A class of processors based on a new graphics processing unit (GPU) has been launched under the NVIDIA Tesla brand, which the company claims "will place the power previously available only from supercomputers in the hands of every scientist and engineer," transforming today's workstations into "personal supercomputers." The Tesla products are designed to fit into existing high-performance computing environments and key elements of the brand include a multithreaded architecture with a 128-processor computing core and the CUDA SDK.
CUDA is the C-language development environment for the GPU and includes a C compiler, debugger/profiler, dedicated driver, and standard libraries. CUDA uses Standard C to create programs that process large quantities of data in parallel. As such, programs written with CUDA and run on Tesla are able to process thousands of threads simultaneously, providing high computational throughput to quickly solve complex computational problems. The CUDA development environment is currently supported on Linux and Windows XP.
The NVIDIA Tesla family includes:
- A Computing Processor, which is a dedicated computing board that scales to multiple Tesla GPUs inside a single PC or workstation. The Tesla GPU features 128 parallel processors and delivers up to 518 gigaflops of parallel computation. The GPU computing processor can be used in existing systems, partnered with high-performance CPUs.
- A Deskside Supercomputer, which is a scalable computing system that includes two Tesla GPUs and attaches to a PC or workstation through an industry-standard PCI-Express connection. With multiple deskside systems, a standard PC or workstation can deliver up to 8 teraflops of compute power to the desktop.
- A GPU Computing Server, which houses up to eight NVIDIA Tesla GPUs, containing more than 1000 parallel processors that add teraflops of parallel processing to clusters -- bringing GPU computing to the datacenter.
"Today's science is no longer confined to the laboratory; scientists employ computer simulations before a single physical experiment is performed. This fundamental transition to computational methods is forging a new path for discoveries in science and engineering," said NVIDIA's Jen-Hsun Huang.
The new hardware and CUDA have been deployed in both academic and application development environments where impressive success has been measured. "Many of the molecular structures we analyze are so large that they can take weeks of processing time to run the calculations required for their physical simulation," said John Stone, senior research programmer at the University of Illinois Urbana-Champaign. "NVIDIA's GPU computing technology has given us a 100-fold increase in some of our programs, and this is on desktop machines where, previously, we would have had to run these calculations to a cluster."
