Short answer: NO. - minimum will be CUDA 5.0.
CUDA 5.0 will be considered optimum for AMBER 14 (as it is for AMBER 12) -
we may move to CUDA 6.5 [skipping 5.5 and 6.0 since, although supported,
they have performance regression] but that will be a while.
I'd RMA the machine as defective (it clearly is unless they can offer you
a firmware upgrade to fix it), get a refund and buy something that is
supported. This is why I have spent the last couple of years working
closely with Exxact to develop and AMBER Certified Hardware Platform:
http://ambermd.org/gpus/recommended_hardware.htm#hardware
This is guaranteed to work with AMBER as a turn key solution (and is what
we develop and test on) and thus avoids weird incompatibilities such as
what you are seeing.
All the best
Ross
On 2/17/14, 10:44 PM, "Eugene Yedvabny" <eyedvabny.berkeley.edu> wrote:
>Hello Amber community,
>
>My group is finally getting into GPU computing with a recent purchase of a
>GPU enclosure and M2090 cards. It seems, however, that the enclose (Dell
>410x if anyone has experience with these) is limiting us to CUDA 4.2, as
>CUDA 5 just results in constant hardware lockups. As such, I was wondering
>if Amber is planning on continuing to support CUDA 4.2 for the foreseeable
>future, or if that support is going away with Amber 14?
>
>Thank you,
>Eugene Yedvabny
>_______________________________________________
>AMBER mailing list
>AMBER.ambermd.org
>http://lists.ambermd.org/mailman/listinfo/amber
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Mon Feb 17 2014 - 23:30:02 PST
