had already read that. However the post is from 2012 testing with a GTX 690 so I am not sure whether the increase would be the same with a 970. What tools would you suggest I use to test myself and make sure that it doesn't bottleneck my system?

 

[17seconds]

Just try it out with the "PhysX Visual Indicator" enabled from the Nvidia Control Panel. A GTX 690 is similar to a GTX 970, so the results should be comparable.

 

[alexisvx]

Ok, I'll try it out this weekend and report back. My main concern is whether using the 650Ti will actually make perforce worse as I have read on some threads.

 

[17seconds]

Most random thread posts tend to underestimate the benefits of a dedicated PhysX card, ignoring the benchmark evidence. The best way to know for sure is to try it for yourself. Be sure to test before and after the installation.

 

[17seconds]

Don't use 3D Mark. You could use the preset benchmark from Metro Last Light, or your own run through of Borderlands 2, Pre-Sequel, or one of the Batman games.

 

[alexisvx]

Thank you. I'll run the tests and revert next week.

 

[alexisvx]

I did a test and here are the results:

Summary: Not a significant increase in performance

Details:
2 tests PLA Game benchmark and Batman Arkham Origins benchmark
All settings at maximum

PLA:
GTX 970 on its own: Max: 62.4 - Avg: 60
GTX 970 +970 TiBoost: Max: 60 - Avg: 62.8

Batman:
GTX 970 on its own: Min: 38 - Max: 65 - Avg: 47
GTX 970 +970 TiBoost: Min: 46 - Max: 74 - Avg: 56

My Rig:
Windows 7
Motherboard: Asrock Z97 Extreme 4
CPU: i5 4690k 3.5Ghz O/C to 4.6Ghz
GPU: GeForce GTX 970 SSC
CPU cooler: Noctua NH-D14
PSU: EVGA 750W ATX12V
RAM: G.Skill - Ripjaws X - 2x4GB (DDR3 2133)
Case: Fractal Design Define R4

 

[17seconds]

That's a 19% boost in Batman for the average, not bad. Sorry, I'm blanking on what PLA is right now.

I guess the key to the dedicated PhysX card is that it's best used when you're not getting playable performance at full settings, and you need the PhysX card to make it playable or allow extra settings. Otherwise, if the game is already performing well, the PhysX card is not going to help your experience much.