PDA

View Full Version : 2nd card still needed with 480 for Physx ?


turdhat
04-15-10, 10:23 AM
I got double in fluidmark with the 480 going from a 280 and 8800gt for dedicated physx however batman is slower with Physx and I did get the new dll for Batman and the 480. I am going to put the 8800gt back in when I get home and rebench with all physx based benches I can find let you guys know.

CaptNKILL
04-15-10, 11:02 AM
I got double in fluidmark with the 480 going from a 280 and 8800gt for dedicated physx however batman is slower with Physx and I did get the new dll for Batman and the 480. I am going to put the 8800gt back in when I get home and rebench with all physx based benches I can find let you guys know.

I'm very interested in seeing how Physx games perform with a 480 vs a 480+8800GT.

If you can, run Cryostasis or the Cryostasis demo.

brunner
04-15-10, 01:46 PM
Physx is always faster if you put it on a seperate card. Even with Quad-SLI, it's better to have a 5th card for Physx. For some reason, there's a huge performance impact when Physx isn't offloaded to a GPU that is just dedicated to it.

CaptNKILL
04-15-10, 07:03 PM
So what's the verdict turdhat? :)

I'm actually considering buying a GT240 on the cheap if I can, as a step toward upgrading my graphics. I can use it with Physx with whatever GPU I get and I'll be able to play the less intensive games right now.

Viral
04-15-10, 07:39 PM
I would assume that a dedicated physx card would help in any situation where a game isn't CPU limited.

Bah!
04-15-10, 07:42 PM
Physx is always faster if you put it on a seperate card. Even with Quad-SLI, it's better to have a 5th card for Physx. For some reason, there's a huge performance impact when Physx isn't offloaded to a GPU that is just dedicated to it.

I saw a benchmark on Guru3d that showed virtually no difference between 480's in SLI and the same with a 280 for Physx:

http://forums.guru3d.com/showthread.php?p=3540436

And here is a single 480 with and without a 240 for Physics:

http://www.overclock.net/nvidia/712321-gtx480-physx-very-interesting-results.html

Finally a 470 with a 240:

http://www.overclock.net/nvidia/711622-nvidia-gtx470-gtx480-owners-thread-7.html#post9058698


I'd say a dedicated Physix card isn't needed with the 400 series, unless you just happen to have one laying around. I certainly wouldn't buy a card just for Physx.

brunner
04-15-10, 08:41 PM
I think the guy screwed up. Look at how much higher my results are with just one 5870 plus a 9800 GT for Physx. That run was taken at 1920x1200 w/max settings:

http://img94.imageshack.us/img94/1484/capture2k.jpg

CaptNKILL
04-15-10, 08:42 PM
I think the guy screwed up. Look at how much higher my results are with just one 5870 plus a 9800 GT for Physx. That run was taken at 1920x1200 w/max settings:

http://img94.imageshack.us/img94/1484/capture2k.jpg

Hmm... yeah that's strange. Maybe he was running at 2560x1600 or was using more AA?

brunner
04-15-10, 08:43 PM
Or maybe the 240 is quite a bit slower for Physx than 9800GT? What's a 240 anyway? I'm not up to date with NV's lower end offerings.

Still, in my experiences GPU speed matters little when talking about dedicated Physx cards. All that matters is that you have one that is solely used for Physx.

That's why I'm surprised that a 480 + 240 can only manage:

http://img441.imageshack.us/img441/3858/gts240physx.jpg

brunner
04-15-10, 08:45 PM
Hmm... yeah that's strange. Maybe he was running at 2560x1600 or was using more AA?

You can't select AA in that benchmark. It's possible he's running it at a higher res but I doubt the 480 will drop that much going up to the next higher resolution.

Bah!
04-15-10, 09:00 PM
Or maybe the 240 is quite a bit slower for Physx than 9800GT? What's a 240 anyway? I'm not up to date with NV's lower end offerings.

Still, in my experiences GPU speed matters little when talking about dedicated Physx cards. All that matters is that you have one that is solely used for Physx.

That's why I'm surprised that a 480 + 240 can only manage:



Apparently a 240 is quite a bit slower than an 8800 GT, so that would make it quite a bit slower than the 9800gt as well.

http://www.tomshardware.com/reviews/geforce-gt-240,2475.html

CaptNKILL
04-15-10, 09:47 PM
The only thing that should matter for Physx is the shader clock and the number of SPs.

The 9800GT has 112 shaders clocked at around 1.5Ghz, the GT240 has 96 shaders clocked at around 1.34Ghz. They shouldn't be that far off since the 240 is also a much newer design.

turdhat
04-15-10, 10:00 PM
Very here is what I found.

I only used Fluid Mark, Batman and the rocketsled demo because I dont have time for much else.

No numbers just observations. Sorry.

Batman for me has a slower minimum FPS that with my 280/8800gt combo. Dont know why and not really worried. I think it is a driver issue because Fluidmark runs twice as fast with the 480 doing all the work. Switching to the 8800gt cuts the Fluidmark score in half.

Rocketsled demo ran about the same but has some stutters. I will try Cryostasis when I have more time.

CaptNKILL
04-15-10, 10:02 PM
There is a patch for Batman for use with the 480\470.

http://www.nzone.com/object/nzone_batmanaa_downloads.html

newparad1gm
04-16-10, 01:08 AM
I'm glad that currently it seems that the 480 isn't really benefiting from a PhysX GPU offload, as I'm not really sure that my 1050W could handle both my GTX 480s and a 9600 GT as well.

jlippo
04-16-10, 01:21 AM
One of the reason for 480 to do so good in comparison is better state change handling between graphics and gpgpu.
280 most likely twidled it's thumbs for a while every time this happened, so physics had quite big impact in performance.

brunner
04-16-10, 01:56 AM
One of the reason for 480 to do so good in comparison is better state change handling between graphics and gpgpu.
280 most likely twidled it's thumbs for a while every time this happened, so physics had quite big impact in performance.

How are you guys drawing this conclusion? I'm getting much better results with a single 5870 + a 9800GT than the guy with the 480 +240.

Max @ 1920x1200, Physx-Max, 4XAA

http://img688.imageshack.us/img688/8111/benchh.jpg

Keep in mind 4X AA has much bigger impact on performance for ATI cards in this game.

I think the 240 is just so slow at Physx that it doesn't make much of a difference for the 480 GTX.

brunner
04-16-10, 02:16 AM
Crossfire results, in case anyone is interested. Scaling is not too impressive in this game but it can be seen that Physx isn't the bottleneck in this game.

http://img219.imageshack.us/img219/4759/bench2.jpg

Bah!
04-16-10, 03:57 AM
There is a patch for Batman for use with the 480\470.

http://www.nzone.com/object/nzone_batmanaa_downloads.html

Maybe that is why this guys Batman numbers with the 240 are so bad?

turdhat
04-16-10, 09:04 AM
Maybe that is why this guys Batman numbers with the 240 are so bad?

No, I patched it. Minimum fps is lower. Used to be 60 fps 98% of the time with all maxed, 4xaa and physx on high at 1080p with my old combo of 280/8800gt. I think the drivers need to mature. Someone posted Dirt 2 stuff but Dirt 2 is not a physx game.

edit: 240 ? Oh thats not me. Still my batman number went south a bit.

brunner
04-16-10, 11:30 AM
Turns out that I was right. Off-loading to a 240 for Physx doesn't do much for a 480 because the 240 is just too slow.

Otoh, a 275 + 8800 GT makes the 275 faster than a 480 by itself. Without the 8800 for Physx, the 275 is significantly slower than a 480, as expected. So the results are pretty much in line with what I expected, which is that even a 480 can benefit substantially from offloading Physx to another card.

http://www.overclock.net/nvidia/711622-nvidia-gtx470-gtx480-owners-thread-7.html#post9058698

jlippo
04-16-10, 01:57 PM
How are you guys drawing this conclusion? I'm getting much better results with a single 5870 + a 9800GT than the guy with the 480 +240.

I'm pretty sure that the state change doesn't happen when using one card just for a physx. ;)
Also things like AA and resolution shouldn't affect that much either.