Quick PhysX Case Study: 660ti + 460

P

pj-schmidt

Limp Gawd
Joined
Oct 12, 2004
Messages
356
After seeing the Borderlands2 PhysX demo I have been inspired to try testing a dedicated card for physx processing.

My computer is a Intel i7-2600k overclocked to 4.4Ghz with 16GB of DDR3-1600 ram. My main graphics card is a GTX 660 Ti. I had a GTX 460 laying around, so I put that in my box as well to be my PhysX card.

My first test was to max the settings in Batman: Arkham City and run it at 1080p. I then ran the built in benchmark. Below are the results.
72 fps avg - GTX 660ti only
73 fps avg - GTX 660ti + GTX 460
The dedicated card is basically not worth the electricity to run it in this scenario.

The second test was to switch to surround mode with my 3x 1080p monitors (5760x1080). Re-running the benchmark gave me my second set of numbers.
39 fps avg - GTX 660ti only
43 fps avg - GTX 660ti + GTX 460
In this case the dedicated card added 10% increase in frame rates which is definitely noticeable.

My Thoughts:
1.) If your main card is not working hard, the added overhead of a dedicated physx card is not worth it.
2.) Quantity of memory on the physx card is almost irrelevant. The max amount used during the test was 175MB. Don't pay for extra memory if you're looking for a dedicated card.
3.) During scenes with large amounts of physx effects, the GTX 460 was momentarily maxed out. I suspect that cards slower than a 460 will have a hard time keeping up with being dedicated physX cards for modern games and drag the main card lower instead of helping.
3.) Scenes with little physx effects left the 460 doing almost nothing while the main card continued to run hard. If you have two identical cards, running them in SLI is a much better use of their combined performance.

Hope this is helpful. Let me know if you have any questions. Keep in mind, this was just a quick test for my entertainment and not intended to be comprehensive or conclusive. I may post more as I do more testing.

~daPhoosa
 
In your testing were the minimum frame rates better, eg 660ti 20 fps min, with GTX460 30fps min?
same question for test two
 
The quest for better low frames. I was toying with the idea of adding a physx card but the number of titles that use it are lacking.
 
n3zyd said:
In your testing were the minimum frame rates better, eg 660ti 20 fps min, with GTX460 30fps min?
same question for test two
Click to expand...

The minimum reported frame rate was inconsistent. I believe this was due to stutters during scene changes. The maximum frame rate increased in all cases when both cards were used, so I suspect the true minimums may have as well.
 
I just bought a GTX 460 myself to try out with my GTX 680. I ran the Batman: AC benchmark to compare to your results. The tests were run with all settings maxed, except for FXAA which I had set to medium. I ran the benchmark 3 times, ignoring the first run results.

Basic specs: 2600k @ 4.5ghz, game running off Intel 510, reference Asus GTX 680 @ stock, MSI GTX 460 1GB @ stock.

GTX 680: 2560 x 1600 PhysX High
Min: 30, 28
Max: 86, 87
Avg: 60, 61

GTX 680 + GTX 460: 2560 x 1600 PhysX High
Min: 31, 25
Max: 93, 94
Avg: 68, 68

GTX 680: 1920 x 1200 PhysX High
Min: 10, 24
Max: 129, 131
Avg: 75, 75

GTX 680 + GTX 460: 1920 x 1200 PhysX High
Min: 11, 32
Max: 138, 138
Avg: 81, 82

I also found the minimums to be unreliable in this basic test. I did, however, show a benefit at both 1920 x 1200 and 2560 x 1600 of around ~9% in average FPS.

To put it in perspective, I ran the test with just my 680 and PhysX set to normal and ended up with a min. of 35, max of 87, and average of 67.

Adding the 460 essentially allowed me to bump PhysX to high and maintain similar results to just a GTX 680 running at normal PhysX levels. Probably not worth it, but electricity is cheap around here so the card is going to stay in!
 
interesting results Jolli.

I may retest my 1080p numbers after seeing your results, seems like I should have seen more of an improvement than I did.
 
foozy said:
how do you guys think it would do with a 9800GT?
Click to expand...

I'm hoping to test some slower cards in the near future. For reference, the GTX 460 768mb is roughly 3x the physx power of a 9800gt.

My guesses:
1) 9800gt is better than a cpu if you're dying to get some physx action on a low end computer.
2) Paired with a relatively slow card (maybe a 560Ti or slower), it may help.
3) Paired with a relatively fast card (570 or faster), it may become a bottle neck.

Again, these are just guesses. When testing with a 460, I don't see amazing increases in performance over using my main card. This leads me to believe that cards much slower than a 460 will be more likely to bottleneck the main card.

See here for an example of a slow physx card slowing down the main card: http://www.youtube.com/watch?v=cbww3dhzK0M
 
Last edited:
I went ahead and ran a few tests with MSAA instead of FXAA to see if loading the main GPU up more would help. I only ran each test twice, so the results might not be the best.

GTX 680: 2560 x 1600 4x MSAA PhysX High
Min: 17, 21
Max: 65, 66
Avg: 46, 48

GTX 680 + GTX 460: 2560 x 1600 4x MSAA PhysX High
Min: 20, 24
Max: 72, 71
Avg: 54, 54

GTX 680: 2560 x 1600 8x MSAA PhysX High
Min: 17, 15
Max: 45, 45
Avg: 35, 36

GTX 680 + GTX 460: 2560 x 1600 8x MSAA PhysX High
Min: 18, 18
Max: 51, 51
Avg: 40, 40

The results look slightly better using MSAA instead of FXAA, but it's nothing massive. Makes sense that offloading PhysX would help when a greater demand is placed on the main GPU.
 
I just ran a benchmark, 1920x1080, all settings maxed except for AA where I used 4xMSAA.

GTX 680 PhysX High
Min : 13
Max : 96
Average : 56

GTX 680 + GTX 480 PhysX High
Min : 19
Max : 110
Average : 71

The difference in my case seems to quite large...
 
blade52x said:
I just ran a benchmark, 1920x1080, all settings maxed except for AA where I used 4xMSAA.

GTX 680 PhysX High
Min : 13
Max : 96
Average : 56

GTX 680 + GTX 480 PhysX High
Min : 19
Max : 110
Average : 71

The difference in my case seems to quite large...
Click to expand...

We have nearly identical computers, but are getting some pretty different results. I'm running the newest beta drivers (306.02).

Here's my results at your settings:

GTX 680 Only @ 1920 x 1080 4x MSAA PhysX High
Min: 13, 31
Max: 110, 111
Avg: 66, 66

GTX 680 Only (OC offset +150 on core, +100 on mem) @ 1920 x 1080 4x MSAA PhysX High
Min: 10, 32
Max: 122, 115
Avg: 70, 70

Now adding the 460:

GTX 680 + GTX 460 @ 1920 x 1080 4x MSAA PhysX High
Min: 12, 19
Max: 122, 123
Avg: 76, 77

GTX 680 (OC offset +150 on core, +100 on mem) + GTX 460 @ 1920 x 1080 4x MSAA PhysX High

Min: 14, 13
Max: 134, 131
Avg: 78, 78

It's unfortunate I sold the GTS 450 and gave away my GT 430, GT 240, & GTS 250. Would have been nice to compare.
 
It could be my drivers - I'm still on 301.42. I'll update to the 306 tomorrow and report back.
 
jolli said:
We have nearly identical computers, but are getting some pretty different results.
Click to expand...

I don't know how you can say you have nearly identical computers. You have a GTX 460 and he has a GTX 480. That's top-end versus mid-range for the 4xx series. It makes sense that his gains are larger, right? I'm going to try a GTX 580 for PhysX with my GTX 680. I should have similar but slightly better results than blade52x.
 
sk3tch said:
I don't know how you can say you have nearly identical computers. You have a GTX 460 and he has a GTX 480. That's top-end versus mid-range for the 4xx series. It makes sense that his gains are larger, right? I'm going to try a GTX 580 for PhysX with my GTX 680. I should have similar but slightly better results than blade52x.
Click to expand...

I was more alluding to the fact that I am getting similar performance with just a GTX 680 to him with a GTX 680 + GTX 480. We both have 2600K's clocked in the same region, so something seemed off with his results.

A GTX 480 or GTX 580 would absolutely be better in any imaginable way.
 
Hey, sorry I forgot about this completely. I've updated to 306.xx drivers some time ago, and they did give me a boost

GTX 680 High PhsyX (stock 1202/6000) : 11 min, 119 max, 67 average

Unfortunately I took the GTX 480 out for now and don't have that benchmark. I hope I can get around to that more quickly this time. :p
 
Well the real question is, its it really worth the trouble?, With today's Technology I have [1x] 670GTX 4GB on a 1920x1080p & I play BF3 & Skyrim with no problems,
I average no lower than 65-fps most of the time it hangs on 70-80fps & sometimes I guess depending on area of game it shoots up to from 120fps! :eek:
& I have thought about a dedicated PhysX card, but after lots of reading & researching I came to my conclusion > If I do go with another card I would rather go SLI Mode without dedicating a card to PhysX, I believe SLI or Crossfire is a much better way to go!
 
exilelrrp said:
Well the real question is, its it really worth the trouble?, With today's Technology I have [1x] 670GTX 4GB on a 1920x1080p & I play BF3 & Skyrim with no problems,
I average no lower than 65-fps most of the time it hangs on 70-80fps & sometimes I guess depending on area of game it shoots up to from 120fps! :eek:
& I have thought about a dedicated PhysX card, but after lots of reading & researching I came to my conclusion > If I do go with another card I would rather go SLI Mode without dedicating a card to PhysX, I believe SLI or Crossfire is a much better way to go!
Click to expand...

BF3 doesnt use any physx.

Skyrim I think is locked to cpu physx like Metro 2033.

So if those are the games you play you would not benefit at all.
 
I don't know why it just dawned on me this weekend that my 480 didn't have to go below my GTX670 on my x58 board, and putting it above alleviates most of my concerns. I didn't want to use the notoriously power hungry card for PhysX and have my 670 (Gigabyte Windforce) breath off the back of it. No where on my board could I put 2 GPUs and have an additional space. The 480 might be cramped up top now, but it's only doing PhysX...

End result, Borderlands 2 runs like fucking glass now! It would chug at 1080p on a single GTX 670. I was worried about these "% increase" benchies because even 35% over these massive dips would still be bad fps... but it's butter now.

It would be nice to get something I pick for just PhysX. My 480 idling (40% fan) is the loudest thing in the computer while surfing etc. It raised the "minimum" sound floor of the computer. Never bothered me before, but I guess I got used to the GTX670. When I game the 480 ramps up to about 55% fan, which is lower than it would be when I used to game on it, I figure it's not something I'm not used to (care more about the idle noise to be honest). GTX670 went up maybe 1 or 2 C.
 
Shadohh said:
BF3 doesnt use any physx.

Skyrim I think is locked to cpu physx like Metro 2033.

So if those are the games you play you would not benefit at all.
Click to expand...
I never said it does, I'm simply compering my [1x] GPU VS their [2x] GPU's with one as Dedicated Physx FPS & I personally don't care if I do benefit Physx from my games, it look f'n amazing :eek: as is & I focus on my FPS when I play my online multilayer games & with no issues,
I do play other games but mostly those [2] I just added, I see these guys are talking about their FPS with dedicated Physx which don't look good to me that's why I said is it really worth the trouble? NO!
I rather go SLI or Crossfire as today's GPU can handle Physx with no problems again not much games takes advantages of hard Physx, I'm waiting for ARMA 3 & by that time I will have another 670GTX 4GB SC
on SLI again which I believe would be better to go that rout, that's my opinion as we all state our opinion's here, I believe list of physX games is misleading, the vast majority of them use PhysX on the CPU, not the GPU, @ the end its up to the you to make the choice, good luck on that :cool:
 
I have an MSI 680 4gb paired with an evga 480 dedicated to physx. I get a 10-15 FPS boost in many games.


GTX 680 Far Cry 3 2560x1440 maxed

Min 20
Max 40
Average 25

GTX 680+480 2560x1440 maxed

Min 15
Max 30
Average 15

680 BF3 Maxed 2560x1440

Min 35-ish
Max 60
Average 45

680+480 Maxed 2560x1440

Min 45
Max 60
Average 55

In Dirt 3 I get an almost constant 60fps with the 680+480, once in a while it will drop to 55fps, but it is very rare. Without the 480 my fps would drop to around 40 and waver in-between 45-60 range, but usually it stays in the 50's.
 
Last edited:
exilelrrp said:
I never said it does, I'm simply compering my [1x] GPU VS their [2x] GPU's with one as Dedicated Physx FPS & I personally don't care if I do benefit Physx from my games, it look f'n amazing :eek: as is & I focus on my FPS when I play my online multilayer games & with no issues,
I do play other games but mostly those [2] I just added, I see these guys are talking about their FPS with dedicated Physx which don't look good to me that's why I said is it really worth the trouble? NO!
I rather go SLI or Crossfire as today's GPU can handle Physx with no problems again not much games takes advantages of hard Physx, I'm waiting for ARMA 3 & by that time I will have another 670GTX 4GB SC
on SLI again which I believe would be better to go that rout, that's my opinion as we all state our opinion's here, I believe list of physX games is misleading, the vast majority of them use PhysX on the CPU, not the GPU, @ the end its up to the you to make the choice, good luck on that :cool:
Click to expand...
why not have it all? i have 2 6850's crossfired and a gt430 for physx. borderlands 2 looks great and uses the gt430 (never more than 50% gpu usage). i tried playing with no physx and it just doesn't look the same, i like seeing all that blood go everywhere when i snipe someone's head off.
 
I wonder how demanding PhysX will be in PlanetSide 2.

http://www.youtube.com/watch?v=lcJlTUgZTT8

It looks like your typical PhysX in the video above, but multiply that x100 in large scale battles and who knows? Right now the GPU physics option is disabled (was available in beta for a short period of time).
 
bigted said:
why not have it all? i have 2 6850's crossfired and a gt430 for physx. borderlands 2 looks great and uses the gt430 (never more than 50% gpu usage). i tried playing with no physx and it just doesn't look the same, i like seeing all that blood go everywhere when i snipe someone's head off.
Click to expand...
I hear ya on that, but my gaming does not demand PhysX to a level where I'm like <<< :eek: >>> so I'm not losing any sleep over it :D I focus more on my FPS that's what count big time for me especially on multiplayer!
 
I've been getting into Alice but with my 470 I have to put PhysX on low. Just bought another 470 to run SLI (or use as dedicated PhysX). The game just looks so much better with the full physics effects on.
 
Here's some more Batman AC numbers. I said I was going to around to doing this eventually, just never when!

Driver version : 306.97 WHQL
Game Settings : Maxed 1080P with 4xMSAA

The following is an average of 3 runs :

Mode : Min, Max, Avg
GTX 680 no PhysX : 33, 163, 118
GTX 680 PhysX : 16, 122, 70
GTX 680 + GTX 560 PhysX : 25, 132, 80

GTX 560 no PhysX : 32, 71, 52
GTX 560 PhysX : 21, 53, 40
GTX 560 + GTX 680 PhysX : 24, 61, 47

Based off Batman AC :

- GTX 680 dropped around 32% in performance with PhysX enabled. An overkill dedicated card boosted PhysX performance around 14%.
- GTX 560 dropped around 23% in performance with PhysX enabled. An overkill dedicated card boosted PhysX performance around 18%.

Now I really wish PS2 to re-enable PhysX soon. :D
 
You must log in or register to reply here.
Back
Top