Re:Intel SPMD: Compiler for High-Performance SIMD Programming
by semi-extrinsic on Monday April 11, 2016 @10:58AM
(#11471920)
The problem with having 100 x86 CPU on one board is NUMA becomes a bad problem. When memory accesses from one CPU to different regions of memory will have quite different bandwidths and latencies, you're much better off by acknowledging that upfront, designing a proper interconnect (Infiniband), and not sharing memory between threads but rather communicating explicitly (MPI).
[ Parent
| on HN ]
Re:Intel SPMD: Compiler for High-Performance SIMD Programming
by CyberDildonics on Monday April 11, 2016 @11:09AM
(#11472001)
The next Xeon Phi has an Intel designed interconnect and a layer of 16GB high bandwidth memory between the main memory and the processor.