SIGN-IN

Cluster graham.sharcnet.ca

Links System documentation in the SHARCNET Help Wiki

Manufacturer Huawei
Operating System CentOS 7
Interconnect EDR + FDR Infiniband
Total processors/cores 33448
Nodes
1‑800
32 cores
2 sockets x 16 cores per socket
Intel E5-2683 v4 (Broadwell) @ 2.1 GHz
Type: Compute
Notes: Base profile compute nodes.
Memory: 128.0 GB
Local storage: 1.2 TB
801‑803
56 cores
4 sockets x 14 cores per socket
Intel E7-4850 v3 (Haswell) @ 2.2 GHz
Type: Compute
Memory: 3072.0 GB
Local storage: 1.2 TB
804‑827
32 cores
2 sockets x 16 cores per socket
Intel E5-2683 v4 (Broadwell) @ 2.1 GHz
Type: Compute
Memory: 512.0 GB
Local storage: 1.2 TB
828‑987
32 cores
2 sockets x 16 cores per socket
Intel E5-2683 v4 (Broadwell) @ 2.1 GHz
Type: Compute
Notes: Accelerated compute nodes with 2 × NVIDIA Pascal P100 GPUs (12GB HBM2)
Memory: 128.0 GB
Local storage: 800 TB
988‑1043
32 cores
2 sockets x 16 cores per socket
Intel E5-2683 v4 (Broadwell) @ 2.1 GHz
Type: Compute
Notes: Cloud configuration
Memory: 256.0 GB
Local storage: 1.2 TB
Total attached storage 14500 TB
Suitable use

Heterogeneous cluster, suitable for a variety of workloads.

Software available

MAP, ADF/BAND, GAUSSIAN, MATLAB, STAR-CCM+, ANSYS, DDT, AMBER, LSDYNA

Current system state details Graphs

Recent System Notices

Status Status Notes
Dec 14 2017, 04:35PM
(2 days ago)

Graham is performing normally now.

Dec 14 2017, 10:32AM
(2 days ago)

Non-default accounts/projects are not available. Working on a fix.

Dec 14 2017, 09:56AM
(2 days ago)

Graham is available for normal use. Please report any oddities you notice.

Dec 12 2017, 01:53PM
(4 days ago)

Graham is currently unavailable for software upgrades. It is expected to return to service before 9am Thursday, December 14.

Sep 28 2017, 05:04PM
(3 months ago)

Graham is fully operational. The CVMFS restricted repository was fixed on Sept 8.

To avoid overloading the login nodes, some resource limits now apply: - no process can exceed 1 cpu-hour. - no process can exceed 4GB in virtual size (VSZ in top/ps). - no user can have more than 20 login sessions at once. - no user can have more than 100 threads at once.

These limits apply only to the login nodes, and do not affect jobs (sbatch/srun) or interactive allocations (salloc). These limits are very much like limits on other Sharcnet clusters such as Orca, and are similar to those on Cedar.

Sign-in to get full status history