Mini-itx motherboard with 8200

This might be interesting to the people that are looking for an embedded, low-power CUDA solution:

[url=“Jetway JNC62K (GeForce 8200 on mini-ITX) | bit-tech.net”]http://www.bit-tech.net/hardware/2008/05/2...8200-mini-itx/1[/url]

I’m not sure whether the speed is reasonable enough to do any real number crunching, but who knows.

That’s a neat looking little motherboard. AM2+ socket and two gigabit ethernet ports?! But, yeah, with only 2 multiprocessors and a 64 bit bus to system memory, it’s hard to imagine the 8200 doing better than the Phenom in the socket next to it.

This is where it would be nice if there were a standard CUDAMark application that we could get reviewers to start running on new hardware. bandwidthTest + a mix of 4 compute bound and memory bound kernels from the SDK with a Windows GUI to summarize the scores should be enough. (Any of you Windows coders out there looking for a project? :) )

But on the other hand, it DOES run CUDA, whereas CUDA performance on a CPU in device_emulation mode are just catastrophic (as repeated everywhere, these are only for debugging and/or running memory checkers - not to do memory calculation).

So, although you can’t efficiently use it as a node in some super-computer, you can use it as a small cheap platform to test CUDA code on a real GPU before uploading and recompiling it on the multi-GPU/multi-CPU monster in the lab.

And I remember having read elsewhere that the newer chipset (both from nVidia and ATI) can collaborate with discrete GPU. Thus plugin in a GPU doesn’t disable the chipset’s graphic output, but doubles the graphic cards (although producing a rather assymetric SLI/CrossFire with the discrete card much more power than the chipset).

Thus you can :

  • Start experimenting CUDA on the ITX board as-is

  • Add later a slighlty more powerfull card (something with 4 multi processors and dedicated GDDR)

  • Experiment with the 8200 running the desktop (even with Compiz/Beryl enabled, yay !) and the bigger GPU running CUDA.

  • As a nice bonus you can also test your code on multiple GPUs, without have buy something expensive as a 9800GX2 card.

  • And once you’re tired and need to relax : reboot the machine under windows, and enjoy both card in SLI which should give just the necessary umph to run decent games.

…or just enough performance run Vista :D

(But maybe not Crysis, at least not at the maximal settings :P ).

Good points, although this particular motherboard lacks PCI-Express slots. (Just 1 PCI slot.)

And actually, I take back my previous slight to the 8200’s speed. This motherboard can’t actually accept any current Phenom chips, since they all have TDP > 65W. If you are putting a low power Sempron into this board, suddenly CUDA is quite a force-multiplier.