OnePlus 5T review: Hands-on
Sun, NEC, and AMD Partner for 50 Teraflops Opteron Cluster
Considering that Itanium-based server supplier and vector supercomputer
maker NEC and RISC/Unix and Itanium server maker Fujitsu pretty much rule
the indigenous supercomputer market in Japan, it was somewhat surprising to
see that Sun Microsystems partnered this week with NEC and AMD to deliver a
50 teraflops supercomputer based on future eight-socket "Galaxy"
Opteron-based servers. You would have thought that Sun would have partnered
with Fujitsu to beat out NEC for the deal at the Tokyo Institute of
Under the deal, Sun and AMD will deliver a cluster running Linux (with
Solaris 10 as an option on the nodes) with 5,240 dual-core processors. The
machine, comprised of 655 servers, will also have 360 Advance co-processors
from ClearSpeed for boosting the speed on HPC algorithms. By next summer,
the Tokyo Institute of Technology plans to add another 240 Advance
co-processors and boost the number of Opteron servers such that it can reach
100 teraflops of performance. This is by far the largest HPC deal--in terms
of processor count and raw aggregate performance--that Sun has ever been
involved in, and it bodes well for the Galaxies. It will have over 21 TB of
main memory and 1.1 PB of disk storage (which is coming from NEC) in the
initial 50 teraflops configuration. The machines will be lashed together
using InfiniBand interconnect from Voltaire. Sun's N1 System Manager and N1
Grid Engine software will be used to manage the cluster.
This thread is now locked and can not be replied to.