Batman Arkham City - Titan + (GT 430, GTX 570, GTX 650, and GTX 770) Dedicated Physx

[Golgatha]

For my own benefit I decided to run some Batman Arkham City benchmarks using some hardware I have laying around. My main rig is in my signature. I added a Zotac GT 430 (128bit, 96 shaders, 16 TU, 700Mhz/1400Mhz) and an eVGA 570 SC (320bit, 480 shaders, 60 TU, 732Mhz/1464Mhz) to the system, to see which card I wanted to install for full time duty. My thinking was the GT 430 would be enough for Physx only and I could sell the GTX 570. Boy was I wrong!

Given as (Min, Max, Avg) FPS. Average of 2 benchmark runs. All graphics options set to their max values, resolution set to 1920x1080, and Physx set to High. I should probably do the benchmark without Vsync on to get a realistic Max value, but I was more interested in Min and Avg results because the game was unplayable at times with Physx set to high on the Titan.

Titan Alone = 16, 60, 42 FPS

Titan + GT 430 = 13, 60, 49 FPS

Titan + GTX 650 = 32, 60, 54 FPS

Titan + GTX 570 SC = 34, 60, 55 FPS

Titan + GTX 770 SC = 35, 60, 56 FPS

In practice, the jump in Min FPS when adding the GTX 570 SC turned the benchmark from looking like a slideshow at times to perfectly smooth. As an added bonus, the 570 SC will be doing Folding@home work when summertime finally passes us by. Right now it can make the room a bit too hot to leave Folding@home on 24/7 with both cards chugging away .

Given that the cards are running almost the same GPU and Processor clock speeds, I would theorize the number of shader units on the cards is making the large difference in FPS. However, some benchmarks I could find Online using a GTX 650 vs GTX 650 TI showed the 650 coming out ahead because it was clocked higher. Difference in architectures? I dunno. If anyone could shed some light for me, it would be much appreciated.

 

[Golgatha]

Tested my GTX 770 from another system just for good measure. Pretty much no change from the GTX 570 SC.

 

[Makaveli]

What resolution was this testing done at?

However, some benchmarks I could find Online using a GTX 650 vs GTX 650 TI showed the 650 coming out ahead because it was clocked higher.

I noticed this also when I was looking at getting a dedicated card for physx the GTX 650 superclock that I picked up showed a nice boost in performance from the stock GTX 650.

 

[toyota]

of course the gt430 is not enough. it may increase your average but I bet its really inconsistent and feels no smoother than just letting Titan do everything.

I would just use common sense and reduce the crazy amount of MSAA and stick with just the Titan by itself.

 

[Rvenger]

Curious to see a gtx650 in this.

 

[Makaveli]

Curious to see a gtx650 in this.

I can only provide numbers for my GTX650SC and Physx set to normal not high.

Running into an hybrid related issue when set to high game doesn't load.

Will see if I can figure it out later.

 

[Golgatha]

What resolution was this testing done at?

1080p

 

[Golgatha]

Curious to see a gtx650 in this.

I was too. They had an open box at Microcenter and I had a birthday gift card, so I decided to do some more testing. It basically has the same performance as the GTX 570. Going to keep the GTX 650, as it uses a lot less electricity for the same performance.

 

[blackened23]

I had to revert back to the 320.18 driver for acceptable physx performance in Batman: AC. Using the 320.49 driver results in very bad performance, presumably from the new physx release included with 320.49. Looking at afterburner, my cards don't boost to their max speed with the new physx release included with 320.49, and get low utilization. 320.18 never had such issues. In fact, with 320.18 and a single 780 I can use FXAA with everything else maxed, and the game never, ever dips below 60 fps. This is with Physx-HIGH at 2560 resolution. But my framerates with 320.49 aren't acceptable.

Reverting back 320.18 and it's smooth as silk. Not sure if anyone else noticed this, but there's definitely something awry with the new beta 320.49 drivers and physx games. I highly suggest reverting back to 320.18 and framerates should be a bit higher. This is without a dedicated physx card of course, I find such things to be a complete waste.

 

[stahlhart]

I had to revert back to the 320.18 driver for acceptable physx performance in Batman: AC. Using the 320.49 driver results in very bad performance, presumably from the new physx release included with 320.49. Looking at afterburner, my cards don't boost to their max speed with the new physx release included with 320.49, and get low utilization. 320.18 never had such issues. In fact, with 320.18 and a single 780 I can use FXAA with everything else maxed, and the game never, ever dips below 60 fps. This is with Physx-HIGH at 2560 resolution. But my framerates with 320.49 aren't acceptable.

Reverting back 320.18 and it's smooth as silk. Not sure if anyone else noticed this, but there's definitely something awry with the new beta 320.49 drivers and physx games. I highly suggest reverting back to 320.18 and framerates should be a bit higher. This is without a dedicated physx card of course, I find such things to be a complete waste.

Since I currently have 320.49 installed, I'm going to look into this here. I don't own this game, but could check Mafia II, Cryostasis, and Metro 2033.

This may be a dumb question I should already know the answer to, but -- is it not possible to leave the 320.49 display driver installed, and just roll back the PhysX driver to 9.12.1031 via separate manual installation? Or do the two have an internal relationship to one another in the package that can't be broken?

Have you compared 9.12.1031 to 9.13.0604 using a benchmark like FluidMark, or would something synthetic like this not show the issue?

 

[blackened23]

Since I currently have 320.49 installed, I'm going to look into this here. I don't own this game, but could check Mafia II, Cryostasis, and Metro 2033.

This may be a dumb question I should already know the answer to, but -- is it not possible to leave the 320.49 display driver installed, and just roll back the PhysX driver to 9.12.1031 via separate manual installation? Or do the two have an internal relationship to one another in the package that can't be broken?

Have you compared 9.12.1031 to 9.13.0604 using a benchmark like FluidMark, or would something synthetic like this not show the issue?

Actually, I could have done this (rolled the physx version back) -- I think it's merely the physx release included with 320.49 that caused strange GPU utilization issues within Batman: AC - probably not the driver release itself. I'll try that later tonight or tomorrow if I get a chance.

I'm really not sure if that was applicable to other games or just something strange on my end. I didn't try anything else at length - But the difference between physx 9.12.1031 and 9.13.0604 was night and day in Batman: AC, when using a single GTX 780. My other settings were FXAA, physx high, and everything else maxed at 2560x1600 resolution. I'll check on this later and see.

 

[stahlhart]

No significant difference in FluidMark:

 

[Erenhardt]

how about gtx650 vs gt640 as a dedicated physx processor

 

[RaistlinZ]

Golgatha,

It's interesting that the 650 works just as well as a 770 for dedicated PhysX rendering. Do you have access to a 4601GB you can test as well?

 

[toyota]

However, some benchmarks I could find Online using a GTX 650 vs GTX 650 TI showed the 650 coming out ahead because it was clocked higher. Difference in architectures? I dunno. If anyone could shed some light for me, it would be much appreciated.

there is no chance that a plain gtx650 will ever come close to matching much less beating a gtx650 ti. a gtx650 ti is 40-50% faster than a plain gtx650.