NVIDIA (http://rbi.ims.ca/5719-542) released the second generation of its fully programmable graphics processing unit (GPU), which serves as the foundation for a new Tesla platform aimed at delivering 2X performance boosts for scientific and graphics-intensive applications, including simulation and 3-D CAD.
The new Tesla T10P Processor packs 240 processing cores and the ability to attach 4 Gbyte of memory, enabling it to deliver 1 teraflop of processing power for CAE/CAD, medical imaging, oil and gas and other applications that demand high-performance computing horsepower for processing large data sets. The Tesla T10P is also the second-generation platform to support CUDA, NVIDIA's C language-based programming environment for its GPU line, which makes it fully programmable and more accessible to meet the needs of a variety of scientific applications.
Leveraging GPUs like the new Tesla T10P is a more cost-effective way to boost high-performance computing horsepower compared to the traditional way of adding additional servers to a data center, according to Sumit Gupta, NVIDIA's product manager for the Tesla business unit. “Computing that was happening on a workstation platform moved to the data center because applications became more intensive and the workstations couldn't keep up,” he says. “What that meant to scientists and engineers was if they didn't work for an organization that could afford a data center, they didn't have access to those precious resources. GPUs (like the Tesla T10P) bring computing back to the workstation.”
Boosting horsepower with GPUs versus traditional multi-core CPUs has other advantages, according to Gupta. GPUs are faster and have shared memory, therefore there are faster interconnections on a chip allowing for better performance than you might have scaling with multiple servers, he says.
The Tesla T10P chip is available as a single GPU, the Tesla C1060, for $1,699; a four-card, rack-mounted system, the Tesla S1070, is also available for $8,000.