Since 1986 - Covering the Fastest Computers in the World and the People Who Run Them

Language Flags
April 3, 2014

How RSC Pushed PetaStream To 1.2 Petaflops With Xeon Phi

Timothy Prickett Morgan
xeon_phi555

Russian supercomputer maker RSC Group is pushing the compute density limits to the extreme again. The company has upgraded its PetaStream system, which debuted at the SC13 conference last year, with a new variant of the Xeon Phi X86 coprocessor from Intel.

With the upgrade, RSC is able to push the peak theoretical performance of a single rack of the PetaStream system by 20 percent, to 1.2 petaflops. The increase in floating point processing power is made possible by shifting to the top-end Xeon Phi 7120D coprocessor, which Intel quietly delivered in March. RSC says that it is the first of the supercomputer makers to get its hands on this device.

The Xeon Phi 5120D that RSC deployed in the original PetaStream machine was itself launched at the ISC13 supercomputing event last June, and it is a stripped down version of the Xeon Phi card that does not have its own fan or cooling heat sinks. Rather, it relies on the packaging and cooling of the system in which it is embedded to keep it from overheating. This allows companies like RSC to cram components a lot more tightly than is possible using the actively cooled packaging common with graphics cards, GPU coprocessors, and other Xeon Phi cards.  The Xeon Phi cards with a D designation are for the highest density, while the X models, which also are bare bones devices, are slightly larger physically; cards with the P designation have passive cooling (heat sinks and enclosures) and those with the A attached to their name have active cooling (heat sinks, enclosures, and a fan).

That Xeon Phi 5120D coprocessor had 60 Pentium-style cores running at 1.053 GHz plus 8 GB of GDDR5 memory with 352 GB/sec of peak memory bandwidth. This coprocessor dissipates 245 watts and delivered 1.011 teraflops of double-precision floating point operations per second. The PetaStream machine pairs eight Xeon Phi coprocessors with a single ten-core Xeon E5-2690 v2 processor running at 3 GHz to make a compute module, which has liquid cooling for all of the components inside of the enclosure and which is designed to pull 400 kilowatts of dissipated energy from those components. The PetaStream rack has 128 of these compute modules, for a total of 1,024 Xeon Phi coprocessors, and that works out to 61,440 cores and 245,760 threads packed into a rack that is 7.2 feet tall and measures 3.28 feet on a side.

With the upgrade to the 7120D, which just started shipping in the first quarter, the Xeon Phi embedded in the PetaStream system has all 61 cores on the die operational and they run at 1.238 GHz. That yields 1.208 teraflops of double precision math per card, and the machine also has 16 GB of GDDR memory so for applications that are memory constrained, this will be useful. (The memory bandwidth stays at 352 GB/sec however.) This device is also rated at 270 watts, so it burns more electricity and generates more heat.

That extra 20 percent of performance could come at a price, however, if list price from Intel is any guide. The Xeon Phi 5120D has a recommended end user price of $2,759, while the 7120D costs $4,235 each. That is an incremental 53.5 percent increase in the cost of the coprocessor for a 20 percent increase in raw floating point performance. This is not necessarily a premium that customers will want to pay. But then again, the cost of number crunching with Xeon Phi coprocessors can be so much lower than for Xeon CPUs (depending on the workload, of course) that there could be customers who will want this top-shelf performance. The point is, RSC is giving them options and, more importantly, the gap between the prices of the two Xeon Phi cards may not be as large as the one-off list price shows.

Oleg Gorbachov, spokesperson for RSC, tells HPCwire by email that the company is still putting the PetaStream machine through the paces and is running benchmark tests on various workloads. These include the GROMACS molecular dynamics simulator maintained by the Royal Institute of Technology and Uppsala University in Sweden; the LAMMPS molecular dynamics simulator out of Sandia National Laboratories; and the MAGMA linear algebra libraries out of the University of Tennessee at Knoxville. RSC hopes to present results of these tests at the ISC14 conference in Leipzig, Germany in June.

As for customers, Gorbachov says that RSC is working with potential PetaStream customers right now and hopes to announce its first deal this year.

Tags: , ,

SC14 Virtual Booth Tours

AMD SC14 video AMD Virtual Booth Tour @ SC14
Click to Play Video
Cray SC14 video Cray Virtual Booth Tour @ SC14
Click to Play Video
Datasite SC14 video DataSite and RedLine @ SC14
Click to Play Video
HP SC14 video HP Virtual Booth Tour @ SC14
Click to Play Video
IBM DCS3860 and Elastic Storage @ SC14 video IBM DCS3860 and Elastic Storage @ SC14
Click to Play Video
IBM Flash Storage
@ SC14 video IBM Flash Storage @ SC14  
Click to Play Video
IBM Platform @ SC14 video IBM Platform @ SC14
Click to Play Video
IBM Power Big Data SC14 video IBM Power Big Data @ SC14
Click to Play Video
Intel SC14 video Intel Virtual Booth Tour @ SC14
Click to Play Video
Lenovo SC14 video Lenovo Virtual Booth Tour @ SC14
Click to Play Video
Mellanox SC14 video Mellanox Virtual Booth Tour @ SC14
Click to Play Video
Panasas SC14 video Panasas Virtual Booth Tour @ SC14
Click to Play Video
Quanta SC14 video Quanta Virtual Booth Tour @ SC14
Click to Play Video
Seagate SC14 video Seagate Virtual Booth Tour @ SC14
Click to Play Video
Supermicro SC14 video Supermicro Virtual Booth Tour @ SC14
Click to Play Video