Here's an interesting approach to parallel processing:
http://www.enterpoint.co.uk/merrick/merrick1.html
They are working on a system using lots of those boards:
http://www.enterpoint.co.uk/merrick/supercomputers.html
I was a bit surprised by the choice of the Spartan-3A, but they are very cheap. The concept of hiring out the hardware is novel; perhaps XMOS could do the same with the XK-XMP-64.
Programming it effectively is going to be a big problem, of course.
FPGA supercomputer
-
- XCore Expert
- Posts: 546
- Joined: Thu Dec 10, 2009 10:41 pm
- Location: St. Leonards-on-Sea, E. Sussex, UK.
-
- Respected Member
- Posts: 377
- Joined: Thu Dec 10, 2009 6:07 pm
Assuming the XMP-64 is Ethernet-connected, it might be possible to build some software on an edge core to just load code in over the web and get results back. You then wouldn't even have to physically rent the XMP-64, you could just buy compute time on it by the (unit of time).leon_heller wrote:I was a bit surprised by the choice of the Spartan-3A, but they are very cheap. The concept of hiring out the hardware is novel; perhaps XMOS could do the same with the XK-XMP-64.
1) XMOS, is this possible?
2) Anyone else with more time who likes the idea want to give it a shot?
Once built, this could run far, far more than just an XMP-64. If it proved popular, you could use it to run a stack-load of XMP-64s.
Or the programmer/customer could just choose how many cores/threads they wanted, and the on-board handler could control the code distribution across as large a compute array as was requested.
-
- XCore Expert
- Posts: 546
- Joined: Thu Dec 10, 2009 10:41 pm
- Location: St. Leonards-on-Sea, E. Sussex, UK.
It could be hooked up to an XC-2 for Ethernet connectivity, if it doesn't have the capability. They might even sell a few more by doing that, as customers might buy one after getting their application working.
-
- XCore Expert
- Posts: 577
- Joined: Tue Sep 08, 2009 5:15 pm
I really like this idea!
Should also hook it up to a data stream from SETI to crunch away at when not being used by someone! Or some other such project that needs spare processing power to help some greater goal.
Should also hook it up to a data stream from SETI to crunch away at when not being used by someone! Or some other such project that needs spare processing power to help some greater goal.
-
- Respected Member
- Posts: 377
- Joined: Thu Dec 10, 2009 6:07 pm
How about climate modelling? Burn power modelling what happens when we burn power. ;-)jason wrote:I really like this idea!
Should also hook it up to a data stream from SETI to crunch away at when not being used by someone! Or some other such project that needs spare processing power to help some greater goal.
-
- Respected Member
- Posts: 363
- Joined: Thu Dec 10, 2009 10:17 pm
That sounds like a really neat idea.
Resesans did a similar approach for promoting. They basically had a little java remote desktop app on there website. You first booked a free slot and then it let you remote desktop in to a machine with all the software development tools installed and the PC connected to a real dev board that users could execute code on.
So a similar approach could be used to promote the other xmos kits. And would be even better if you had a XC-1 with some hardware connected to it like a LCD, switches,audio output... and have a live webcam show you the board over the internet. In my opinion it would be as close to a hands on experience as you can get without the actual hardware.(Meaby a good project to assign to a few interns at xmos)
Resesans did a similar approach for promoting. They basically had a little java remote desktop app on there website. You first booked a free slot and then it let you remote desktop in to a machine with all the software development tools installed and the PC connected to a real dev board that users could execute code on.
So a similar approach could be used to promote the other xmos kits. And would be even better if you had a XC-1 with some hardware connected to it like a LCD, switches,audio output... and have a live webcam show you the board over the internet. In my opinion it would be as close to a hands on experience as you can get without the actual hardware.(Meaby a good project to assign to a few interns at xmos)
-
- Respected Member
- Posts: 377
- Joined: Thu Dec 10, 2009 6:07 pm
Yes please.Berni wrote:That sounds like a really neat idea.
Resesans did a similar approach for promoting. They basically had a little java remote desktop app on there website. You first booked a free slot and then it let you remote desktop in to a machine with all the software development tools installed and the PC connected to a real dev board that users could execute code on.
So a similar approach could be used to promote the other xmos kits. And would be even better if you had a XC-1 with some hardware connected to it like a LCD, switches,audio output... and have a live webcam show you the board over the internet. In my opinion it would be as close to a hands on experience as you can get without the actual hardware.(Meaby a good project to assign to a few interns at xmos)
XMOS - if there aren't "people" resources to do this (ideal), would you consider sponsoring a project to set this up on one or two boards?
Would anyone be interested in building this for the XMP-64?
In exchange, the prize for completing it would be that you get to keep the XMP-64. I think that's win-win for everyone. Sound like a good deal?
-
- Junior Member
- Posts: 5
- Joined: Fri Feb 19, 2010 9:09 am
Hmm, just because you could, should you?
A Geforce 8800 gives you 370 GFLOPS for <$500 compared to 25 GIPS for $1.5K as well as bit more than 64K RAM shared between 8 threads. About 200 times more memory to be precise.
I wonder who would pay for time on a 32b integer CPU with moderate arithmetic performance and memory more suited to an embedded system, in these days of open CL.
Surely XMOS is all about deterministic computing and emulating I/O?
A Geforce 8800 gives you 370 GFLOPS for <$500 compared to 25 GIPS for $1.5K as well as bit more than 64K RAM shared between 8 threads. About 200 times more memory to be precise.
I wonder who would pay for time on a 32b integer CPU with moderate arithmetic performance and memory more suited to an embedded system, in these days of open CL.
Surely XMOS is all about deterministic computing and emulating I/O?
-
- Respected Member
- Posts: 377
- Joined: Thu Dec 10, 2009 6:07 pm
Have you programmed a GeForce 8800?cereal_entrepreneur wrote:Hmm, just because you could, should you?
A Geforce 8800 gives you 370 GFLOPS for <$500 compared to 25 GIPS for $1.5K as well as bit more than 64K RAM shared between 8 threads. About 200 times more memory to be precise.
I wonder who would pay for time on a 32b integer CPU with moderate arithmetic performance and memory more suited to an embedded system, in these days of open CL.
Surely XMOS is all about deterministic computing and emulating I/O?
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...
-
- Respected Member
- Posts: 377
- Joined: Thu Dec 10, 2009 6:07 pm
I think building this and running it over the web (free or not) would attract quite a bit of interest. If you want to try to turn it into a business model, there are several ways to do this: you don't have to charge for the privilege of using it. I can think of a few, but frankly I think this kind of system is of more use to XMOS as a promotional tool than as a business...
Which is why I suggested XMOS sponsor someone an XMP-64 for building it and getting it running.
It's easy to find problems with new ideas - because they are new. If people didn't decide to do them anyway, we'd never make progress!
Which is why I suggested XMOS sponsor someone an XMP-64 for building it and getting it running.
It's easy to find problems with new ideas - because they are new. If people didn't decide to do them anyway, we'd never make progress!