• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

CPU PhysX?

Associate
Joined
29 Dec 2009
Posts
649
Location
Germany
So why can't I just run PhysX on my CPU instead, without losing functionality?

Does PhysX really stress so much that a 4 GHz quad can't handle it?
 
It's just NV's way to make you part with more cash for effects that are pap, and still with a big performance hit, forget phsyx, it will die on it's ass, just look at the current and future title that will support it.
 
Depends on quite a few things... if your just chucking around 10s of RBs and a handful of ragdolls then the CPU could handle it fine, if the CPU was just doing physics and no other game processing then it could handle quite a bit of physics fine... in a scenario tho where you have a multi thread capable game thats making decent use of the CPU and the game contains a number of soft body, cloth, fluid, etc. physics effects it would be brought to its knees no matter what current CPU - even the new 12 core ones... whereas a decent GPU could pull it off and still be running 40+fps.
 
I actually quite like PhysX, but it would be much better if it was available on both manufacturers' cards. Because then it would be utilised a heck of a lot more.

Nothing worth crying about though.
 
Hardware physics is a great idea (and something future games will need) sadly nVidia are stiffling its progress and intel and ATI/AMD making little effort towards it. PhysX has the potential to be one of the next big things in gaming but everyone on all sides hold it back :(
 
Last edited:
Depends on quite a few things... if your just chucking around 10s of RBs and a handful of ragdolls then the CPU could handle it fine, if the CPU was just doing physics and no other game processing then it could handle quite a bit of physics fine... in a scenario tho where you have a multi thread capable game thats making decent use of the CPU and the game contains a number of soft body, cloth, fluid, etc. physics effects it would be brought to its knees no matter what current CPU - even the new 12 core ones... whereas a decent GPU could pull it off and still be running 40+fps.

What about UT3 or Mirror's Edge?

Also, are there any modern PhysX-only cards? Getting a whole NVidia card just to have hardware PhysX seems like a bit of a throwaway.
 
PhysyX is mostly crap. It will not make a bad game good, or a good game better. A few extra bits of flying debri within a few average games is not worth the hassle. Forget about it.
 
What about UT3 or Mirror's Edge?

Also, are there any modern PhysX-only cards? Getting a whole NVidia card just to have hardware PhysX seems like a bit of a throwaway.

Very little in either game that would tax a CPU that badly - tho mirrors edge does have some nice cloth and glass effects that really would need GPU... they aren't a major enhancement of the game but it adds to the immersion a bit.

PhysyX is mostly crap. It will not make a bad game good, or a good game better. A few extra bits of flying debri within a few average games is not worth the hassle. Forget about it.

PhysX itself isn't crap - with the right useage it will very very deffinatly make a good game better, its capable of so much more than a few bits of debris flying around.
 
Depends on quite a few things... if your just chucking around 10s of RBs and a handful of ragdolls then the CPU could handle it fine, if the CPU was just doing physics and no other game processing then it could handle quite a bit of physics fine... in a scenario tho where you have a multi thread capable game thats making decent use of the CPU and the game contains a number of soft body, cloth, fluid, etc. physics effects it would be brought to its knees no matter what current CPU - even the new 12 core ones... whereas a decent GPU could pull it off and still be running 40+fps.

To be fair - I think most fast quad cores should be able to handle anything GPU physX is currently doing in games. It's just that it's implemented to run on the GPU. Have you seen how low CPU utilisation is in PhysX titles?

Dual cores seem to struggle in Ghostbusters but quad cores do very well and are mostly utilised. That game throws around an awful lot of stuff - more so than some GPU physX titles.
 
Ah, I was just interested in that small bit of "Extra Immersion" though. I've read some posts by people that were absolutely stunned when they enabled PhysX when playing, for example, Mirror's Edge - so I thought the game would be 100x better with it on.

Is there a way to force PhysX effects to be used even with no PhysX-capable card installed? I want to see if I can realistically run it on my CPU.
 
Throwing around 3000 RB primitives and a few ragdolls in a highly specialised solver is one thing, performance on general purpose middleware physics engine another. Very few studios have the skill, time and resources to create their own highly specialised physics solver.

A fast quad core can quite easily handle far more rigid body physics and simple ragdolls than we have seen in most games to date - even PhysX titles. A fast quad core absolutely can't handle the physics workload in a proper game environment with a reasonable level of both rigid and soft body physic effects (a GPU can) - which no one would say didn't enchance the game considerably if they actually played such a game but won't happen because no developer wants to narrow down their potential market that much.
 
Throwing around 3000 RB primitives and a few ragdolls in a highly specialised solver is one thing, performance on general purpose middleware physics engine another. Very few studios have the skill, time and resources to create their own highly specialised physics solver.

A fast quad core can quite easily handle far more rigid body physics and simple ragdolls than we have seen in most games to date - even PhysX titles. A fast quad core absolutely can't handle the physics workload in a proper game environment with a reasonable level of both rigid and soft body physic effects (a GPU can) - which no one would say didn't enchance the game considerably if they actually played such a game but won't happen because no developer wants to narrow down their potential market that much.

PhysX would certainly be interesting in a fully-sandbox environment where nearly every object has physics, but it seems like there are no games to date that actually use this.

Note: Interesting enough, the most physics-heavy game I have is Oblivion, and it's only optimized for running on a single core, which a 1 GHz can easily play at 30+fps, and my current setup eats it up. Every single object has physical properties in Oblivion, even if they aren't 100% realistic, but collisions etc. are very convincing, especially if picking something up and dragging it against some other object.

This is only rigid body though.
 
Just curious, are there any benchmarks showing PhysX acceleration, CPU vs. GPU when the CPU and GPU both have very similar theoretical levels of throughput? Say having a 3.6GHz Q6600 (110.2 GFLOPS single precision peak throughput) vs. a graphics card of similar peak levels of throughput (say an 8600 GT, 113.28 GFLOPS) in software vs. hardware accelerated modes and seeing what 'wins' - presumably with a dedicated rendering card not involved in the PhysX acceleration process.
 
You can run the PhysX screensaver on L5 with both CPU only and GPU modes... the difference between a Q6600 @ 3.6gig and a stock GTX260 was in the region of 4 v 55fps - and remember the GPU is also rendering.
 
Seems to be making reasonable use of all 4 cores on my Q9550 - not maxing them out but just because its not pushing 100% kernel usage doesn't mean its not making full use of what the CPU can bring to the table.
 
LMAO with future processors offering more cores. One would only assume that Physix would become obsolete within time.

If/when we end up with a unified processing array rather than discrete CPUs/GPUs that would happen... even intel is struggling with that tho so we are a long way off unfortunatly.
 
If/when we end up with a unified processing array rather than discrete CPUs/GPUs that would happen... even intel is struggling with that tho so we are a long way off unfortunatly.

I don't think that's entirely accurate. Sandy Bridge (scheduled for early 2011 release) should be more than adequate for high performance physics with all the eye candy turned on, at least in its 8-core form. Actually Sandy Bridge (and AMD's Bulldozer architecture, actually) is interesting because it has doubled the SIMD unit width over contemporary processors, whilst also doubling the number of cores available. Actually, I wouldn't be surprised if Intel was planning on stealing Fermi's thunder by releasing a 16-core glued-together part for the HPC space (which could potentially have performance in the 800+ GFLOPS range, which is enough to make it competitive with Fermi even if 'theoretically' it doesn't beat it).
 
Back
Top Bottom