ShARC specifications¶

Total capacity¶

Worker nodes: 121
CPU cores: 2024
Total memory: 12160 GiB
GPUs: 40
Fast network filesystem (Lustre): 669 TiB (/home and /data are shared with Iceberg)

Note that some of these resources have been purchased by research groups who have exclusive access to them.

98 nodes are publicly available (not exclusive to research groups).

Machine: Dell PowerEdge C6320
CPUs: 2 x Intel Xeon E5-2630 v3
- Haswell processor microarchitecture;
- 2.40 GHz;
- Support for AVX2 vectorisation instructions (simultaneously apply the same operation to multiple values in hardware);
- Support for Fused Multiply-Add instructions (expedites operations involving the accummulation of products e.g. matrix multiplication).
- Hyperthreading is disabled on all nodes bar four that are reserved for interactive jobs.
RAM: 64 GB (i.e. 4 GiB / core)
- 1866 MHz;
- DDR4.
Local storage: 1 TiB SATA III HDD
- /scratch: 836 GiB of temporary storage;
- /tmp: 16 GiB of temporary storage.

Four nodes are publicly available (not exclusive to research groups).

These are identical to the general CPU nodes but with 256 GiB RAM (16 GiB per core).

Two nodes are publicly available (not exclusive to research groups):

Machine: Dell PowerEdge C4130
CPUs: 2 x Intel Xeon E5-2630 v3 (2.40GHz)
RAM: 64 GB (i.e. 4 GiB / core); 1866 MHz; DDR4
Local storage: 800 GiB SATA SSD
GPUs: 8 x NVIDIA Tesla K80
- 24 GiB of GDDR5 memory (12 GiB per GPU; 192 GiB per node)
- Up to 2.91 Teraflops of double precision performance with NVIDIA GPU Boost
- Up to 8.74 Teraflops of single precision performance with NVIDIA GPU Boost

One node is publicly available:

OS: Centos 7.x (binary compatible with RedHat Enterprise Linux 7.x) on all nodes
Interactive and batch job scheduling software: Son of Grid Engine
Many applications, compilers, libraries and parallel processing tools. See Software on ShARC

Two login nodes (for resilience)
Other nodes to provide:
- Lustre parallel filesystem
- Son of Grid Engine scheduler ‘head’ nodes
- Directory services