Interview: Nvidia's Carillo On The Physics Revolution

Author: by Chris Remo, Leigh Alexander

In February 2008, graphics card maker Nvidia acquired Ageia, with the intent to integrate its PhysX physics technology onto its graphics cards. Now, Nvidia hopes the hardware-enabled physics approach will proliferate, supported by its own GPU, as it gently steers away from the PPUs of old. Nvidia's Tony Carillo says he didn't expect hardware-enabled physics to catch on so quickly. Warmonger, one of the earlier PhysX-enabled titles, was "basically a tech demo," he says -- publisher NetDevil later said it wanted to take a second look at that title's Ageia hardware exclusivity. When Carillo first encountered PhysX three years ago, "There [were] lots of titles that [were] being developed without hardware support, but since they didn't have a dedicated hardware, they [were] limited in how much physics they put in there. So to be honest, I didn't see it going very far," he says. But Nvidia has since made getting behind PhysX -- which includes a software SDK as well as the hardware card -- a priority. "They've dedicated so many resources to make sure that everybody has what they need to create PhysX titles," says Carillo. "We've got a list a mile long of developers who are jumping on the bandwagon in order to make PhysX titles and create physics titles from the ground up." Since then, Realtime Worlds has announced it's using PhysX for its upcoming MMO APB, while GRIN also employed it in Bionic Commando. Other hardware companies have offered their vote of confidence as well; earlier this year, NaturalMotion hitched its Morpheme animation engine to PhysX's star, joining up to improve integration, while more recently Emergent integrated the tech into its Gamebryo engine (Ageia was previously an Emergent partner). "Everything at Nvidia now is focused on PhysX and giving developers the tools that they need," Carillo says. "We're trying to get developers to create PhysX titles, and that helps us by selling cards, because it needs more processing power, and you need a better card to go through with that. To be honest, there's all these huddles out there that are using it ground-up for physics." Does that mean Nvidia will back-burner the PPU side of the equation in favor of working directly with the GPU? That depends, Carillo says. "If you've got a low-end GPU, you can still have a PPU in there and get a decent framerate with the graphics. and then use the PPU for PhysX." But Carillo says Nvidia's no longer pushing the PPU. "We're going to support it for I don't know how many driver releases -- maybe a year or two or something like that -- but we're not pushing sales on it," he says. "It's so much better and faster on a GPU. You can get the minimums with a PPU, but it just runs so much better on a GPU." Carillo expects that the hardware-enabled physics battle will heat up, once games reach a threshold visually -- meaning that physics, not graphics, will become the major deciding factor in realism. With that in mind, how does he see industry standards shaping up? "The CUDA SDK and the Nvidia PhysX [are] all free. There's nothing stopping anybody from going in and making it work on an ATI card. It's up to them. There's no way to stop them. However, ATI and AMD may work with Havok, and make something work on their GPU," he says. Carillo claims that a recent Intel demo simulated 50,000 particles running at 15 frames a second -- by contrast, he claims, Nvidia with PhysX can run a fluid simulation of the same number of particles at 250 frames a second. "PhysX just works better on dedicated hardware than it does on a CPU," he says. "So in my opinion, they have to go in that direction." So in his view, the necessities of the technical situation will dictate the direction of the industry -- "A game that works across all GPUs... I couldn't answer that. It depends on who makes what available." Carillo notes that competitor ATI, for example, could ensure that games made on their tech stay on their tech, but it's too early to tell how the space will shape up. Eventually, however, Carillo expects they'll settle out, particularly as the PC prioritizes both ease of development and user compatibility to stay competitive against consoles -- limiting developers' options is no path to success in the current climate. "That's why we had so little traction when it came to PhysX processing, because you can't make a game that's tailored for PhysX that's going to run great on a PhysX card but is going to run really crappy on a CPU and we only have 10,000 cards out there," he says. These days, it's intended to be different: CUDA allows processes like PhysX to be run directly on GPUs, and the benefits are huge. "Consoles, there's hundreds of millions of consoles and GPUs, and with the 8000 series and above, there's 100 million people that PhysX will now run on." That userbase didn't exist with the PPU, Carillo says, but with the GPU, it opens options -- " A game developer's never going to say, 'I want my game to only run on the Nvidia 8000 series.'" "They're never going to want to do that, because it's limiting how much of their game is going to get out. If anything, when more games become available, they're going to insist that it's going to be on a broader-range type thing. But it's going to be the GPU, as where the PhysX goes."