Advanced Micro Devices is teaming up with SiSoftware to create a way for developers and businesses to measure the performance of systems using the OpenCL standard.
AMD and Nvidia have been driving graphics processors into more mainstream computing environments, and officials with both vendors have said that organizations running heavy computational workloads are turning more often to systems with GPUs (graphic processing units) or a combination of GPUs and CPUs.
AMD, which makes GPUs through its ATI business, offers a development framework based on the OpenCL standard, which lets applications be run in both GPU and CPU environments.
Through the SiSoftware partnership, announced Dec. 3, organizations will now have a benchmark for measuring performance.
“This is a major development OEMs and developers,” Michael Chu, product manager for AMD’s Stream computing software, said in an interview.
The OpenCL GPGPU benchmark suite is part of SiSoftware’s Sandra 2010 offering, which includes remote analysis, benchmarking and diagnostic features for PCs, servers, mobile devices and networks. It now can test OpenCL performance on AMD’s ATI Stream technology.
AMD officials said the company has optimized the performance of the OpenCL benchmarks for its GPU implementations. AMD also offers its ATI Stream Software Development Kit for OpenCL.
Officials with AMD, which bought graphics chip maker ATI in 2006 for $5.4 billion, sees the company’s graphics capabilities as a key differentiator in its competition with Intel. The chip maker also is becoming a larger rival to GPU maker Nvidia, which is using its CUDA technology to drive its graphics chips into mainstream computing environments.
Both AMD and Nvidia see opportunities in the HPC (high-performance computing) space, where there are large numbers of highly parallel computing workloads.
Intel also is looking to increase the graphics capabilities of its CPUs. Its upcoming “Larrabee” GPGPU is expected to compete with AMD’s Radeon and Nvidia’s GeForce products.
At the Supercomputing 2009 show Nov. 17, Intel demonstrated an overclocked Larrabee chip topping the 1 teraflop (trillion floating point calculations per second) mark.
However, despite that performance and the GPU push by AMD and Nvidia, Intel officials don’t see a strong a need for graphics processors in mainstream computing. In an interview last month, Boyd Davis, general manager of Intel’s server platforms group marketing, said that when Intel’s upcoming eight-core “Nehalem EX” Xeon chip is released early next year, organizations with highly parallel workloads will be able to run many of them on the CPU.
The demand for GPU-CPU co-processing is limited, Davis said.
“It’s really pretty narrow workloads,” he said. “They are important, but it’s a minority of the workloads.”