Hi people,

I've been hearding about physics processin alot Nvidia ATI and Ageia are producing solutions.
Since i'm interesting in buying an card from ageia when its comes to cost about $200
I ask:

Did boinc will use the extra power for its mathemathics calculations?
If not i suggest you add in the capbility becou evem if its not a mager sucesess if 1000 boincs user buy them it will empower a lot the system!

---

Hi people,

I've been hearding about physics processin alot Nvidia ATI and Ageia are producing solutions.
Since i'm interesting in buying an card from ageia when its comes to cost about $200
I ask:

Did boinc will use the extra power for its mathemathics calculations?
If not i suggest you add in the capbility becou evem if its not a mager sucesess if 1000 boincs user buy them it will empower a lot the system!

Seti or Boinc or any other Boinc application won't use this extra computing power, because it would need a redesign and a reprogramming of these applications. I don't think that the PhysX card would be a big improvement for crunching times, because PCI is too slow and hasn't the bandwith for a lot of these computations. Even PCI Express 1x wouldn't help here much more.
In the case of graphics cards something like Havok FX isn't needed. This would only make programming easier. There have been some efforts for gpus to be used for general purpose computations (you can visit gpgpu.org for more information). But the main problem is that there aren't any fully completed libraries which support programming graphics cards for general purpose computations. Brook GPU is still stuck in version 0.4 for months or years now and it has only an unsufficient documentation. Sh is currently in Version 0.8rc1 available, but it doesn't provide the full functionaltity to support all kinds of general purpose computations.
hope this helps you a bit

SpaceTrucker | team #probe

Edit: A big problem is also that computations on graphics cards are difficult to be done, if they are datadriven.

Yeh but graphics cards aren't supossed to do general processing. You nedd a heavy one qhem you heavily use games, mostly of the time it should be off (and vista like OS will require some power from them).

The question is that GPU started as PCI and even them the performance increase was frytening (its spelled corectly?) And physics is just starting. Most people won't be required to have one to play games for yers to come. But if you use one can increase your performace.

Im just thinking that supporting this kind of aplication will improve performance to every one who choose to have one PPU. ANd they are in the midle of seti enchased. it will need a complete rewrite of most of the code ?

---

I had a quick look at the Ageia PhysX SDK some weeks ago. Yes the whole code which is into data analysis would have to be rewritten, since for use of the PPU some Ageia PhysX libraries must be used for programming.
But there are even more questions which should be considered if the PhysX card should be used for computations in the seti client or another scientific application. The processor used on the PhysX card is as you correctly said a physics processing unit and therefore not made for general purpose computations. This makes programming a scientific application for the PPU a bit more complicated than for just a cpu (but maybe not as complicated as with graphics cards).There are nearly no information about the architecture of the PhysX processor available (or at least I haven't found something). So it is hard to write hardwareefficient code. And another question is, if there is this information hiding about the hardware, should this be supported by the open source community by using this in an application?

SpaceTrucker | team #probe