One more thing…
AMD just announced that they are working with Havok on physics acceleration, but probably not how you thought. Nope, nothing about GPUs and Havok working together again, it seems that ship has sailed. But did you know that Havok’s APIs work with CPUs too??
We get a LOT of press releases here at PC Perspective, and not very often do we reprint them as articles, but this one deserved some analysis on our part to completely understand. And I should note that even I don’t completely understand the information we were provided – but let’s look at the actual press release first.AMD and Havok to Optimize Physics for Gaming
AMD and Havok announce plans to optimize physics processing on the AMD platform
SUNNYVALE and SAN FRANCISCO, Calif. — June 12, 2008 — AMD (NYSE: AMD) and Havok today announced plans to jointly investigate the optimization of physics effects utilizing AMD’s full line of products.
With over 100 developers and 300 leading titles already using Havok’s physics engine – Havok Physics – the company has clearly defined its position as the leading developer of game physics. By working together, both companies are demonstrating their commitment to open standards and continued support for the needs of the game community.
“As the complexity and visual fidelity of video games increases, AMD wants to take advantage of opportunities to improve the game experience,” said Rick Bergman, senior vice president and general manager, Graphics Products Group, AMD. “By working with the clear market leader in physics software, AMD can optimize our platforms to consistently deliver the best possible visual experience to the gamer.”
Havok Physics scales extremely well across the entire family of AMD processors, including quad-core products such as the AMD Phenom™ X4. As part of the collaboration, Havok and AMD plan to further optimize the full range of Havok technologies on AMD x86 superscalar processors. The two companies will also investigate the use of AMD’s massively parallel ATI Radeon GPUs to manage appropriate aspects of physical world simulation in the future.
“The success of Havok as a cross platform software company is predicated on our willingness to listen to the needs of our customers,” said David O’Meara, managing director of Havok. “The feedback that we consistently receive from leading game developers is that core game play simulation should be performed on CPU cores. The clear priority of game developers is performance and scalability on of the CPU. Beyond core simulation, however, the capabilities of massively parallel products offer technical possibilities for computing certain types of simulation. We look forward to working with AMD to explore these possibilities.”
Today’s game experience demands a balanced platform approach, one that combines the right CPU and GPU horsepower. A number of game aspects, including advanced physics processing, are optimized for CPUs, while the latest DirectX 10.1 games demand the latest GPUs. AMD is striving to deliver the best of both worlds, with highly capable CPU and GPU technology that partners like Havok can use as their canvas to deliver the best experience possible.
“Physics is an important part of today’s game experience, and in the quest for greater realism the usage and requirements of physics computation is expected to increase,” said Dean McCarron, principal analyst, Mercury Research. New technologies that improve physics simulation will enhance the gaming experience.”
(All emphasis is mine.)
During a phone briefing with AMD on the 11th, this “Havok announcement” was one of the keys discussed in AMD’s compute strategy going forward. In reality, this press release and AMD’s own statements to me make very little sense as they are presented.
First, AMD mentions that Havok is market leader for physics technology in today’s games and this is a fair statement. Even the AGEIA technology is well behind the number of game engines and technologies that utilize the Havok software; over 250 titles on the shelves today for PC and consoles depend on Havok for their physics and other simulations with 135 more coming soon. So yes, partnering with Havok in some form or fashion makes a lot of sense even when one of their key competitors, Intel, purchased Havok last year.
Force Unleashed – an upcoming title using Havok’s technologies
Also keep in mind that the first AMD executive quoted in this story is from the Graphics Products division of AMD – essential all the ATI technologies that exist in the company. The next paragraph discusses how the entire family of x86 superscalar Athlon and Phenom processors scale very well with the current Havok software and that both companies are working closely to make sure that more optimizations are made to improve performance on AMD’s CPUs. Again fine, this makes sense, and we have nothing but respect for these two companies making an effort to cross the divide and shun politics to develop better experience for the gamer.
It’s this next line that really puzzles me: “The two companies will also investigate the use of AMD’s massively parallel ATI Radeon GPUs to manage appropriate aspects of physical world simulation in the future.” So…what they are saying is that AMD and Havok WILL look into ways that the GPU can help with physics simulations; NOT that they already have investigated or that they successfully found some areas that can utilize Havok on the GPU or anything like that at all. Instead, we are supposed to gobble up a news release that says ‘we’ll get to it, maybe, if it makes sense.’
With that out of the way, some of what AMD is saying after that is relevant: most of the core gameplay physics and simulations are going to be put on the CPU for some time to come. We have analyzed the problems with a dedicated physics processor many times before and always come to the conclusion that there is inherent latency between the CPU and the PPU/GPU that current prevents such a device from really affecting “game play” physics. That could change as hardware and software improve, but for today AMD’s assertion that the core physics will remain on the CPU is accurate. And thus, announcing that Havok and AMD are collaborating for optimization is good – but you don’t seriously think these same optimizations aren’t going to be made for Intel’s line of Core 2 and upcoming Nehalem processors?
Is this the real physics power house?
This entire AMD ‘announcement’ seems very contrived and meant to add in a “me too!” to the various announcements coming from Intel and NVIDIA. NVIDIA has made no secret of the point that they are going to be implementing AGEIA PhysX technology on their GeForce 8-series and upcoming GT200 graphics cards courtesy of CUDA, their general-purpose computing software architecture. And when Havok was purchased by Intel in September of 2007, it was assumed that AGEIA would fall to one of the competitors, NVIDIA or ATI/AMD and because AMD fell short on that they are now the only of the three major compute companies to not have a physics “partner” in this very specific definition. AMD doesn’t want its customers (or worse, the market) to think they are falling behind in that realm so this ‘news’ makes it out.
In truth, Intel sees a benefit from AMD and Havok working together that at first might not seem obvious. If the combined Intel/AMD optimized Havok engines could completely wipe out the acceptance of NVIDIA’s PhysX API from AGEIA by dominating market share, then any advantage that NVIDIA might have had on them before with their acquisition would be nullified. Of course, if that were to pass, then AMD would be in a similarly tight spot when Intel’s Larrabee architecture was revealed with much better/faster Havok support than either of AMD’s computing technologies could offer. It’s a classic Catch-22 for AMD though it would appear they’d prefer to side with the “here and now” of Havok than the “then and maybe” of GPU-accelerated PhysX.
During the conference call it was asked why AMD has chosen to not implement CUDA support on their GPU, thus allowing the AGEIA PhysX technology to run on Radeon cards as well as GeForce cards. AMD’s reps stated that they were “willing to work” with NVIDIA on the issue, but that they were not approached by anyone at the company expressing interesting in making the standard truly “open” – instead it was insinuated that it was just NVIDIA lip-service trying to appear to be the consumer-friendly combatant in this fight. Havok on the other hand, approached AMD, I was told, about this relationship and because Havok is the clear market leader in physics software today the decision AMD made is pretty logical from most angles. That doesn’t mean we wouldn’t LOVE to see CUDA running on AMD’s GPUs – I am currently crushing on the products that NVIDIA is enabling with such a programming paradigm though AMD should also have something along the lines of a fully open-standard API of their own soon. That is a talk for another day though…
I don’t mean to completely dismiss this information from AMD and their continued partnership with Havok in the world of physics, but it is incredibly interesting to parse through the marketing data and dig down to what is really happening at these computing giants. We are glad to see ANY kind of cooperation between them and the fact that AMD and Havok appear to be “playing nice” is welcome news – many assumed that after the purchase by Intel Havok would begin “locking AMD out” of critical performance issues, etc. And that may be the real message behind this press release; that AMD is still alive and well in the CPU processing world and gamers can count on future revisions of the Havok’s products to utilize AMD’s x86 processors to their fullest extent.
And that, I think all of our readers can agree, is a good thing.
Be sure to use our pricing engine to find the best prices on NVIDIA and AMD graphics cards and anything else you might need: