SGI announced the availability of a complete, managed GPU solution of its SGI Rackable servers with the new high-performance NVIDIA Tesla K10 GPU Computing Accelerator. Coupled with SGI Management and Performance Suite software, the solution is built, completely integrated, and tested in SGI's manufacturing facility so that it can be installed at the customer site and begin running application codes in less than a day.
The next-generation NVIDIA Tesla K10 GPU Accelerator is designed for throughput and performance per watt. It features two ultra-efficient GK104 GPUs that provide up to 2X performance per watt for single precision applications than its predecessor, the M2090. The NVIDIA Tesla K10 features two GK104 GPUs with 1536 CUDA cores each and 4GB of memory operating at 160 GB/second per GPU. These GPUs are shown to be particularly well suited to single high precision workloads, including seismic processing, signal, image and video analysis, and radio astronomy.
"The research team at the Swinburne University of Technology has benefited from significant improvements in processing power and research output since our SGI Rackable system was installed in July 2011," said Dr. Jarrod Hurley, manager of Swinburne's supercomputer. "That system was over ten times more powerful than its predecessor, and with the introduction of the new NVIDIA Tesla K10 GPU Computing Accelerator we'll look forward to building even greater functionality and performance into our own system as our requirements evolve."
The baseline server for the solution is the SGI Rackable C1104G-RP5, a 1U standard depth chassis, and the SGI 2110-RP5, a 2U standard depth chassis. Each server features one dual-socket server powered by two Intel Xeon processor E5-2600 series, and 8 DIMM slots. The C1104 can accommodate up to four 2.5 inch hard disk drives, three PCI-E 3.0 x 16 double-width expansion slots and dual-port GigE networking controllers, while the C2110 can accommodate up to ten 2.5 inch hard drives, four internal PCI-E 3.0 x 16 double-width expansion slots and dual-port GigE controllers. SGI is also announcing the availability of the SGI GPU Starter Kit with NVIDIA Tesla K10 GPU Accelerator offering a pre-configured bundle of 10 nodes.
"As a leader in design-to-order high performance computing (HPC) solutions with over three decades of experience in accelerators applied to high precision workloads, SGI is uniquely positioned to bring its customers hybrid solutions of industry-standard microprocessors with GPU accelerators," said Bill Mannel, vice president of product marketing at SGI. "Our GPU clusters are all you need for a powerful deployment. We're also announcing the SGI GPU Starter Kit, a 10-server cluster with 20 NVIDIA Tesla K10 GPUs for those who want to start enjoying the benefits of CUDA programming today in a cost-effective solution."
"Adding GPUs to a server delivers significant application acceleration, and this is having a profound effect on many important industry segments," said Sumit Gupta, senior director of the Tesla business at NVIDIA. "SGI's new K10-enabled SGI Rackable servers provide researchers with a powerful and flexible platform upon which to innovate."
The NVIDIA Tesla K10 GPU is available in the SGI Rackable C1104G and C2110G server product lines. The SGI GPU Starter Kit with NVIDIA Tesla K10 GPU Accelerator is available in the SGI Rackable C1104G server product line. All are available for order today.