Dedicated PhysX Card Experiment - How Powerful Does it Have to Be? Linus Tech Tips

Linus Tech Tips ·Linus Tech Tips ·2011-05-08 · 745 words · ~3 min read
Floatplane YouTube

Transcript

JSON SRT VTT 55
0:00 so this is actually probably going to be a fairly interesting video here on my
0:04 test bench today I have two things I have a GTX 580 from
0:08 EVGA and I also have an
0:12 8600 GTS so that's a bit of a mismatch
0:15 you'd think what I will be looking at today is the performance impact of a
0:20 high performance graphics card versus a low performance graphics card as a
0:26 dedicated FX processor so this GTX 58 is
0:30 going to be doing all of my Graphics processing regardless and then I'm going
0:33 to run two different scenarios I'm going to run with the 8600 GTS as my dedicated
0:38 physx card and then I'm going to try running with another GTX 580 which is
0:43 obviously ludicrously Overkill as my
0:47 dedicated physx card so if you guys check out up here I have my physx
0:51 configuration set so I have selected the 8600 GTS as
0:57 my dedicated physx card and I'm going to try with both scenarios and I'll let you
1:00 guys know how it goes well I've already
1:04 run into some interesting results so I'm running at 1920 x 1080 on Mafia 2 I'm
1:08 running with high details 8X AF anti-aliasing is on and I've already run
1:15 only two scenarios so I've run the GTX
1:18 580 with the 8600 GTS as a dedicated FX
1:21 card and then I've run the GTX 580 ignoring the 8600 GTS so using itself
1:27 for physx you can force that in the driver as well and and it actually
1:31 scored or the GTX 580 actually scored
1:35 66% Higher by itself than it did by tying it down with
1:41 this 8600 GTS to do the physx processing so let's see what else we can
1:46 investigate here so at this point guys I've run
1:50 quite a few more fascinating scenarios so here you can see my GTX 580 running a
1:55 GTX 550 TI as a dedicated physics card
1:58 but I've actually finished all the Run throughs that I am going to do and I
2:02 just want to share with you my results so this is also going to be a a sneak
2:06 peek into what my next episode is going to be looking at PCIe bandwidth and the
2:10 effects on gaming performance but here I have my full results so the GTX 580
2:15 alone gets about 50 frames per second you throw in a FX card that's too slow
2:21 and it goes all the way down to 30 FPS
2:25 so what that means to me is that you're far better off to just let this card do
2:29 both the graphics and the FX by itself
2:33 then to give it something that's so slow that it bottlenecks it and that the
2:37 physx calculations are going to be behind the rate at which this card can
2:40 draw the frames okay so that is clearly a problem okay so then I tried a few
2:46 cards of varying power as a dedicated
2:49 physx card so I tried an additional GTX 580 I tried a GTX 560 TI as well as a
2:56 GTX 550 so I don't know where the threshold is but it looks like this is
3:01 how it works for physx because these three are all within margin of error of
3:05 each other uh regardless of how much physx power we're throwing at it I mean
3:09 a GTX 580 is clearly overkill for physx
3:12 processing but you can see that the 550 TI and the 560 TI are also both just
3:18 fine so the answer is I don't know exactly where that threshold is and it's
3:22 going to depend from game to game how much physx processing is needed but what
3:27 you need for physx is a card that is fast enough not too slow because that
3:32 will bottleneck your entire setup and not so fast that you're spending way too
3:37 much money on it so something like a 550 or even something a little bit lower end
3:41 or maybe a last generation card that you can get your hands on on the cheap is
3:45 probably the best bet if you have something like a GTX 580 as your main
3:49 card so thanks for checking out this little fizx episode of Linus Tech tips
3:53 don't forget to subscribe for more unboxings reviews computer videos and
3:56 other good stuff like that