80 likes | 89 Views
Fall 2010 Jih-Kwon Peir Computer Information Science Engineering University of Florida. CIS 6930: Chip Multiprocessor: GPU Architecture and Programming. Tutorial: How to Use the HPC environment for this course. How to Apply the HPC account Login to HPC account Setup CUDA environment
E N D
Fall 2010 • Jih-Kwon Peir • Computer Information Science Engineering • University of Florida CIS 6930: Chip Multiprocessor: GPU Architecture and Programming
Tutorial: How to Use the HPC environment for this course • How to Apply the HPC account • Login to HPC account • Setup CUDA environment • devicequery: the first CUDA application • MultrixAdd: the sample code
Apply HPC account • Go to http://hpc.ufl.edu/ • Click “Support” • Click “Account Requests” • Login with your gatorlink • When Applying • Input Dr Peir as your Sponsor • Input GPU Course as Type of research • Input CIS6930 in Comments • Remember to active your Bugzilla account when you receive the email.
Access the HPC environment • Use SSH tool as putty etc; • Host name: • submit.hpc.ufl.edu • You may be on either submit1.hpc.ufl.edu or submit2.hpc.ufl.edu • Submit is only to submit jobs or do simple tasks; • You need ssh to test01/test02/test03/test04/test05 to do time-consuming jobs;
Build the CUDA • The CUDA-SDK directory on HPC is • /opt/cuda-sdk • You may download the cuda_test.tar.gz from the course website. • The source directory is …/mycuda/myproject/ in the cuda_test.tar.gz; • The binary files are in …/bin/ or …/mycuda/bin/linux/release/ • You can also build your own workspace.
Qsub • You need first use the Qsub to get the tesla machine. The command is as follows: • qsub -I -l nodes=1,walltime=01:00:00,gres=gpu -q tesla • You will be login into one of the tesla machines.
Compile and Run • Run the deviceQuery • >cd …/cuda_test/mycude/myproject/deviceQuery • >make • > ../../bin/linux/release/deviceQuery
Another Example • Matrix Add: • See the matrixAdd.2 and matrixAdd.3 code in the cuda_test