By now, many have heard of AGEIA and its startling announcement: they will produce a processor used exclusively to process physics related computations. Called the PPU, or Physics Processing Unit, its role will be to offload highly intensive mathematics such as realistic water movement, realistic character physical reactions to objects and the world, from the CPU to a dedicated processor. This all seems like the natural progression of things, since dedicated sound, network and other processors are commonplace.

Today, however, most processors spend their time mostly idling - you're rarely ever pushing your hardware to its limits consistently. Thus Havok, a company that's well known to game developers, has announced that it has plans to do for you what AGEIA promises, but save you money and maximize your dollar spent at the same time. Indeed, Havok has confirmed with us that they are competing with AGEIA.

The Havok FX engine is what Havok claims will provide the functionality of a PPU, but its approach is entirely different than AGEIA's. What's special about Havok FX is that it's a software engine that is currently based on Havok's widely used physics engines. However, Havok FX is designed to offload many intensive physics functions from the CPU to the GPU. Using technology available in Shader Model 3.0 and beyond, the Havok FX engine will be able to take advantage of unused resources of today's powerful GPU's and put them to use.

Many games today are intensive, but while they don't use a lot of what your graphics card is capable of doing (features, etc.), they do stress the majority of GPUs that are out there - often to their very limits. For example, a given scene can be utilizing 80% of GPU rendering resources (bus, memory bandwidth, etc.) while another scene uses 20%. How Havok aims to utilize the resources in a GPU to accelerate physics calculations remains to be seen. Havok did mention to us however, that they believe having two GPUs would be better suited to handle such duties as load balancing would be the key issue. Microsoft's Shader Model 3.0 requires compliance for full programmability, so with today's DX9 compliant GPUs, it's entirely possible for Havok FX to program a Radeon X1800 or GeForce 7800 GT (and beyond) on the fly, with specific physics processing instructions.

Havok also pointed out to us that its Havok FX engine will allow a Shader Model 3.0 compliant GPU to accelerate "game-play" physics and not only the resulting visual effects of such physics, which Havok says, AGIEA's product only does. Havok explains that its engine is able to offload such physics operations as collision detection, which on today's general purpose GPUs are very slow to compute. We spoke to Havok and they said:
"It is definitely the case that load-balancing is a key challenge for both effects physics and graphics. Enabling effects physics via the GPU offers much greater flexibility for addressing that type of problem versus a proprietary physics hardware device that will inevitably sit idle while the GPU may be overtaxed. We believe that two GPU's stand a far better chance of collaborating more effectively."
One important fact to keep in mind is that Havok is playing a significant role in the development of console technology, and has not specifically stated that the PC desktop is the ideal platform for Havok FX. Consoles have the advantage of being a closed system, in which dedicated solutions such as Havok FX will have long term applications. It will be interesting to see if Havok can achieve this and how well.

Havok's physics engines are featured in many of today's games, including F.E.A.R., Age of Empires 3, Brothers in Arms: Road to Hill 30, Halo 2 and many others. The full list can be found here on Havok's website.



View All Comments

  • realphysics - Thursday, November 10, 2005 - link

    It is easy to jump to the bandwagon, but don't be hasty - those who create games develop and publish them have two issues related to physics. One: how much they have to pay companies such as Havok to get that physics toolkit..It isn't free folks.. Havok gets paid for every game for using their physics toolkit. Ageia has an SDK for physics which is FREE. There goes the economics for the game developers. Argument number two: Physics can never be AS REAL,VIVID OR IMMERSIVE or as envisioned or fancied by the game developers, or those who enjoy a gaming experience when the game itself is limited by the hardware that executes it (remember CPU and GPU were designed and developed more than a decade ago at least - with a completely different purpose and still serve those purposes). Don't confuse the issues. If you want mediocre physics - go for what you have in the market, if you want real physics and experience go for Ageia. You get what you pay for. Remember, in the future those who create, develop and publish games are going to realize they have even greater potential to express their creativity with the physics offloaded to hardware than can really perform and the SDK they use for it is FREE - not something that Havok offers - Go figure Reply
  • Nighteye2 - Thursday, November 03, 2005 - link

    Havok FX vs AGEIA is like integrated graphics vs a video card. I expect developers will use Havok FX as a substitute for gamers that do not have an AGEIA card.

    I find it hard to believe that a GPU already stressed by dozens of other things to calculate could outperform a dedicated PPU. Keep in mind that scenes with lots of physics have lots of moving components in them and those cast lots of shadows, so the scenes with the most physics calculations are likely to also be the ones that tax the GPU the most - even without cramming some of those physics calculations unto the GPU.

    You'll probably see that the highest physics level in games will start to require an AGEIA PPU, just as some detail modes once required a 3DFX chip.
  • SynthDude2001 - Wednesday, November 02, 2005 - link

    I like the idea in general, but I worry about performance. First of all, how would this perform compared to the discrete PPU (=how well would the GPU perform these types of calculations and how fast); and how will this affect 3D rendering? We don't know much about how this works right now, but you'd think that the computing cycles for these sorts of heavy calculations have to come from somewhere...the fact that they even admit that this would be better on dual GPU systems worries me. Unless you can use your old video card as a 'dedicated GPU-PPU', I'm definitely worried about performance. And since they say SM3 is required, only those upgrading from 6600/6800 cards will be able to do that anyway. I'm wondering if I'll be able to keep my AGP 6800GT and use it just for physics with this...but at that point, it would seem to make more sense to sell it and get a real PPU for about the same price.

    In any case, like someone else said, I'm glad that another company (and not just a tiny one nobody has ever heard of) is starting to get serious and push physics processing.
  • KristopherKubicki - Wednesday, November 02, 2005 - link


    AGP 6800GT and use it just for physics with this

    Unfortunately, AGP + any sort of physics computation doesn't seem very likely due to the asynchronous bandwidth.

  • SynthDude2001 - Thursday, November 03, 2005 - link

    aha, thanks for the clarification. I suppose it doesn't make too much of a difference since I'll be upgrading the video card by next summer, and I doubt any of this physics processing stuff will surface at retail before then anyway.

    Still not sure how I feel about this, I guess it's really too early to know much of anything. I was looking forward to the stand-alone PPU, but if this idea catches on and provides similar physics performance without having to spend extra on a dedicated card...I'm all for it. I guess we'll have to wait and see.
  • tfranzese - Wednesday, November 02, 2005 - link


    Havok explains that its engine is able to offload such physics operations as collision detection, which on today's general purpose GPUs are very slow to compute. We spoke to Havok and they said:

    I do believe you meant CPUs.
  • KristopherKubicki - Wednesday, November 02, 2005 - link

    Unfortunately, GPU is correct. Collision detection on a GPU is very poor because branch prediction on a chip that essentially only calculates scalar math is not very good.

  • tfranzese - Wednesday, November 02, 2005 - link

    Within the context of the article then, that doesn't make a whole lot of sense considering Havocs position on a GPU vs. a PPU.

    It'd be great to see an actual analysis on this stuff rather than all this endless speculation about how useful it could be.
  • JarredWalton - Wednesday, November 02, 2005 - link

    I think the point is that SM3.0 makes such computations more feasible (i.e. it wouldn't have been worth pursuing in the DX8 days). Reply
  • shabby - Wednesday, November 02, 2005 - link

    Those poor poor ati users with sm2 cards... Reply

Log in

Don't have an account? Sign up now