New Ageia PhysX processor on PCIe pictured

"use something a bit less offensive to the year"

 

Is anyone still excited by Ageia Physx in the face of ever increasing cores and processing power from Intel and AMD CPUs, and ever increasing cores and power from Nvidia and AMD GPUs?

If I was building a new PC any time in the next six months, I can't think of anything that would be bottom of my list of requirements under a PhysX board, especially given the massive lack of support we've seen for Ageia PhysX in upcoming games.

I just can't see any developer putting time and effort into this when they could be putting time into CPU cores or GPU boards. At least those can do more than one thing, and are guaranteed to exist inside every target PC in one form or another.

 

Nope, not even in the slightest.

 

Why do they even bother...

Also, I am wondering, what is the overhead of sending thousands of object positions between this card and the gfx card? I'm guessing, a lot?

 

Hm, that bigger and longer die resembles two cores sliced together, just like Smithfield.
Longer PCB, external power plug, more heat & noise, less free expansion slots... damn, I'll just take a piece of Barcelona or Yorkfield next time, and skip this.

 

I said it when the card came out and I believe it even more strongly now: there's nothing fundamentally unsuitable about physics for current GPUs, and even multicore CPUs can keep up nowadays.

Thus to be honest Ageia cannot compete with the efficiency and mass-market funding of GPUs... nor should they try! Physics can be done very efficiently on the GPU (or Cell SPEs) so it's better to just do it there, which allows load balancing between physics, graphics and other tasks (and reduces data movement). Having a CPU, GPU, PPU, AIPU and god knows what else simply means that you're going to *always* have a good percentage of your hardware idle.

Anyways I have a lot of respect for Ageia's physics SDK, etc. but this custom hardware path is a dead end IMHO. Unfortunately they're stuck with a business model that means either the hardware succeeds or the whole company is in trouble...

 

Their hw is going no where, imho they should re-focus on GPGPU applications and even GPGPU graphics (alternative rendering methods as ray tracing, reyes, etc..)

 

How is Havok doing right now

 

The big difference is that Havok is a middleware software product sold to software developers, whereas the Ageia business model relies on selling specialised hardware (in the form of the PhysX add-in board) to gamers.

PhysX is just not a compelling product in the current market, against competing hardware capable of the same job, and with the current low level of developer support in games.

 

PCI-e x8/x16? What a horrible decision. They've shrunken their market for this product down to include only those that have a spare PEG slot, which basically means new-gen ultra-high-end 3/4 PEG slot boards or under-utilized dual-slot boards that don't have the 2nd slot populated. Should've gone with x1.

 

by "next generation Ageia PhysX processor" do you actually mean " the same Ageia PhysX processor just on on different bus ?"

they retail for £50 now (£150 when they first come out) although this could just be the retailer trying to get rid of stock

somewhere (sorry forgot the link) took another look at it and this time they sorta liked it they said it is now a physics accellerator whereas it used to be just a physics enabler ageia have sorted out the slowdowns and apparently games now run faster with the card instead of slowing down like they used to
but i cant see it taking off though escpecially with the poor performance of dx10 gfx cards are struggling now imagaine what it would be like if they are rendering many thousands of peices of debris

also got to agree whith shaidarharan they should of gone with pci-e 1x

ps: gfx card physics is that dead now ?

 

ps: maybe they should make a havok wrapper

 

Just look on the chip-size, if it is not in 130nm like the first PhysX and for example 90nm, than it has a much higher transitorcount and so more much more power.

 

mm....

One thing that is of interest is Epic's (somewhat) recent incorporation of PhysX into UE3.0, though I don't know if that is only for their own games or if licensees get that as well. Given how popular UE3.0 seems to be, the latter case may be enough to present competition with Havok. But then again, maybe Epic used PhysX because it was cheaper and not necessarily better in any way compared to Havok's software.

 

Looks like it's going to be a UT3 mod, so again very limited support that won't be standard.

Given that the vast majority of the online community that Epic will try to build with UT3 won't have PhysX boards, I don't see it having much impact.

I don't see why any gamer would spend money on Ageia ahead of a faster CPU with more cores, or a second GPU, or more memory, etc, and these will all have a bigger impact on the whole gameplaying experience.

 

My understanding is that their Novodex(sp?) stuff is actually quite solid(that's the name of their SDK AFAIR), asides from being free, and it actually does a number of things better than Havok does. The trouble for Ageia, as someone pointed out above, is that whilst their SDK is solid, it's also free, and a developer using it isn't exactly forced to code for their card, which is their primary source of income.

I'm still not totally sold on the GPU doing physics...I haven't yet seen a single current or upcoming implementation of this approach. It should do it quite well, but the question remains wheter or not there's going to be GPU muscle available to spare for physics?Multi-core CPUs are likely to be king-of-the-hill, if only for the fact that a CPU upgrade is a no-brainer and everybody is likely to have one, thus giving the largest possible installed-base.

 

The PhysX SDK is only free for noncommercial use.

 

HavokFX has had production-quality stuff for years now. Although they only use it for "effect physics", there's no real reason why it couldn't be used for everything (they'd need to make it a bit more full-featured, but that wouldn't be hard). ATI/AMD has also been demoing full broad+narrow phase collision detection on the GPU for several years.

The upcoming game "Hellgate: London" will use a lot of GPU physics, both in terms of HavokFX and also real-time fluid sim (a la NVIDIA smoke demo) and more. It'll be out in the next few months IIRC.

If you need more power, add another GPU (or get a faster one, or both) I've never understood this argument about "how can the GPU do physics if it's doing graphics?" since conceptually it's no different from the CPU doing bits of both, and indeed the GPU is already much *more* parallel.

This I can certainly agree with... with quad-core CPUs being <$300 (!!!) now people will certainly be buying them. That said, they're still technically less efficient than GPUs for many physics computations (both in perf/$ and perf/W), but the fact remains that the average person seems to like purchasing fairly fast CPUs and relatively slow GPUs.

That said physics isn't as taxing as graphics for instance, even with thousands of objects. Thus a dual/quad-core CPU is usually enough, and thus is the obvious target especially with a lot of x86 physics code still sitting around.

 

Hypothesis:

SLI/Crossfire among heterogeneous chipsets will result in serious adoption of GPU physics, since you can then either get physics or faster graphics.