NCCS GPU
A small set of nodes configured with Graphical Processing Units (GPUs) has been added to the Discover system specifically designed for application porting, testing, and some small amount of scaling. Users who wish to access this system for these purposes may contact the User Services Group. In the request for access, please provide a description of your application, potential uses for the GPUs, and any specific tools that may be required for support.
NCCS GPU System Details
System Architecture
This system was designed to be quickly used by Discover users and is coupled to the Discover IBM GPFS file system. Therefore, the operating environment is consistent with the Discover operating system, software stack, tools, and file system.
The Portland Group Inc. (PGI) CUDA compiler with FORTRAN support is available for use on these nodes.
Detail on how to run CUDA compiler, Click here.
Overall Specifications
- Manufacturer: IBM
- 36.4 Tflop/s
- 384 Total Cores (General Purpose Intel Cores)
- 28,672 Total Cores (GPU Streaming Cores)
- Production: 2Q 2011
32 Compute Nodes each configured with the following:
- 2 Hex-core 2.8 GHz Intel Xeon Westemere Processors (4 flop/s per clock)
- 48 GB of memory per node
- 2 NVidia M2070 GPUs each connected through a dedicated x16 PCIe Gen2 connection
- Interconnect: Infiniband QDR
64 Graphical Processing Units each configured with the following:
- 1 Tesla GPU (M2070)
- 448 CUDA cores
- ECC Memory
- 6 GB of GDDR5 memory
- 515 Gflop/s of double precision floating point performance (peak)
- 1.03 Tflop/s of single precision floating point performance (peak)
- 148 GB/sec memory bandwidth
- 1 PCIe x16 Gen2 system interface