ATI... 3 Card Solution for Physics

[jAkUp]

Interesting, check this out:

http://www.nvnews.net/vbulletin/attachment.php?attachmentid=18591&stc=1&d=1149437898

[RiciN]

Looks interesting. Hopefully nVidia comes out with something similiar.

[Banko]

So what makes this different from a PPU then?

[zer0]

So what makes this different from a PPU then?

its powered by motherF*ckn nature...(nana2)

[nutball]

So what makes this different from a PPU then?

This solution is about twice the price, and it's doing something it wasn't designed to do.

But hey, at least you'll be giving your money to the Right Team rather than the Wrong Team!

[slick]

This solution is about twice the price, and it's doing something it wasn't designed to do.

But hey, at least you'll be giving your money to the Right Team rather than the Wrong Team!

Read up on Havok FX.

[Banko]

Read up on Havok FX.
Yes and that will help how? There haven't been benchmarks between the AGEIA PPU and a dedicated video card for physics. Hell Sweeney says the PPU will be a much better solution. From what I can see they also recommend another X1900 series card, wtf is the point? It will be even more pricey.

[nutball]

Read up on Havok FX.

I'm well aware of Havok FX. It doesn't change the fact that GPUs are designed to do a very specific task, and that that task is not physics (or any other sort of general-purpose vector processing). The fact they can do it at all is all credit to them, but it says nothing about the efficiency they achieve. It would be interesting to see a comparison of the efficiencies achieved by CPUs, GPUs and PPUs, but given the level of marketing hype and secrecy surrounding the FP capabilities of GPUs and PPUs I wonder if anyone has achieved this.

GPUs also lack some features which are important in certain classes of algorithm (eg. random writes, which were introduced in R5xx but carry big flashing health warnings from ATI over their usage).

[Zelda_fan]

So what makes this different from a PPU then?

A PPU will be faster, and less expensive.

[a12ctic]

as if one card isnt expensive enough *Rolls eyes*

[Zelda_fan]

as if one card isnt expensive enough *Rolls eyes*

$600 is quite a bit different from $300.

[pakotlar]

I'm well aware of Havok FX. It doesn't change the fact that GPUs are designed to do a very specific task, and that that task is not physics (or any other sort of general-purpose vector processing). The fact they can do it at all is all credit to them, but it says nothing about the efficiency they achieve. It would be interesting to see a comparison of the efficiencies achieved by CPUs, GPUs and PPUs, but given the level of marketing hype and secrecy surrounding the FP capabilities of GPUs and PPUs I wonder if anyone has achieved this.

GPUs also lack some features which are important in certain classes of algorithm (eg. random writes, which were introduced in R5xx but carry big flashing health warnings from ATI over their usage).

SM3.0 GPU's are extremely efficient stream processors. The ALU capabilities of the x1800+ should be fairly similar to the Ageia's vector processors going by Ageia's description of their design. GPGPU has already been demostrated to a great degree on x1800+, and this is another application that makes sense. In addition, if it is running off of a PCI-E 16x bus, there shouldn't be as many issues with latency as with Ageia's PPU. And this has nothing to do with team right or wrong, as imo there is no such thing.

[pakotlar]

$600 is quite a bit different from $300.

Going by what Havoc says, the whole point is that you won't have to spend even 300 to get a nice boost in physics demonstration. No, it won't affect the game world, but neither does Ageia's solution so far. And I get the feeling that industry will go with the lowest common denominator, which in this case, would be Havoc FX. Modifying geometry in a way that affects gameplay is not an easy thing to do anyways (in any more significant way than cellfactor does it...i.e barrels bouncing into players) and is significantly more comp. intensive than Ageia would have us believe. Anyways, so far, their solution, quite frankly sucks, and not only does it not add to the experience in any meaningful way, but it cuts performance by a good amount. Check anandtech's benchies.

[jAkUp]

I dunno about that... really a halfway decent card will cost you close to $300.

[brady]

IMO, this physics accerlation stuff is a mess right now. I'm glad that the Crysis team is just offloading all the calculations to the CPU and taking advantage of multi-core. I'm staying away from PPU's and even GPGPU usage untill the industry figures out what it's gunna stick with.

[SH64]

3 cards into 1 setup! i have enough heat with SLI now 3!!!11

interesting nevertheless.

[Marvel_us]

This physics stuff is really getting kind of confusing. Hopefully when all 3 solutions, nV/Havok, ATI and Ageia, are on the market we'll be able to see one clear cut winner.

[jAkUp]

Yea I definetely agree we need MS to step in here and set up DirectPhysics, we now have 3 different physics solutions.. its out of control.

[pakotlar]

I dunno about that... really a halfway decent card will cost you close to $300.

Jackup, we don't know how much processing power is required for the benefits of a GPU based physics acceleration to make sense. It could be a cheap $100 dollar SM3.0 capable card or a $300 dollar one, but logic would dictate that in order for Havoc FX to compete with Ageia, they will try to sell comparable benefits at a lower price point. Otherwise they won't win. And I agree with your comment about the hardware community's need for something akin to Direct3D. These proprietary standards will never get us the tight integration that a single framework will allow.

[Red_Shift]

In addition, if it is running off of a PCI-E 16x bus, there shouldn't be as many issues with latency as with Ageia's PPU.
I still don't understand why AGEIA has made a pci card as the performance probs physX card is having seem to be due to the low bus bandwidth, haven't they tested the card in a pci-e environment?

Yea I definetely agree we need MS to step in here and set up DirectPhysics, we now have 3 different physics solutions.. its out of control.
That should make most game devs jump into the same wagon. How M$ makes that step will determine AGEIA and HAVOK success. It will depend on how DirectPhysics API will be: similar to gpu language or similar to AGEIA's physics dedicated language? My guess is it will be similar to AGEIA's approach and HAVOK will lose in that case. ATI and NVIDIA are allways safe cause they're hardware vendors and because they are battling each other on the grpahics front they won't lose time, effort and money on a hybrid architecture which would allways make graphics processing less efficient. All AGEIA has to do then is make a PPU fully supporting DirectPhysics features which shouldn't be hard. Physics engines would proliferate as game dev. companies would also make their own engines (e.g. Epic). To succeed Havok would have to make an engine based on DirectPhysics and so, an engine to be run on AGEIAS's PPUs, ironic right?

[CaptNKILL]

Yea I definetely agree we need MS to step in here and set up DirectPhysics, we now have 3 different physics solutions.. its out of control.
I agree. 3 different ideas (4 if you count traditional CPU only physics) will not get us any further. It will make a huge mess of the whole gaming scene.

This is why we have standards, and the only company I can see having the resources and "power" to create one that will stick is Microsoft.

Theres a reason we dont have 10 different graphics APIs being used in game engines today. There are 2 but OpenGL is used very rarely these days. A single standard, no matter how much better or worse other companies' ideas may be is much MUCH better for the gaming industry than a crap load of competing ideas that never catch on.

Just my 2 cents.

[slaWter]

Right CaptNKILL.

But I still prefer a normal PPU and nothing on a GPU. A thrid GPU or a PPU, hello?

[mrzeld]

So what makes this different from a PPU then?i believe what ATI is trying to say is that as you upgrade, you dont have to throw out that old card, you can now use that "High-end" card from a year ago to do your PPU stuff. get a little more use out of the 500-600 dollars you spent on your high-end GPU's.

[shabby]

Physics schmisics, so far there is no killer app to showcase this. Without one no one will buy any physics cards.
Graw? Please.

[DansFace]

I can't agree with that at all! It's just way too much to buy.

IMO all you should need is ONE graphics card and a 150$ if not cheaper physics card..
Hopefully you guys agree with me on this one.. I really don't like the direction this ppu **** is going