[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [HTCondor-users] matching gpu devicename



You should be getting back a CUDACapability attribute, something like this:

CUDACapability=2.0
CUDADeviceName="GeForce GTX 480"
CUDADriverVersion=4.20
CUDAECCEnabled=false
CUDAGlobalMemoryMb=1536
CUDARuntimeVersion=4.10

if you pass the -extra flag, then you should also be getting attributes CUDAClockMhz, CUDAComputeUnits and CUDACoresPerCU
like this

CUDACapability=2.0
CUDAClockMhz=1400.00
CUDAComputeUnits=15
CUDACoresPerCU=32
CUDADeviceName="GeForce GTX 480"
CUDADriverVersion=4.20
CUDAECCEnabled=false
CUDAGlobalMemoryMb=1536
CUDARuntimeVersion=4.10

If you are using the GPUs metaknob, you can set the -extra flag by setting GPU_DISCOVERY_EXTRA like this:

GPU_DISCOVERY_EXTRA = -extra
use FEATURE : GPUs

-tj

On 6/2/2015 6:39 AM, Michael Di Domenico wrote:
On Mon, Jun 1, 2015 at 3:17 PM, John (TJ) Knoeller <johnkn@xxxxxxxxxxx> wrote:
These are effectively the same device?  It's annoying that they have
separate names then. You should complain to NVidia... ;)
effectively they are the same card, but there are some design changes
concerning thermals and airflow between the two.  it's annoying, but
unfortunately there's a valid reason.

But seriously, your best course for now would be to condor_gpu_discovery
program in a script that re-writes it's output
to change the various CUDA[n]DeviceName attributes to a single
CUDADeviceName = "Tesla K10" attribute.
ugh, i was hoping to avoid this, but i guess it's the only option

Or tell your users to target the GPUs based on capability rather than by
name.
i'm unclear how this matching takes place.  i don't see the
CUDA_CAPABILITY(?) classads bound to my slots when i run condor_status
-l <slot> are they hidden somewhere or is condor_discovery_gpu not
spitting out everything it should
_______________________________________________
HTCondor-users mailing list
To unsubscribe, send a message to htcondor-users-request@xxxxxxxxxxx with a
subject: Unsubscribe
You can also unsubscribe by visiting
https://lists.cs.wisc.edu/mailman/listinfo/htcondor-users

The archives can be found at:
https://lists.cs.wisc.edu/archive/htcondor-users/