News PPU Powered Alienware arrives

OK that was essentially what I was trying to say... although you did it a lot more elegant

 

it would be a math co-processor. thats what it is (did you not follow my link?). and i think a math co-processor could alleviate a lot of the physics calculations, considering it does math 5 times faster than the fx60 can. AMD doesnt seem to think its out of place today.

 

I would have to see a difference before I would consider it. I would think it would be better to have it on the graphics card because then it and the gpu could work directly together, no need to route data unnesscarily. And then you could sli it.

 

OK, here's the thing - I had another meeting with AGEIA this morning but their hands are still somewhat tied because they can't announce / talk about stuff until the developers of various stuff do. Like, the authorisation to discuss Ghost Recon only came the day before the dedicated pre-GDC event.

The slow flow of information, data, and screenshots is frustrating everyone, everywhere - us, you guys, the PR team etc. We will get where we need to be to evaluate this who idea properly eventually, but it will take until the launch of the individual retail cards in May before that happens probably

 

Hope it stays a separate add-on card

I'm still hoping that this PPU add-on card will be useful OUTSIDE of the gaming world. There's a very small minority of people in the world (research scientists in fields of plasma physics, CAD/FEA, Fluid Mechanics/Aerodynamics, etc) that could really use some speeded up processing power for physics.

If anybody thinks they know how to exercise a computer (yes, I'm sure most of you do...but this might be interesting to you nonetheless), I've got a plasma simulation (simulates a Hall Thruster) that takes roughly 40 hours to run on a 3.8GHz P4 (single core...but the fastest ever made, in terms of clock speed)...and in that time, it only simulates about 1 millisecond in real time. If an add-on physics card with the right command libraries could cut that down by a factor of 10, we'd be in business for sure.

Of course, if these end up embedded in high-end graphics card, we'd be looking at not just buying the $250 physics chip, but the $500 (or $1000 in SLI) in graphics cards too...and probably, at least for non-gaming users, be crippled by whatever compromises had to be made to cram the two chips on the same card (to save on cooling, share the ram, etc).

I'm hoping the folks at AGEIA might think about developing a relationship with university/lab researchers to bring us the computing power we've been dying for for years!

 

BTW, I remember reading a while back that asus had signed on to intrigrate the cards features into some future mobos, so that my be like having the math coprocessor socket was well. Personally, I'll wait until unreal tourny and the pci-e versions have been released.

 

Nvidia vs. Ageia physics

There's a pretty good article on Ars that talks about the difference between NVidia's GPU physics and that of Ageia's PPU...especially see the "update" section toward the bottom of the page:

Ars. Physics Acceleration Article

The gist of what they say is that: "Ultimately, GPU-based physics acceleration will complement, not compete with, physics co-processors like PhysX" because the physics processor will be able to talk back directly to the CPU to create effects that alter gameplay...while physics in the graphics card will be mainly for appearances. I can't personally vouch for the validity, but I think Ars is generally pretty reliable.

Also: with the requirements of talking back to the CPU...it would seem that it would not be a good idea to eventually incorporate the PPU onto the graphics card....as then the two will have to share bandwidth (along with space, ram resources, power, heat dissipation). One day, as these PPU's ramp up, they very well may need their own PCI-express x16 slot. PPU SLI anyone?

 

Mate you do realize that your graphics card doesn't to anything to help physics along. The bottle neck there is the CPU. The idea is that you can get the physics off the CPU onto the card which means you can have more physics calculations without bringing everything to a crawl.

Seriously where are they going to fit a good physics chip into a GPU aren't they big enough already? I think that the physics card will have an advantage over graphics card integrated ones for awhile at least.

 

Oh the irony