Any word on matlab acceleration?

Sorry if this has been asked before. (A quick forum search for matlab turned up nothing.)

I have e-mailed ageia asking about possible matlab acceleration, and gotten no response. All I want to know is if there are plans to use the vector porcessing power of physx to accelerate matlab. Even just long term plans. I'm actually not that hung up on the environment being matlab or not. That's just what is most accessible (and portable) to me right now as a math grad student.

If PhysX stays around for more than 3 years, I guess it's inevitable that tools will be adapted to use the card in some mathematical programming environment. After all, the API is available; it should just be a matter of adapting it to the compiler in question. I just wonder if ageia will work to actively help this process out or not.

That's going to be the major point of interest for me. Also, just imagine the product placement they'd get if they got PhysX cards installed in university math/CS/eng labs across the country. You can't get better advertising than that. After all, we know what university computer labs are used for...

[edit]

The more I think about it, the more I'm convinced that getting real power out of a PhysX card would require using it in a real programming environment as opposed to as interpreter (like matlab). The again, writing bits of C (or whatever) to plug into matlab to do customized vector tasks should be doable. That's purely speculative, but I can't imagine it'd be too hard.

I doubt that Ageia will actively develop for matlab.

Consider that there is virtually nothing in terms of 'science community specific' hardware, probably due to the fact that the applications are very much non-standardized and often times it appears smarter just to throw more CPU power at the problem.

This does not mean that I think that the academic community will not develop a Matlab plugin to harness the PhysX power.

drizzt81 said:

I doubt that Ageia will actively develop for matlab.

Consider that there is virtually nothing in terms of 'science community specific' hardware,

Click to expand...

You mean other than pretty much the entire supercomputing industry - Cray et. al.?

probably due to the fact that the applications are very much non-standardized and often times it appears smarter just to throw more CPU power at the problem.

Click to expand...

That's not really true. There is a huge gap in computing capabilities when it comes to large vector calculations. More generic CPU power can't help much when dealing with very large vector claculations. However, a true vector processor can. Historically this has been the exclusive domain of supercomputers, but Ageia is trying to develop a niche market for vector processing on the (gaming) desktop. I see this as a potential for new kinds of computing power to be available that once were the sole domain of supercomputing.

This does not mean that I think that the academic community will not develop a Matlab plugin to harness the PhysX power.

Click to expand...

That's what I see as probably happening.

Also you got to think Agiea's card isn't ran for exact motion and gravity so you can't expect as much acceleration from the card then it would provide in games.

I asked Manju about this at the last warfactory science center lan. He said that they want to get it out and operating with games first, but scientific applications are being considered, and will likely be supported in future driver or hardware.

[RCKY] Thor said:

I asked Manju about this at the last warfactory science center lan. He said that they want to get it out and operating with games first, but scientific applications are being considered, and will likely be supported in future driver or hardware.

Click to expand...

By Manju you mean the CEO? That's good to hear.

That makes sense from a business perspective. The matlab connection won't really help their initial market positioning. They need to build critical mass quickly, and keep the company focused on that one target. It wouldn't make sense to start a side project of this nature at this point.

It's good to know that it's in the longer term strategy, though. Hopefully it's in the mid-range strategy: <2 years would be nice.

nonlnear said:

That makes sense from a business perspective. The matlab connection won't really help their initial market positioning. They need to build critical mass quickly, and keep the company focused on that one target. It wouldn't make sense to start a side project of this nature at this point.

It's good to know that it's in the longer term strategy, though. Hopefully it's in the mid-range strategy: <2 years would be nice.

Click to expand...

Bearing in mind that this is unsubstantiated rumourmill churn, but...

Everyone's understanding right now is that Ageia is only directly supporting games for the time being. HOWEVER, they're also licensing the API to anyone who wants it. That said, the rumourmill says that Dassault is talking to Ageia about it - Dassault being CATIA and SolidWorks among others. I've also heard some rumblings that Autodesk might have it in AC2K8. There are still further rumblings, but I'm not willing to disclose them as they may anger people I have to work with. The big rumour though is that Dassault and/or Autodesk do have some PhysX's in house already - I won't ask anyone to confirm or deny there. I don't know myself.

Suffice to say, while Ageia is very much focusing their strategy on games at current, yes, there is a lot of potential in the professional arena, and they are looking. With Creative basically shutting down 3DLabs, a lot of CAD/CAE/CATIA/GIS/CGI developers are being forced to look at options other than the video card. nVidia and ATI does not cut it in this arena, period. (8 lights? Come on.) The PhysX's capabilities are flexible enough that it has the potential to fill these gaps. Whether it can or will though, remains to be seen.

AreEss said:

Bearing in mind that this is unsubstantiated rumourmill churn, but...

Click to expand...

Thanks. That was a fun read. From the lack of meaty info available on the matter, I don't expect anything better than unsubstantiated rumor mill churn. So in my estimation, that's as good as it gets at this point.

You're looking at the wrong place.
Apparently, this coprocessor: http://www.clearspeed.com/
accelerates MATLAB from the infos on their site.

MissingGhost said:

You're looking at the wrong place.
Apparently, this coprocessor: http://www.clearspeed.com/
accelerates MATLAB from the infos on their site.

Click to expand...

That's an interesting product, but that is prroof that I'm looking at the right place.

The Clearspeed board ggivees 45-55 GFLOPs for AFAIK somewhere between $1000 and $10 000. (This is all I could find for pricing info.)

An Ageia board has (I think) ~100GFLOPs* of vector crunching power for $300. The numbers seem pretty cut and dry to me. If Ageia bothers to write a solid set of drivers for the PhysX card to accelerate matlab (and similar packages), they could find themselves beign the new kings of workstation coprocessors.

* If anyone has better info, please enlighten me. I can't find much online right now that looks authoritative. I see 100GFLOPs being talked about on forums, but I can't find a source.

I just posted a link over here from some guy who developed a PPU-accelerated normal mapping tool. So it's definitely possible to do a lot more than game physics even without official support.

Ok, I'll ask the "stupid" question.

have you asked the matlab people if they intend to persue support for the physx card? That's probably the more appropriate folks to ask.

raz-0 said:

Ok, I'll ask the "stupid" question.

have you asked the matlab people if they intend to persue support for the physx card? That's probably the more appropriate folks to ask.

Click to expand...

Not a stupid question at all. I never got a response. I really ought to keep trying though.

LuminaryJanitor said:

I just posted a link over here from some guy who developed a PPU-accelerated normal mapping tool. So it's definitely possible to do a lot more than game physics even without official support.

Click to expand...

Thanks much, Janitor. I was wondering who might be first. Haven't got a working PPU here right now, but indeed looks quite nice.
As far as 'official' support, that's still hazy. They're pushing the SDK pretty much free, including support, so it's a little hard to be certain who does and doesn't have access and support right now.

Honestly the matlab folks are pretty lame. They won't even add multithreaded support to matlab, so all these enginerds running around with 4+ core workstations don't get any benefit. I don't see them adding PhysX support, it's going to have to be something either done by the community or by Ageia