[prev in list] [next in list] [prev in thread] [next in thread] 

List:       npaci-rocks-discussion
Subject:    [Rocks-Discuss]  We need help with GPU cluster
From:       marlis fulgueira <marlis.fulgueira () gmail ! com>
Date:       2015-06-19 12:12:07
Message-ID: CAP43N1FbnHJqM=j-6s9W6fLKaiNG1FK8PA3VSV2G-8FvJghLhA () mail ! gmail ! com
[Download RAW message or body]

We have a cluster with rock cluster 6.1 installed. It is built with
components COTS, each computer have an Intel core I 7 and 2 GPU, NVIDIA GTX
260. We need use it, for run job writing in OpenCL for that reason we
install manually, in all the nodes, the 295.53 driver for NVIDIA GPU and
openCL runtime for the CPU. When we run a simple job in 1 NODE ( that
means, query the devices and show the name of the plataform and the name of
the device) it works fine, show us the name of the 2 GPU but when we want
other job in 1 NODE too (the same), like a matrix multiplication, show us
the name of device, but is like the GPU do nothing because the result
matrix is empty, only with 0. We did then a few test, we ran the matrix
multiplication job in the server witch have the same configuration, but it
have the driver 337.25 because the 295.53 does not work in the server (when
we installed the server don't wake up) the GPUs are capable to execute  the
multiplication. As I say that the driver in the server is not the same that
the nodes have, we changed, and the result was that the simple job, no
recognized the OpenCL plataform, of course, the name of the device either.

After that we reinstall the node and put driver 295. 53, with toolkit 3.2
and gpu computing sdk 3.2 for cuda, and still happend the same, this time
we ran a job with cuda, and did not recognized cuda device. The same job we
ran in the server and execute well.

We think that the problem is the kernel installed in the nodes or something
relating with the operating system, or something that we are doing bad,
please, help us.

Best Regards,
    Ing. Marlis Fulgueira Camilo
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.sdsc.edu/pipermail/npaci-rocks-discussion/attachments/20150619/34cbe665/attachment.html \



[prev in list] [next in list] [prev in thread] [next in thread]