NVIDIA Tesla C2050 Demonstration Unit Call for Participation
UPDATE: Information on how external and commercial researchers can trial the VPAC demonstration unit 'Enrico' can now be found at the Enrico information page.
24th June 2010
VPAC previously announced the availability of our Enrico node for users to evaluate the latest NVIDIA Fermi architecture GPGPU.
Globally, an increasing number of applications have been ported to GPGPU technologies over the previous year. Much of that work can be found at the nVidia CUDA Zone . If any VPAC users desire to compare an applications performance on a traditional cluster with Enrico, VPAC encourages you to acquire and load the CUDA-ised application, and run comparatively.
The VPAC Helpdesk can provide assistance regarding applications installation and use.
Those users taking advantage of Enrico are asked to provide their performance results, user experiences, and suggestions regarding the GPGPU part and related software to the Helpdesk.
Papers regarding your work are also solicited.
VPAC trust you will have good experiences that are expanding your horizons of what you can accomplish with high performance computing.
1st June 2010
VPAC, with sponsorship of Xenon Systems, has brought the node Enrico on-line for member users to evaluate. Enrico is a 2.27 GHz quad core Nehalem server equipped with 12GB RAM and two of the latest NVIDIA Fermi architecture Tesla C2050 general purpose graphics processing units (GPGPU).
The C2050 extends GPGPU technology through improving double precision floating point performance and adding error correcting memory, making it suitable for 64 bit computational applications, not just graphics or video applications.
GPGPUs are becoming the successors to vector processors for their ability to perform scientific computations at very high speeds. Global experiences with GPGPUs produce performance gains from 20~100 times that of the fastest computer cores available.
Applications written using the NVIDIA Compute Unified Device Architecture (CUDA) or the OpenCL framework can be executed on Enrico to access the Tesla C2050. Only one job will be allocated to Enrico at a time, with each job being able to use all 4 cores and both the C2050 units under the application program’s control with a 24 hour limitation to wall time.
Those without VPAC accounts can apply for an account. Prospective users without affiliations with our member universities can apply for an Enrico evaluation account by contacting Phil Tannenbaum on the contact details below.
Jobs will only be executed on Enrico when the user has been approved and validated for project Enrico, and also includes the fermi queue parameter on their job script as follows:
#PBS -q fermi
Users will be sent an acceptance email within 1 business day of making their request to join the project. Once an acceptance email has been received, users must logout and login back into their session for the project to accept jobs.
CUDA or OpenCL jobs not having the proper queue specified will be launched on VPAC traditional clusters, and will fail because the GPGPU architecture is not supported. The 4 node Tesla C1060 based GPGPU cluster is not affected by the Enrico demonstration system.
The availability of Enrico will be reviewed periodically and assessed depending on demand. Users having special requirements are encouraged to contact Phil Tannenbaum, HPC Centre Manager at 03 9925 4733 or firstname.lastname@example.org for information.