Novel processor accelerates compute-intensive simulations

Complex reservoir and seismic-data simulations are commonplace in today’s oil and gas exploration sector, and it is equally common to wait weeks for an answer. This is set to change with Acceleware’s solution based on graphics processing unit (GPU) technology, which reportedly accelerates processing of large data sets while also providing significant power savings.

Like many industries, the oil and gas exploration and production sector relies on computationally intensive simulations and processing capabilities to make business decisions. While hardware manufacturers continue to develop more powerful processors to run these simulations more efficiently, much of the simulation software is not currently configured to take full advantage of these efficiency gains.

“Acceleware’s business model is to partner with software vendors such that they can rapidly bring their solutions to market and take greater advantage of the higher processing efficiencies offered by more advanced processors,” said Steven Joachims, Vice President of Business Development for Acceleware.

“Hardware providers are continually developing new multicore central processing units (CPU’s), which essentially function as separate brains,” Joachims continued. “These multiple processors are increasingly operating in parallel, rather than having to operate in a serial fashion. This dramatically increases the potential processing speed for a software application, but if that application has not been reconfigured to run in parallel, then many of these multiple cores will sit idle and not get engaged in the computation.”

Acceleware engineers have addressed this problem with an integrated hardware/software solution. “This company was founded by university researchers who had the same processing limitations as our customers,” said Robert Miller, Vice President of Marketing and Product Management for Acceleware. “They settled on using GPU’s as their hardware of choice, because of the advanced parallel-core processing that GPU’s deliver.”

GPUs offer dramatic parallel-processing power

While the newest CPU’s might have 4 to 8 parallel cores, GPU’s currently possess 128 cores. “By using the graphics processors in conjunction with a CPU, we can deliver parallelism that provides an order of magnitude boost for computationally intensive applications,” Joachims said.

Acceleware does not make the graphics hardware, but instead has focused on strategic partnerships with GPU manufacturers. In early 2007, Acceleware partnered with NVIDIA, one of the largest global manufacturers of GPU’s, for the hardware component of the technology offering. Currently, the processors are available in a desk-side (the DS30 with two graphics cards) or a rack-mounted (the QS30 with four cards) configuration, and are easily plugged into existing computer units containing an available PCI Express x16 slot.

The software component of the processing solution comes from Acceleware’s own engineers. “Our software acts as the interface between the data-processing software from independent software vendors and the GPU hardware that will be running their applications,” Miller said. “It contains libraries of programs designed to help string the particular vendor software’s algorithms in multiple sequences and take advantage of the parallel processing routes afforded with the GPU. From an end user perspective, it is the same simulation program or algorithm, but it is running much faster.”

Compared to a simulation run on a single processing core, Acceleware reports at least a 20-fold increase in processing speed using the DS30, and a greater than 40-fold increase using the QS30. For very large seismic processing jobs, the DS30 and QS30 typically provide a 10-fold and 20-fold increase in processing speed, respectively, when added to an existing CPU processor.

Power and space reductions are significant

Acceleware reports dramatic power consumption savings as well. “One-half of a rack of our hardware and software gear can deliver the same performance as 3 or 4 racks of CPU’s,” Joachims said. “This translates to five to six times lower density in terms of rack space and power required.”

This power and space savings is a particular benefit to medium-sized clients in the Calgary market. “Many of these companies have a power limit imposed upon them by the local power generation utility, which is designed to curb the amount of residue or waste they can generate,” Joachims continued. “They have space-constrained computing rooms as well, so our ability to provide the necessary computing requirements at lower space and power consumption is a greater value proposition for them.”

While Acceleware has supplied its accelerated processing technologies for a range of applications in industries as diverse as telecommunications and the automotive sector, the company’s oil and gas focus has centered on two applications. “The two major compute-intensive applications we have focused on are seismic data processing and reservoir simulation,” said Joachims. “While both of these markets are projected to grow by 15-20% in the coming years, we have put more focus on the seismic processing, since it is currently the larger of the two.”

The industry workhorse for seismic data processing is widely considered to be migration, in particular Kirchhoff Pre-Stack Time Migration (KPSTM), a robust method for imaging complex geological structures. It is here that Acceleware has devoted much of its effort, developing seismic libraries that are specifically tuned to address KPSTM data in order to increase processing speed with GPU’s.

Pages: 1 2