ATI's Take On Physics March 23, 2006 Brandon Sandman Bell

Introduction

Page:: ( 1 / 3 )

Earlier this week NVIDIA made quite a splash with their announcement that they were partnering with Havok to bring physics acceleration to their latest GeForce 6/7 GPUs. According to the PR, through Havok’s GPU-accelerated “effects physics” SDK dubbed Havok FX, Havok plans to use Shader Model 3.0 GPUs to simulate “the interactions of thousands of colliding rigid bodies, a fundamental technique of physics computation seen in today's latest games. It's now possible to compute the components of friction, collisions, gravity, mass, and velocity, that form the basis of rigid body physics.” NVIDIA made it pretty clear that Havok FX won’t affect gameplay objects, rather objects in the environment that don’t affect gameplay.

Say for instance, you toss a grenade into a room and it explodes. With Havok FX, the GPU will simulate the collisions of thousands of objects within the room once the grenade explodes (shards of glass, boxes, papers, etc).


FiringSquad: In light of NVIDIA’s announcement with Havok today on using HavokFX with GeForce cards, we were wondering if you guys had a response to NVIDIA? Is ATI working with Havok on getting Havok FX working with ATI’s shader model 3.0 Radeon X1K line? Have you looked into running this with CrossFire? Will you be working with Havok on getting Radeon support integrated into Havok FX, or do you plan on going your own way?

Will Willis: Regarding Havok, while we’re not making any announcements at the show, I can tell you that ATI will support all major physics engines. The beauty of ATI’s X1K architecture is that it gives game developers unmatched processing power for physics – it’s got everything you need to push parallel processing hard: powerful MIMD multithreaded engine, strong performance in dynamic branching, and low latency. All that gives developers the creative freedom to push the limits of realistic physics in their games. So we’re supporting the big players, and we’re delivering superior performance. As for specific cards and CrossFire, you can bet that we’re exploring all the options when it comes to physics on the GPU.

Havok itself has said that they’re a cross-platform software provider and that they don’t want to lock developers into any one specific hardware choice. They also recognize the strength in using the GPU to handle game physics. So all that said, you can probably see where this is going, but right now ATI isn’t announcing anything specific.

We will be talking about physics on the GPU behind closed doors at GDC, and if you’d like I can send that presentation to you later this week. It goes into more detail on the differences hardware can make in driving physics performance, and where ATI is aiming to take this market going forward. I’ll also make sure to get in touch as we get ready to make some announcements on this stuff down the road.

Page 2

Page:: ( 2 / 3 )

Will Willis: Physics on the GPU essentially uses the same graphics pipeline used for rendering 3D, but makes use of the pixel shaders as opposed to the vertex shaders. The 48 pixel shaders definitely contribute to the strong performance of ATI's GPU, but really it's ATI’s handling of dynamic flow control that's spurring the difference in the benchmarks.

[image]

Final thoughts

Page:: ( 3 / 3 )

As it stands now, the idea of GPU-based physics is an interesting one. It has the inherent advantage of not requiring any additional hardware to buy: that Shader Model 3.0 GPU in your PC now is all you need to get up and running. With millions of Shader Model 3.0 cards already out on the market, Havok FX has a huge installed base of potential users that any game developer would be foolish to ignore. At the same time though, the million dollar question that everyone wants to know is what kind of performance impact will GPU-based physics acceleration have on your game performance?

According to NVIDIA, in order to realize the best performance, you’ll want to have a GeForce 7600 GT at a minimum, with GeForce 7900 GT/GTX delivering even better performance, and the most optimal solution being two GeForce 7900 GTX cards running in SLI. If this is indeed the case, you’re going from an audience of potentially millions of users, down to just tens of thousands. This is where ATI’s solution could be interesting, as they’re proposing that instead of running, say two Radeon X1900 XTXs in CrossFire mode to juggle the graphics rendering/physics load, you can instead combine a $500 Radeon X1900 XT for handling graphics, with say a $150 Radeon X1600 XT card for handling physics.

On top of all this, enter AGEIA’s PhysX PPU. Will argues that while on paper AGEIA may look better, in actuality they aren’t doing anything that can’t be handled on the GPU today, simply because there are no games on the market that are truly taking advantage of all of the PhysX PPU’s capabilities outside of AGEIA’s Cell Factor game demo.

I agree that if true, this is a pretty compelling argument in favor of GPU-based physics, but the truth is that we won’t know the answer to this until later this year once more physics-accelerated titles ship. As we all know, the benchmark is going to be Epic’s Unreal Tournament 2007. If AGEIA’s PhysX card can deliver better performance than GPU-based physics, AGEIA would have a pretty compelling argument in their favor. No one knows what kind of a performance hit accelerating physics effects is going to have on the GPU’s performance at 3D rendering; the hardcore crowd isn’t going to want their GPU handling anything other than 3D, and will want a dedicated processor for physics, while more casual gamers probably wouldn’t mind the performance hit if it isn’t too significant. I also wouldn’t be surprised if Epic were to provide a custom gameplay experience for AGEIA users.

The bottom line is that it’s too soon to come to any definitive conclusions on either technology. There’s no denying the potential cost advantage of the GPU-based approach if the performance impact isn’t too significant. In this regard I really do think that AGEIA may be out of touch with the market. From the outside looking in, it really does seem to me that AGEIA’s based their entire approach around 3dfx’s original Voodoo Graphics accelerator. There’s nothing wrong with that, as the Voodoo Graphics was the pioneer of an entirely new industry. And while it’s true that Voodoo Graphics initially launched at $300, it was a niche product at that price point. Personally, I gladly paid $300 for my Diamond Monster 3D when it first came out, but Voodoo Graphics didn’t really take off until:

1. Memory prices came down. Remember that the original Voodoo Graphics required 4MB of memory
   
2. Additional board partners came onboard. 3dfx initially launched with only two board partners, Orchid and Diamond Multimedia
   
3. GL Quake. Before GL Quake debuted, there were only a handful of titles with 3dfx support. Most notable of them was Tomb Raider. Once id finally released GL Quake, 3dfx finally had a killer app that was relevant among the hardcore crowd: I was the buzz of my college dorm once I booted up GL Quake for the first time. Everyone who was into PC gaming dropped by for at least one demonstration. GL Quake looked better than VQuake too, and thus 3dfx’s Voodoo 1 was a more justifiable purchase (remember that 3dfx’s card sold for considerably more than Rendition’s).