The NSF has finished a $30 million upgrade to its Stampede2 supercomputer at TACC, the Texas advanced computing center at the University of Texas at Austin. The system now peaks at 18 petaflops and comprises 4,200 Intel MIC-based Knights Landing nodes and 1,736 Intel Xeon Skylakes. Access is through NSF’s XSEDE science and engineering development environment. TACC’s Aaron Dubrow told Oil IT Journal, ‘It’s early for Stampede2 results but our oil and gas partners will have some really good data around on Stampede2 by the year end.’
Saudi Aramco, Ansys and KAUST claim to have shattered a supercomputing record, scaling Ansys Fluent across nearly 200,000 processor cores, a 5x increase over the record set three years ago. Simulation of a multiphase gravity separation vessel was performed on the Shaheen II Cray XC40 supercomputer hosted at the KAUST Supercomputing Core Lab. More from Ansys.
A recent presentation by Jack Dongarra (Oak Ridge National Lab)
at the EU DG
Connect symposium addresses the global race for
exascale (over 1,000 petaflops) HPC. The race is taking place in three
‘swim lanes,’ CPU, GPU and ‘lightweight’ cores*. The exaflop target is
expected to be hit some time around 2020. Today, China hosts 36% of the
TOP500.org
systems and has begun making its own processors. Dongarra concluded
with a critique of the ubiquitous Linpack benchmark which no longer
accurately tests HPC performance. The high-performance conjugate
gradient method is a candidate for Linpack’s replacement.
* Like the ARM chips used in cellphones.