PDA

View Full Version : How much GPU does PhysX use?


Yaboze
06-27-08, 11:25 AM
Since PhysX is now routed to the GPU in the newest driver sets, how does performance compare to running a game with PhysX off vs. on (although with nicer particle effects) ?

ChrisRay
06-27-08, 12:22 PM
Just like n the CPU any type of PhysX processing must take place before the data reaches the GPU pipeline. The GPU can also do these calculations before rendering occurs. The question isnt how much GPU does PhysX use. But if its faster to have GPU draw physX first then render. Or the CPU to draw PhysX first and then render.

Chris

tweaked
06-27-08, 02:35 PM
yeah, good question.

I have an 8800gts(640) I'm thinking of throwing in once the nvidia physx gets more widely adopted.

Can you tell the drivers to only use Physx on one card?

i originally laughed when Nividia bought physx... the concept of "just use your OLD card" seemed silly.

but here i am, in that exact position..
and I'll admit...
I'm a little excited by the idea.

SH64
06-27-08, 02:54 PM
I wonder how the physx processing will impact the performance in any game in general.

if a game is taking say 100% for the GPU workload , what happens when the physx processing starts to kick in ? is there going to be any kind of processing distribution ? i.e. 70% of the GPU workload/SP's for graphics & 30% for physics .. ??

any impact on the performance/framerate ?

Ninja Prime
06-27-08, 04:29 PM
Physx never announced the performance of their chip so its hard to say, if they are using the same kind of workloads. However, they were using 128mb of ram on their boards so assuming they do the same work, I would imagine you're losing 128mb or so. Unless they are using system ram, but I dont know if that would be fast enough, seem like Physx boards used fairly fast GDDR3, PCIe bus probably wouldnt be able to keep up the same rates.

This all assumes they are doing a similar amount of work.

Ninja Prime
06-27-08, 04:48 PM
Hmm I read this:

http://en.wikipedia.org/wiki/PhysX

Peak Instruction Bandwidth: 20 billion per second

Which may indicate 20Gflops, which isnt much really, however, I can imagine it takes more power to do the same thing on a GPU as compared to a custom physics chip, like the way GPU's have to do extra calculations in Folding@Home.

So, taking that all in I would expect it to take somewhere between 30-40 Gflops for physics calculations equal to a PhysX chip. Thats maybe 4-5% of the the GTX 280 GPU's power, and about 3% of the 4870s power.

Of course, theres other factors there that will effect actual performance, like how much bandwidth its eating up, and on chip factors that could make big differences, so that really doesn't show the whole picture.