News bit-tech revists PhysX in UT3

http://www.bit-tech.net/news/2008/01/18/bit-tech_revists_ut3_physx/1

When we first looked at the PhysX features in UT3, we were a bit non-plussed. It turns out though that our hardware was a little glitchy, so we're having a little revisit.

 

That's fair enough.

Still can't see myself buying a PhysX card any time soon, though - There's just no appeal.

 

not sure why this isn't apart of a card, like a sub-processor. I would by a HD3870 with PhysX, but not just a PhysX card

 

Shouldn't need to be part of a graphics card. We have quad-core processors. Just take 1 core and dedicate it to physics. Problem solved. If you have a dual core you should be able to decide if the game uses both cores or 1 core and you get the advance physics processed on the other. Physics cards should be dead already.

 

I would buy a card, but the price would have to hit a sweet spot in order to do so. It's just not economically feasible to do so at this current moment.

 

A CPU is a generic data processing machine. It is designed to work on many sizes and types of data. The PPU in a PhysX card would be designed for a particular purpose and as such be much better at that job. If your argument were to be correct we wouldn't need a graphics card. Just dedicate one of the CPU cores to graphics. The same counter argument stands -- a GPU is specifically designed to do a job which is much faster than any CPU at doing the same job.

 

One core on a CPU for graphics? maybe one for PPU? two for everything else? Wouldnt be suprised if we had these chips in 12 months (thinking Nehalem)

 

Well, a processor just does sums, right? There is no difference at base level between drawing a person and working out where his shot off arm will fly, it's just maths in the end. Piling only physics calculations into a processor will give only physics 'answers'. That's why a 'physics processor' can do so many sums so quickly, because it has nothing else to do*

If a processor is fast enough that it can have a redundant core while playing a game, then that core could do the physics calculations. Indeed, if the thing was hugely powerful, you could have a core dedicated to graphics processing. It's the same deal as putting graphics or physics processors on the PCI-E bus. Put them inside the CPU package and you eliminate that bus so potentially it's much faster, but I guess that's probably a way off.



I'm pretty surprised they are still clinging to this product and trying to get us to buy them. Back in the day when I ran a 386, we got a co-processor to help with the sums in CAD and I'll be honest it flew.

Now, with CPUs as fast as they are and graphics card as ridiculously powerful, I think trying to re-light the co-processor market (which this effectively is) seems like madness.



* admittedly the data might be arranged and delivered in a more favourable manner by some other technical wizardry and little magic chips, but that'd be doable inside a cpu if someone really wanted to.

 

Well, Intel's got Larrabee coming at some point in the future (it's probably less than 12-18 months away, max)... that's a many core architecture using IA and I wouldn't be surprised if, given Intel's acquisition of Havok, Intel started using some of those cores for physics. After all, Intel has already demoed vortices and other physics in Alan Wake using a first-gen quad-core chip. By the time that game ships, the physics will be pretty damn fast on whatever the fastest Intel processor is then.

Welcome to the forums.

 

I'm with Jamie on this. Those co-processors didn't go away at all - they just got integrated. You'll notice that sometimes CPU benchmarks are broken up into ALU & FPU scores. (Read the links provided if you doubt.)

 

Interesting that your PPU had failed, its not like you have use it that much. I would be furious if my graphics card died after 1-2 years of light use (that said, my motherboard did )

 

jamie is 100% right, I'm not a fan of add-in hardware based physics solutions, but that doesn't mean a General purpose processing would be nearly efficient at the calculations that a dedicated hardware solution would be. It's the same for graphics cards, with dedicated hardware, drivers and algorithmns for the specific task. you also must remember that a physx card is running an optimized physics engine designed to work well with those cards. there really is no competition

the same goes for sound cards and graphics cards. Yes general purpose CPU's are speeding up and having increasing cores, however a dedicated hardware solution that is meant to specifically handle a certain type of calculation can be more focused and optimized as opposed to a CPU which has to handle everything imaginable.

 

Regards sound cards, the amount of parallel computations needed for that task are far fewer than realtime lighting & physics interactions. So the same does not apply to sound cards to the same extent.