System is an older Dell 4-socket server using AMD Opteron 6220 processors.
Each processor has 8 cores (4 pairs), running at a nominal 3.0 GHz (3.3 GHz max all-core Turbo).
Each processor has 4 DDR3 DRAM channels.
The system is fairly old -- my first results on this machine are from December 2011 (the Opteron 6220 processor was released in November 2011).
Those original results are broken in a number of ways, so I re-ran the tests with proper array sizes and proper process binding.
If I am interpreting the PCI configuration space data correctly, the system is running the DRAMs at 1333 Mbs, giving a peak memory bandwidth of
4 sockets * 4 channels/socket * 8 Bytes/channel * 1.333 Gtransfers/s = 170GB/s
This is divided up into 8 NUMA nodes, with 2 channels each (21.3 GB/s).
These results were compiled with gcc, so they do not have streaming stores.
--
John D. McCalpin, Ph.D.
Texas Advanced Computing Center
University of Texas at Austin
https://www.tacc.utexas.edu/about/directory/john-mccalpin
Received on Wed May 27 2015 - 17:35:39 CDT