Thanks to all of our IMSS coworkers who volunteered much of their time and energy. Without you all it would of taken us forever to finish. The 5am pick up of the old Pangu cluster went smoothly except for one of the trucks being to tall to fit under the driveway overhang (oops).
The floor has been cleaned up and the electrical crew is now in there making headway with the new 30amp receptacles and whips.
Some pictures
http://imss-hpc-pics.s3-website-us-west-1.amazonaws.com
The CITerra cluster will be composed of 254 traditional computes nodes. Each
has 12 cores and 4 GB of RAM per core. The whole cluster will be connected with
a low latency, high bandwidth network called infinibad. In addition to
the traditional compute nodes, it also will have 60 GPU based nodes with a total
of 180 Nvidia M2090 GPUs. The system is backed with a 720 TB raw Lustre
high performance parallel filesystem. This filesystem can perform at
around 9 GB/s. The traditional processors have a peak performace of 40
Teraflops and the GPU processors have a peak performance of 110 Teraflops.
IMSS now offers a High Performance Computing (HPC) pay-as-you-go service. This service is ideal for faculty and research groups that do not have the resources to run their own HPC systems, which usually require large financial resources to maintain the required hardware and infrastructure. With IMSS’ HPC service, you simply pay based upon usage and storage. For additional details and pricing, please visit http://imss.caltech.edu/help/hpc or email help-hpc@caltech.edu.
A new GPU cluster has been installed into South Mudd.
This cluster has 72 Nvidia M2070 GPUs. It was built for running MD code.
The NeuroEcon cluster has been installed and being tested. This cluster has 12 compute nodes, and a 70 TB Lustre filesystem. The cluster also is running the Matlab distributed comutation server.
over the last 2 weeks we have been upgraded the filesystem on the CITerra cluster. This has been completed and the cluster is back online. This will hopefully solve the quota and dangling entry problems we have seen in the past. All data was copied off of the array onto loaner equipment before proceeding. The DDN storage array in use on the cluster was also reconfigured during the downtime and 3 new login nodes were added.
We have been working on the design for the HSS cluster. This will haveabout 120TB of storage and 12-16 compute nodes. It will likely run the Lustre file system.
On Monday June 16th, Dell arrived to start the “racking and stacking” phase of the new miller cluster. This cluster will be housed in the Powell-Booth Machine room thanks to CACR. Next week we start on the software installation and configuration of the cluster.
Date: June 25, 2008
Time: 9:00am- 2:00p
Dell and our partners invite you to come learn about new
products, recent technology advances in HPC clusters and
solutions to enable the next generation of simulation. Come
participate and learn about such topics as:
Processor and GPU technology
High performance storage solutions
Cluster Interconnects
Compliers, software and more
AGENDA
9:00am HPC: Dell, NVIDEA, Data Direct Networks,
Intel & Myricom
9:00am Dell- New Products roadmap
12:00pm Lunch
12:00pm Demonstrations, Open Discussions & Information
LOCATION:
Winnett Center (Bookstore)
Lunch will be served in the courtyard…
RSVP to:
Genelle Meyer x6276
-or-
Amy Knowles:
Amy_Knowles@dell.com
(818) 422-8440
