Print 42 comment(s) - last by RobberBaron.. on Jul 3 at 8:54 AM

ATI's physics goes on the back burner

During ATI's 2006 third quarter financial results conference call this morning, ATI CEO Dave Orton revealed to analysts that despite demonstrating physics processing earlier this year, volume availability will not be for another "9 to 12 months." Orton indicated that the technology was still undergoing testing and preparation. Interestingly,  ATI indicated earlier that its physics technology was already on the market and needed only driver updates.

DailyTech has attempted to test some physics-enabled demos on a pair of Radeon X1900 XTX boards but drivers were not ready. Interestingly, ATI demonstrated physics on its RD600 chipset at Computex earlier this year but a year from now, ATI's RD600 will more than likely be phased out. In fact, ATI said during the call that it was already developing and testing RD700, its next generation core logic.

After ATI's announcement about its own physics solution during Computex, AGEIA responded by indicating that ATI's method was crude and did not really address the core needs of physics processing. AGEIA also indicated that its own widely accepted technology was already available on store shelves. AGEIA's availability argument was somewhat called upon by ATI customers, citing that ATI too had boards already in the market. Orton's announcement today unfortunately reaffirms AGEIA's point. Many ATI customers will have to wait a while longer.

Details on ATI's R600 core -- the replacement for its X1900 family -- is expected to be available late this year or in early 2007. It is unclear whether or not ATI's physics API will continue to support the X1900 family a year out from now.

Update 06/30/2006
:  We earlier reported that "retail availability" was not to be available for 9-12 months.  Instead, Orton's exact meaning was for "volume availability," and as such DailyTech has changed the article text. Will Willis, Senior PR manager for ATI sent us the following comment about ATI's physics solution:
Dave’s reply was that we won’t see physics in volume (i.e. hundreds of thousands of cards being used for physics) and therefore a material revenue stream from physics for 9-12 months (i.e. another 3-4 quarters) as that’s when there will be some substantial volume of ATI graphics cards being used for physics purposes.

ATI GPU’s that support physics acceleration are already in retail (i.e. the X1900 and X1600 series of cards for example).  All that’s required is a driver update that enables physics processing, and more importantly, game content which we expect around the holidays.  We’re very likely to see revenue from physics before 9-12 months, but as Dave said it won’t be in volume, it will likely be early adopters.
Willis claims content for ATI physics will be available this holiday season already.  However, there is no word yet on the driver other than it is "several weeks out."

Comments     Threshold

This article is over a month old, voting and posting comments is disabled

RE: Slow And Steady Wins The Race
By Trisped on 6/29/2006 11:48:00 AM , Rating: 2
Microsoft has never actually said they are creating a physics API. For all we know they might have been covering their bases with that job posting, or they could still be looking for someone to fill it. Even if they are working on it, it will be at least 3 months after the leak that Microsoft confirms it, and at least a year before cards that support it start to ship. ATI is not going to put its cards on hold for the Microsoft API.

As I have said before, PPU and GPUs are a large number of cores on one die. For example, x1900s are 64 cores in 4 core groups (for the 1:3 pipeline ratios). Most multi core processors are dual core (2 cores), with at least one reserved for the main game code, and the other partially used by system resources to manage data transfers etc (unless they game is properly threaded, in which case all cores are going to be pushing the game engine). In most cases you will not have sufficient CPU power left to quickly process a large number of dependent calculations necessary for complete physics.

RE: Slow And Steady Wins The Race
By bersl2 on 6/29/2006 1:42:18 PM , Rating: 2
SIMD instructions on CPUs only go so far. If we had a general calculation library that abstracted the mass calculations needed from any specific chip... but of course, we can't, because we don't have the documentation necessary to create such a thing.

RE: Slow And Steady Wins The Race
By saratoga on 6/29/2006 2:05:00 PM , Rating: 2
Conroe and the K8L should roughly double the vector throughput per clock of the K8, single threaded. Add to this quad core and you have another 4x. That works out to an 8 fold increase over the current single core, A64/P4 generation, assuming no clock speed increases. Couple with large caches, lightening fast access to system memory and single digit latency for interprocess communication, I think this will more then give dedicated PPUs a run for their money.

The biggest issue though is going to be the bus. I don't know how fast Ageia's future PCI-E cards will be, but thats going to seriously limit the range of problems they can be applied to because at a certain point its just faster to let a local core do it then wait hundreds to thousands of clock cycles for a PCI-E device to do it.

RE: Slow And Steady Wins The Race
By Hydrofirex on 6/29/2006 6:42:04 PM , Rating: 1
Reading the reviews of the Ageia PPU I start to think that quadcore processors, more cache, and better system performance overall might actually manage to make a PPU worth it! As it is now they're horribly bottlenecked by current systems. I'm sure the coding will get better, but if you think about it all these dependent and related calculations - they must start to bump up overhead.

Just my opinion, but PPU's will come in (Come on, they make way more sense then spending ~$1000 for 2 VGA's!) - The real question is if they'll be intigrated into some other component or will systems get so fast processors can just handle it. I would wager intigration in the mid-run. I'm sure everyone's happy to let Ageia be the 'front-guard' here.

RE: Slow And Steady Wins The Race
By saratoga on 6/30/2006 8:28:43 PM , Rating: 2
Reading the reviews of the Ageia PPU I start to think that quadcore processors, more cache, and better system performance overall might actually manage to make a PPU worth it! As it is now they're horribly bottlenecked by current systems.

I don't think you understand what a PPU does. It offloads functions from the system. The system doesn't bottleneck the PPU since you're not interested in PPU performance, but rather are interested in overall system performance. What actually happens is the faster the system gets (particularly in vector performance), the less interesting a PPU gets simply because you don't get such a large advatage from offloading things. As more and more cores become available, the utility of coprocessors in general will drop.

“We do believe we have a moral responsibility to keep porn off the iPhone.” -- Steve Jobs

Related Articles
AGEIA Responds To ATI's Triple Play
June 9, 2006, 6:00 AM
ATI GPU 2006 Roadmap
June 6, 2006, 3:20 PM

Copyright 2014 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki