[Neurodebian-users] fsl-5.0-gpu

Jonathan Berrebi Jonathan.Berrebi at ki.se
Mon Sep 29 13:30:44 UTC 2014


Thank you for your answers!

Well as I said before, it was working for me before the summer on an nvidia Tesla K20. Then, as I am bad in taking notes, there is uncertainty on which cuda I installed then, i.e. if it was 5.0 or 5.5. It seems not possible to have a K20 with cuda 5.0 so I probably installed the 5.5.

Anyway, it worked for me and I got bedpostx to work in 30 minutes compared to 24 hours on a xeon E5645. So the interest was great for me, or rather for the users. Previously I did the same install on debian wheezy on my laptop that has a quadro 1000M and it took 6 hours instead of 24 hours on the E5645.

On the GPU K20 server I instaled a dual boot with debian wheezy and Centos 6.5. Curiously it worked fine on debian (with a few modifications that I sent to neurodebian about 6 months ago) but not on CentOs which is the original OS for FSL.

I would hate to use CentOS on that server as debian gives so many conveniences with neuro.debian. 

Is there a possibility to downgrade the whole FSL to version 5.0.6 instead of 5.0.7? Then I would probably get FSL gpu to work back again.

If you want me to help I would be happy to. I would need some guidance though as I guess more than I know and the server is going to be public to the users soon so I will have to avoid restarting it.

Thank you,

Best,

Jonathan




________________________________________
De : Yury V. Zaytsev [yury at shurup.com]
Envoyé : lundi 29 septembre 2014 11:05
À : Michael Hanke
Cc : Jonathan Berrebi; neurodebian-users at lists.alioth.debian.org
Objet : Re: [Neurodebian-users] fsl-5.0-gpu

On Mon, 2014-09-29 at 10:27 +0200, Michael Hanke wrote:
>
> I am afraid that the situation will remain at the status quo until
> someone can help, as I am not planning on buying the necessary
> hardware.

Hi Michael,

Just for your information, you can test with

    http://code.google.com/p/gpuocelot/

even if you don't have the hardware, even though, of course, it's very
limited and difficult to use.

Also, we might be able to help if hardware is the only concern, but if
it's more about time and interest, then it's a different problem :-)

--
Sincerely yours,
Yury V. Zaytsev





More information about the Neurodebian-users mailing list