FPGA supercomputer

Off topic discussions that do not fit into any of the above can go here. Please keep it clean and respectful.
User avatar
___
Member++
Posts: 30
Joined: Wed Feb 03, 2010 5:04 pm

Post by ___ »

jonathan wrote: Have you programmed a GeForce 8800?

These all sound like arguments for why you shouldn't bother building an XMP-64 in the first place.

Don't get too bothered by who would pay for it... (you could just to it free of charge). Instead think about whether it would be a useful, interesting demo that might attract attention to the technology, the company etc.

I can think of many interesting things to do on this that would be a pain in the ass on the GeForce. It's not all about peak GIPS...
The programming model for using GPU's for scientific computing is actually pretty well thought out and takes into account the fact that you need to get the data where it needs to be fast. It also fits pretty nicely into the fact that applications are mainly x86 based in this area.

You are are correct about it not being about peak GIPS ..... in this area its all about peak GFLOPS.


User avatar
jonathan
Respected Member
Posts: 377
Joined: Thu Dec 10, 2009 6:07 pm
Contact:

Post by jonathan »

:roll:

In summary: you aren't sure why you might pay to use it so it's not worth doing. :lol:
Image
User avatar
otitov
XCore Addict
Posts: 207
Joined: Thu Dec 10, 2009 11:00 pm
Location: Mexico
Contact:

Post by otitov »

@jonathan i am interested.
User avatar
jonathan
Respected Member
Posts: 377
Joined: Thu Dec 10, 2009 6:07 pm
Contact:

Post by jonathan »

"cereal_entrepreneur" and "___"

Sorry - are you guys from XMOS? Hard to tell. Previous posts suggest that you are! But if you are, I'm not sure why you would be suggesting that the XMP-64 is a useless kit...?

@otitov Cool.
Image
User avatar
daveg
Member++
Posts: 28
Joined: Thu Dec 10, 2009 7:25 pm

Post by daveg »

Sounded more like a suggestion that things which require heavy floating-point maths applications are not suitable for an integer-only architecture to me...
User avatar
Berni
Respected Member
Posts: 363
Joined: Thu Dec 10, 2009 10:17 pm
Contact:

Post by Berni »

That sounds like a excellent idea. You have to consider tho that this project requires a wide range of knowledge to implement. First you need some kind of login based php or similar website for booking time on it, then you need some server and client software to acualy establish the connections and in the end also the boot loader for the dev board.

The quickest aproach is proabobly to just use big fast PC and run a few vmware virtual machines and then each virtual machine have its own XMP on it. Virtual machines are very useful for this as they allow multiple systems run on one pc and it dose not give you direct accesses to the physical machine while providing a covenant restore to the original state after log off in case someone did a bad thing to it. Its also possible to run Win Mac or Linux so the user can chose what system they want. But users will want to run the IDE on there machine too so its not the best solution. Its more useful as a demo for new users as they can try it out without having to download anything and on real hardware too.

Oh and yeah nvidia makes those Tesla supercomputing cards for PCs that are ridiculously fast but xmos is not really aiming at making a supercomputer. I would absolutely love to try one of the kits over the web when i first found out about Xmos. It can instantly give you the impression if its something for you almost as you would bring one home and try it for real. Since you need to actually buy it to try it at home. A lot of people will take it too much of a risk to buy a XC-1 for the prince only to meaby find out its not really what they expected, especially cause its a completely new architecture (If you buy like a ARM dev board you know what your gonna get).

btw i think this web based stuff should be free. Altho you have to put some limitations on it like every user getting like 10 or 20 h of "play time" per week so they cant just take all the slots. Then make extra time or reserving large amounts of time cost a few bucks so if someone really wants a lot of time they can have it or even better if a university decides to teach programing on these boards without even buying them(Since they would need large amounts of time reserved on a specific schedule).
User avatar
___
Member++
Posts: 30
Joined: Wed Feb 03, 2010 5:04 pm

Post by ___ »

jonathan wrote::roll:

In summary: you aren't sure why you might pay to use it so it's not worth doing. :lol:
Not at all, just raising the point that an embedded microprocessor with limited memory and no floating point hardware is unsuitable for areas of scientific computing such as climate modelling. Although an integer based climate model would make it simpler for everyone.
User avatar
jonathan
Respected Member
Posts: 377
Joined: Thu Dec 10, 2009 6:07 pm
Contact:

Post by jonathan »

daveg wrote:Sounded more like a suggestion that things which require heavy floating-point maths applications are not suitable for an integer-only architecture to me...
But this totally misses the point. Everyone here knows what the XMP-64 can and cannot do. And I'm not convinced anyone suggested running floating-point scientific computations on it (except possibly in jest).

The proposal is to hook it up to the web and make it possible to distribute code across it through this web interface. That's the interesting part. Focus on the interesting part, please.

If you don't know what you would use it for, that's fine! Others do. :-)
Image
User avatar
Berni
Respected Member
Posts: 363
Joined: Thu Dec 10, 2009 10:17 pm
Contact:

Post by Berni »

Or someone could use it to test out there computer vision project and see if xmos is a good choice, so they can use a xmos chip or a few in the final design insteed of a big powerhungy PC.
User avatar
jonathan
Respected Member
Posts: 377
Joined: Thu Dec 10, 2009 6:07 pm
Contact:

Post by jonathan »

___ wrote:
jonathan wrote::roll:

In summary: you aren't sure why you might pay to use it so it's not worth doing. :lol:
Not at all, just raising the point that an embedded microprocessor with limited memory and no floating point hardware is unsuitable for areas of scientific computing such as climate modelling. Although an integer based climate model would make it simpler for everyone.
What seriously? I was joking about the whole concept of using powerful computers that burn power to model climate change etc. Ack never mind.

:arrow: :arrow: :arrow:

Moving on.
Image
Post Reply