New Intel chip hailed by upstream developers

Ansys, CMG, Schlumberger report significant speedup from parallelizing and Xeon 5500 chip.

In an interesting exchange on the new LinkedIn Petrel user group, Schlumberger’s Russ Sagert offered some advice on getting the most out of the multi-threading capabilities of modern processors. Parallelizing Petrel is work in progress. Today, geophysics is the most highly threaded domain and Sagert recommends using as many cores as you can. Other applications, such as kriging and volumetrics, are not yet ready for multi-core, but Sagert expects this to be fixed ‘real soon now.’ His recommendation (for modeling) is to go for a fast quad core processor in the medium term. Sagert also noted that the new Intel Xeon 5500 Nehalem chips are ‘vastly superior’ to the 5400 series. Parts of Petrel and ECLIPSE that have been parallelized are seeing three-fold speedup. For virtual machine environments he suggests checking out Parallels Workstation Extreme VM software which has shown six fold speedup on the 5500 chip.

Other companies see similar speedup on the new Intel chip. Engineering software specialist Ansys reports speedup of ‘up to’ 2.15 times over previous platforms for its mechanical and computational fluid dynamics packages. Computer Modeling Group has successfully leveraged the 5500’s on-chip memory controller and improved cache, to effect a ‘big jump’ in parallel performance compared to the 5400 for its OpenMP-based applications. CMG president Ken Dedeluk also noted Intel’s ‘excellent support’ for the company’s parallelization effort. Schlumberger’s Eclipse reservoir simulation unit has also been trialling the new processor and have found it ‘up to’ 3.13 times faster than the 4400 chip running its black oil simulator.

Where does that leave AMD? Intel’s competitor is still in the race and, to celebrate the Opteron processor’s sixth birthday this month, was showing a new 12-core ‘Magny-Cours’ processor to launch in Q1 2010. A 16-core, 32 nanometer ‘Bulldozer’ architecture will follow in 2011. By which time, Intel’s own multi-core Larabee chip should be in production. So it will be parallel or die (or maybe just parallel on-die?)

Click here to comment on this article

Click here to view this article in context on a desktop

© Oil IT Journal - all rights reserved.