LABORATORY RESEARCH COMPUTING
Berkeley Lab provides Lawrencium, a 358-node (3448 computational cores) Linux cluster, equipped with a high performance, low latency Infiniband interconnect, to Berkeley Lab researchers needing access to scientific computational resources. The system, which consists of shared core nodes and PI-contributed Condo nodes, has a theoretical peak performance rating of 37 teraflops and delivers over 21M processor hours to researchers every year.

SCIENTIFIC CLUSTER SUPPORT
With over 35 clusters in production, HPC Services also offers comprehensive Linux cluster support, including data center resource planning, pre-purchase consulting, procurement assistance, installation, and ongoing support, for PI-owned clusters. Our HPC User Services consultants can help you to get your application running well to make best use of your new cluster. UC Berkeley PIs can also make use of our services through the Cal HPC program available through IST.

NEWS

Jan 24, 2012 - Bootstrapping Institutional Capability
HPC Services Manager Gary Jung talks about the issues institutions may encounter when developing new or enhancing existing infrastructure to support data intensive science at the Winter 2012 ESCC/Internet2 Joint Techs Conference in Baton Rouge, LA this week.

Nov 16, 2011 - Warewulf3 at SC11
Greg Kurtzer will be hosting a round table discussion to talk about the new Warewulf3 cluster toolkit release at SC11 in Seattle. Drop by the Berkeley Lab Booth #512 at 3pm on Weds Nov 16 to find out why you should be using it.

Nov 3, 2011 - Supercomputers Accelerate Development of Advanced Materials
Researchers from Berkeley Lab and MIT have teamed up to develop a new tool,as part of the Materials Project, to speed up the development for new materials. The project incorporates the use of supercomputing resources including Lawrencium to characterize the properties of inorganic compounds. More

Oct 25, 2011 - Supercomputing As A Service
LBL CIO Rosio Alvarez and HPC Services Manager Gary Jung present their experiences using cloud services for HPC at InformationWeek's GovCloud 2011 conference in Washington DC. More

Sep 19, 2011 - Lawrencium LR1 relocated to SDSC
Did you know that the Lawrencium LR1 compute nodes have been relocated to the San Diego Supercomputing Center? This was done as part of the IT Division's quest to best optimize data center space. The LR1 nodes are connected via a dedicated 10 gigabit ESNET virtual circuit into the Lab's HPC infrastructure here in Berkeley. The impact of having LR1 remote should be minimal for most users.

Apr 19, 2011 - UC Cloud Summit
HPC Services staffers Greg Kurtzer talks about the new Warewulf Version 3 Cluster Toolkit release and Krishna Muriki and Kai Song present their work using Amazon CCI and Amazon GPU at the first annual UC Cloud Summit hosted by UCLA.

Mar 28, 2011 - Lattice Optimization Using Cloud and GPU. ALS physicists Changchun Sun and Hiroshi Nishimura along with HPCS staff Kai Song, Susan James, Krishna Muriki, and Yong Qin recently explored the use of Amazon's Cluster Compute Cloud and GPU computing to perform Lattice Optimization for particle tracking at the ALS. Their work will be presented during the poster session at the Particle Accelerator Conference (PAC11) in New York later this week.

ANNOUNCEMENTS

GPU Compute Nodes Now Available. We have a total of 4 second generation (lr2) nodes each equipped with two Nvidia C2050 GPU Fermi cards. Each lr2 node has a total 12 cores and 24GB of memory. Use these PBS job submission arguments to get one more of these nodes:

" -l nodes=X:ppn=Y:gpu -q lr_batch "

where X - num of nodes from 1 to 3 & Y - num of cores/processors per node from 1 to 12. One of the four GPU nodes is available for quick 30 mins debug jobs and users can access this node by using:

"-l nodes=1:ppn=12:gpu -q lr_debug"

Lawrencium Large Memory Nodes. Three large memory nodes are now available as part of the Lawrence segment of the Lawrencium cluster. Each node has 48 AMD processor cores and 256GB(2) and 512GB(1) of memory. Use these PBS job submission arguments to get one of the nodes:

" -l nodes=1:ppn=Y:bigmem_256 -q lr_batch "

" -l nodes=1:ppn=Y:bigmem_512 -q lr_batch "

" -l nodes=1:ppn=Y:bigmem -q lr_batch " # Use any bigmem node

HPC Services has formed a new general interest HPC user mailing list for users of Lawrencium and of clusters maintained by HPC Services. Members of the Berkeley Lab community working with high-performance computing technologies are also invited to participate in this user mailing list to discuss topis like HPC applications, best practices and future technologies. If you like to participate in the discussion we encourage you to subscribe to the mailing list here: https://lists.lbl.gov/sympa/info/hpc-users