Dragon1

Hosted at UMons, the Dragon1 cluster consists of 28 compute nodes:

• 26 nodes with two Intel Sandy Bridge E5-2670 (2×8 cores @ 2.6 GHz)
• 2 nodes with two Intel Sandy Bridge E5-2650 (2×8 cores @ 2.0 GHz)
  All nodes feature 128 GB RAM and 1.1 TB local scratch.

The cluster is interconnected with Gigabit Ethernet (10 Gigabit for the 36 TB NFS file server).
GPU resources include:

• 2 nodes with 2× NVIDIA Tesla M2075 (512 GFLOPS float64 each)
• 2 nodes with 2× NVIDIA Tesla K20m (1.1 TFLOPS float64 each)

Suitable for:

Long shared-memory parallel jobs (OpenMP, Pthreads) or resource-intensive sequential workloads.

Max wall times:

• Long queue: 41 days
• Batch queue: 5 days
• GPU queues: 15–21 days depending on GRES

Resources

• Home directory (20 GB per user)
• Local working directory /scratch ($LOCALSCRATCH)
• No internet access on compute nodes
• Queues*:
  • Long queue: Max 41 days, 40 CPUs/user, 500 jobs/user
  • Default queue (batch): Max 5 days, 40 CPUs/user, 500 jobs/user
• GPU GRES*:
  • gpu: Max 15 days, gres=gpu:kepler:1 or gres=gpu:tesla:1
  • lgpu: Max 21 days, gres=gpu:1

Access / Support

This cluster has been decommissioned