Slashdot Mirror
Full Details Uncovered on Chinese Tianhe-2 Supercomputer
An anonymous reader writes "With help from a draft report (PDF) from Oak Ridge National Laboratory's Jack Dongarra, who also spearheads the process of verifying the top of the pack supercomputer, we get a detailed look at China's Tianhe-2 system. As noted previously, the system will be housed at the National Supercomputer Center in Guangzhou and has been aimed at providing an open platform for research and education and to provide a high performance computing service for southern China. From Jack's details: '... was sent results showing a run of HPL benchmark using 14,336 nodes, that run was made using 50 GB of the memory of each node and achieved 30.65 petaflops out of a theoretical peak of 49.19 petaflops, or an efficiency of 62.3% of theoretical peak performance taking a little over 5 hours to complete.The fastest result shown was using 90% of the machine. They are expecting to make improvements and increase the number of nodes used in the test.'"
Half an hour after you get the results, you have to run the simulation all over again.
I want to delete my account but Slashdot doesn't allow it.
The machine uses a mixture of Xeons and Xeon Phi's communicating over PCIe, and the nodes seem to communicate over a home-grown chipset hanging off the PCIe bus. The efficiency number (65%) while not stellar is typical of such machines (i.e. about the same as Titan.
It would seem therefore that the interconnect is competitive, or at least not restrictive. Given how close it is to Titan, I would not be surprised if the lack of efficiency comes from the relative pain of using fast co-processors which lack decent single threaded speed or memory.
Anyway, it sll seems respectable enough, though I was hoping it would be a bit more interesting and use the Longsoon variant they've been trying to build a supercomputer out of.
SJW n. One who posts facts.
...can it run Crysis 3?