SLI vs Dedicated Physx GTX 560 Ti

[Skoty27]

So basically what the title says...

I'm wondering if it's better to run the 2 GTX 560 Ti's in SLI or set one up for Dedicated Phsyx

I have noticed that my cards run cooler when its setup for physx, and I'm using the EVGA Precision X program to delegate fan speeds at certain temps .....

my current system specs:

i7-930
24 gigs ram
2 EVGA GTX 560 Ti
750w PSU

Thanks in advance

 

[clutchc]

PhysX uses so little GPU power, it would be a waste to dedicate one card to that. SLI and let Nvidia set the card that handles the PhysX to its recommended default. It will probably be the 2nd card since it is used the least.

 

[TriBeard]

For sure do SLI. Let nVidia software just use whatever it wants for Phys-x. You'd be wasting a ton of horsepower with one of those cards dedicated to just phys-x. Enabling it on crossfired cards will probably give you a minimal performance hit anyway in most games.

 

[17seconds]

Even in GPU accelerated PhysX games, the performance advantage for SLI will be much greater than the performance advantage of a single card doing graphics and another only doing simple PhysX calculations. In non-GPU accelerated PhysX games, of course, there would be zero advantage for the setup with a dedicated PhysX card, while SLI will be boosting your performance ~75-95% in a well-balanced system.

 

[bystander]

There are rare exceptions, but most the time SLI far outweighs a dedicated PhysX card.

 

[thaartist]

From the tests I conducted my conclusion seems to be that Dedicated PhysX can offer a performance boost in some games.

3d Mark Pro Firestrike tests with graphic intense games in multi-player where fps is important my findings conclude that with two 560 cards it is better to use Physx on a dedicated card to get a significant boost in FPS in today's top games like COD: Ghosts, Battlefield 4, and Titanfall while needing to also running resolution lower more near 720 instead of 1080.

The combined Firestrike test I got only 3-6fps, with dedicated PhysX I was getting 8-11fps. The rest of the tests were significantly higher as well. Any other configuration was much lower in FPS.

Figured I'd post my findings using Benchmark tests.

Benchmark software used: 3d Mark Professional
Cards Tests: TWO EVGA 1gb 560

My System:
GRAPHICS:.......TWO NVidia GeForce 560 GTX
CPU:.................AMD PHENOM 2 X6 1100T
CPU COOLER:..CoolerMaster Hyper 212EVO
MOBO:.............MSI 970A-G46
RAM:................GSkill 16GB PC3-12800 (800MHz)
DRIVE 01:.........Corsair Force 3 120GB SSD (Window's only)
DRIVE 02:.........Western Digital Caviar Black 640GB 7200 RPM SATA 3.0Gb/s HHD
PSU:.................OCZ ModXStream Pro 700W
MONITOR:.........25" LCD HP2509m 3ms Response 1920x1080, 60Hz
OS:...................Windows 7 Professional 64-bit

8 tests on Firestrike mode (Highest stress test besides extreme mode) with different settings at 1080 resolution determined that it was best with dedicated 2nd card PhysX and preformed 5-10fps better which is a big deal for these cards being so outdated.

Now I kept all Firestrike settings but reduced resolution to around 720 and the fps increase was about 15fps higher except on combined tests where fps was only about 5fps higher.

When left settings in SLI mode and physx left to auto the benchmarks were almost the same. (Different CPU's may differ). I think Auto mostly uses CPU for PhysX in this mode with these cards.

If you are having gaming problems, it might be worth it to try dedicating one card for PhysX to see if that makes an improvement. Not all games are coded the same or implement the same support for SLI or PhysX so it's hard to say if there will be an improvement or not.

 

[17seconds]

Auto PhysX never uses the CPU. Firestrike does not use GPU PhysX acceleration and therefore says nothing about the performance of a dedicated PhysX card.

Using a low 720 resolution is going to highlight the effects of CPU bottlenecking and negate the advantages of dual cards in a SLI configuration. Naturally, at such a low resolution, your single card will run at just about the same speed as two in SLI, particularly when paired with an underpowered AMD CPU.

 

[bystander]

Firestrike is NOT the test to perform for this sort of experiment, and as mentioned "auto" will use PhysX on the GPU's and use SLI. Ultimately that is the setting you should use, as most game will not even use a dedicated card, and if you have it in dedicated mode, it won't be used in all but 30ish games ever made.

 

[clutchc]

I'm afraid I have to agree with 17seconds. My own testing with Firestrike in SLI did not include GTX 560 Ti's, and probably wasn't as thorough as thaartist's, but from my limited testing, dedicating one card to physX was always a waste. But I will remain open minded on it.

 

[thaartist]

Auto PhysX never uses the CPU. Firestrike does not use GPU PhysX acceleration and therefore says nothing about the performance of a dedicated PhysX card.

Using a low 720 resolution is going to highlight the effects of CPU bottlenecking and negate the advantages of dual cards in a SLI configuration. Naturally, at such a low resolution, your single card will run at just about the same speed as two in SLI, particularly when paired with an underpowered AMD CPU.

The tests concluded that when running the CPU dedicated for physx the benchmark scores were about the same as when on Auto. Hard to believe it doesn't use the CPU when on auto, when putting the 1st or 2nd card for non dedicated physx usage changes the scores a noticeably higher. If on auto it doesn't use the CPU then it's odd that the scores are so much different than selecting a non-dedicated card to handle physx.

Also, if physx isn't a factor in 3dmark benchmarking then I don't know why I got such a big improvement when I put dedicated PhysX to one of the cards. I tried every scenario possible and this was the only one that got a big boost in FPS.

However, I will agree not all programs and games will benefit from dedicated PhysX but from running 3d Mark Firestrike it would seem that graphic intense games that are big budget releases, will probably have the manpower behind it to code to the correct standards for Nvidia cards and people will see an improvement. If anything, if someone is trying to improve game play, it's probably worth a try.

I have yet to run any in game FPS, will fraps help verify my FPS with different settings or is that null and void too? I'll report the findings of in game FPS soon.

 

[thaartist]

Firestrike is NOT the test to perform for this sort of experiment, and as mentioned "auto" will use PhysX on the GPU's and use SLI. Ultimately that is the setting you should use, as most game will not even use a dedicated card, and if you have it in dedicated mode, it won't be used in all but 30ish games ever made.

What's the point for them to even make it an option then? And also can you list some of those games? Do you know if Ghost's is one? I would think so since they put a PhysX setting in the settings.

I was just trying to see if I could find a way to tell if it even made a difference having SLI or dedicated PhysX with 560's. Because games still suck with 560's in SLI. 560 cards are super outdated and pretty much worthless.

 

[17seconds]

www.physxinfo.com
For a list of games.

 

[bystander]

Firestrike is NOT the test to perform for this sort of experiment, and as mentioned "auto" will use PhysX on the GPU's and use SLI. Ultimately that is the setting you should use, as most game will not even use a dedicated card, and if you have it in dedicated mode, it won't be used in all but 30ish games ever made.

What's the point for them to even make it an option then? And also can you list some of those games? Do you know if Ghost's is one? I would think so since they put a PhysX setting in the settings.

I was just trying to see if I could find a way to tell if it even made a difference having SLI or dedicated PhysX with 560's. Because games still suck with 560's in SLI. 560 cards are super outdated and pretty much worthless.

17seconds got the right link. It seems at the moment, it is exactly 30 games. The reason I said Firestrike isn't a good example, as it will stress the physics far harder than most games will. It would be an outlier. It may also not even be better in all their tests (I'm not as familiar with 3dMark as I am with 3dmark11).

Most games will perform best leaving it in SLI with "auto" set on PhysX. There are rare exceptions, but just set it to auto will get you where you want most the time.