NCSA has now powered up Forge, a 153 teraflop heterogeneous supercomputer for use by scientists and engineers across the country. The system features Dell PowerEdge C6145 servers (AMD Opteron Magny-Cours 6136 2.4 GHz dual-socket eight-core) accelerated by NVIDIA Tesla M2070 GPUs.
Seventy percent of the compute time Forge offers will be allocated through the National Science Foundation’s Extreme Science and Engineering Discovery Environment (XSEDE) program. XSEDE, a cross-country partnership of nearly 20 institutions, is led by NCSA and provides digital resources, services, tools, and support to the nation’s science and engineering research community. The remaining 30 percent of Forge’s cycles will be allocated to NCSA’s Private Sector Program and to faculty, staff, and students at the University of Illinois at Urbana-Champaign.
Read the Full Story.
Doesn’t the Poweredge C6145 have the opteron 6100’s instead of xeons? I think this is a typo in the story, it could mean Xeon 5600’s on Poweredge C6100’s instead. I just got a bunch of these 6100’s. That’s why i noticed this. Thanks
According to their detailed specs, Forge is based on AMD Opteron Magny-Cours 6136 2.4 GHz dual-socket eight-core. We regret the error.
Interesting! The C6145s have dual IO hubs per node which should give better HBA->CPU performance than the 6100 with so many GPUs attached per node. That’s only true though if the NUMA related issues are sorted out (ie making sure that threads are properly matched to cores with the right IOH/GPU attached). I hope they publish some benchmarks on this setup with things like NAMD STMV with differing numbers of GPUs.