Xciter at 250,000

[pythagoras]

Thanks very much for the boost Xciter, it is appreciated by all the team members :toast: Only another 250,000 to go, to take down Angra :D

Only joking, you have done more than anyone could ask. On behalf of the team: Great Job my friend.

Regards

John.

[Ackbar]

Xciter needs to change his avatar... :D

[Ugly n Grey]

LOL...change Avitar :)

Thank you Xciter, the rush has been wunderbar for cert!

:toast:

[mtb856]

Congrats bud :)

I just passed that milestone as well, although for another team.

[Ugly n Grey]

he did it in about 2 weeks ;)

[twiggy]

:toast: Xciter!

Incredible power you have! :fact:

[Ackbar]

:toast: Xciter!

Incredible power you have! :fact:

"The Force is with you, but you are not yet a Jedi..."

[Xciter]

Thanks all,

Not going to give Angra a run... he's the top folder on this team. :toast:

I recently passed the 25 million mark also... but that's under another ID for my home team ;)

Time to put my machines with their normal home... till next time John needs to be passed... :)

[twiggy]

Thanks for the help Xciter!:toast:


See you next time.:D

[Ackbar]

Thanks all,

Not going to give Angra a run... he's the top folder on this team. :toast:

I recently passed the 25 million mark also... but that's under another ID for my home team ;)

Time to put my machines with their normal home... till next time John needs to be passed... :)

Maybe we should all put some nodes under John's name so we can get Xciter to stay. ;)

[Ugly n Grey]

LOL, there a man who is thinking :cheer: Ackbar

[mtb856]

good grief haha, our sub-team definitely didn't do it that fast. Big congrats are in order then...... feel like transferring to the OCforums? :D

[Ugly n Grey]

no poaching even in jest :) :toast:

[twiggy]

Maybe we should all put some nodes under John's name so we can get Xciter to stay. ;)

:clap:

:rofl: It'd take the whole team! We have 544 active CPU's(within 50 days), 390 of those are X's! :fact:

[angra]

Thanks all,

Not going to give Angra a run... he's the top folder on this team. :toast:



aww, I'll take a shot for the team :)

you can even make a hypnotoad target avatar if you want, I won't be offended :).

on a different topic, I wish I had a little more funding to do GPU work - I would like to explore converting the FAH cores to make use of GPUs. I have a decent idea of how to do it, but I bet there are a _lot_ of cores that would need to be converted, and, of course, every GPU would require its own program to do well (probably).

[Ugly n Grey]

That is a problem with GPU's, where x86 procs follow an instruction set laid out in days gone by ensuring reasonable code compatibility, GPU's have no such standards. Having said that, there are ways to generalize the spectrum and compile reasonably well for the top four GPU's in a market segment. The application could even be made so the "engine" plugs in. After all, the code still has to launch and bootstrap from the CPU. The same holds true with the clearspeed math coprocessor.

[angra]

That is a problem with GPU's, where x86 procs follow an instruction set laid out in days gone by ensuring reasonable code compatibility, GPU's have no such standards. Having said that, there are ways to generalize the spectrum and compile reasonably well for the top four GPU's in a market segment. The application could even be made so the "engine" plugs in. After all, the code still has to launch and bootstrap from the CPU. The same holds true with the clearspeed math coprocessor.

yes, I am actually working directly on such things in my research. the problem with GPUs is that the capabilities differe so much from chip to chip that the optimizations required are significantly different. One thing that can be done is to write the applications using a portable/standard domain-specific API, such as <a href="http://www.vsipl.org">VSIPL</a> or BLAS or somesuch, or to determine a physics API or something like that, and have optimized implementations for each new platform.

There are approaches that could work, but the field is very new. research is mostly going on in the "specific-optimized-functional-kernels" stage, rather than the productivity phase. There is Brook for GPU, which is a somewhat general programming language for GPGPU.

There is work also in the Polymorphous Computing Architectures program that relates to programming model for highly parallel and reconfigurable architectures (of which clearpseed, GPU, cell, etc, are (degenerate in some cases) proper subsets). It will be very interesting to see where this stuff goes.

Unfortunately, the folding kernels are not open source, so we can't experiment directly in the context of the FAH project. We can construct similar kernels for testing, but who has that kind of spare time? :)