Subscribe / Unsubscribe Enewsletters | Login | Register

Pencil Banner

HP sees GPUs taking more computing load

Agam Shah | April 12, 2011
Graphics processors have always been popular in workstations, but are increasingly handling a larger computational load and bringing supercomputer-like capabilities to complex scientific and math applications, according to a Hewlett-Packard executive.

Graphics processors have always been popular in workstations, but are increasingly handling a larger computational load and bringing supercomputer-like capabilities to complex scientific and math applications, according to a Hewlett-Packard executive.

Graphics processors have been primarily used to support highly visual tools such as computer-aided design, but some new math and science applications can take advantage of hundreds of graphics processor cores in workstations, said Josh Peterson, product marketing manager of workstations for HP. Companies are running complex calculations by harnessing the parallel processing capabilities of CPUs and GPUs, which speeds up workstations while reducing overall power consumption, Peterson said.

GPUs have recently been harnessed in supercomputers to perform specific calculations. The world's fastest supercomputer, the Tianhe-1A system at the National Supercomputer Center in Tianjin, China, uses Nvidia's Tesla GPUs and Intel's Xeon CPUs to deliver 2.5 petaflops of performance. HP on Tuesday started shipping new desktop and mobile workstations with optional Nvidia or Advanced Micro Devices GPUs, which include hundreds of processing cores.

"The next inflection point we see [for workstations] is probably in the GPU space. We're starting to see more software vendors look at the GPU as the next thing to design for," Peterson said.

Demands in the workstation market have changed over the last decade. In the mid-2000s, companies moved from RISC (reduced instruction set computing) chips such as HP's PA-RISC and Sun's Solaris to x86 chips, which delivered improved performance thanks to an increased number of cores and power-efficient design.

CPUs remain important thanks to multiple cores and features like multithreading, but GPUs offer better raw computing power, Peterson said. The faster execution of specific programs on GPUs also helps cut energy costs, which is important to customers, Peterson said.

But there are programming challenges as applications need to be written for parallel execution across GPUs and CPUs with tools such as OpenCL, Nvidia's CUDA or Microsoft's DirectX. HP works with graphics companies such as Nvidia, which provides programming tools, to ensure that applications are certified to work correctly.

Workstations are sophisticated machines that are used for very specific computation, innovation and analysis tasks, said Nathan Brookwood, principal analyst at Insight 64.

Applications such as Autodesk's Autocad and Adobe's Premiere Pro and Photoshop are taking a larger set of workloads off CPUs and moving it to GPUs, Brookwood said. Queries can also be set up for math packages such as MatLab and Mathematica, with calculations taking place on GPUs.

 

1  2  Next Page 

Sign up for Computerworld eNewsletters.