NVIDIA Launches Tesla K40 GPU Accelerator

NVIDIA (NVDA) unveiled the NVIDIA Tesla K40 GPU accelerator, which the company called its most efficient architecture ever, achieving 4.29 teraflops single-precision and 1.43 teraflops double-precision peak floating point performance. The new K40 features double the memory and up to 40 percent higher performance than the K20X GPU, and 10 times higher performance than today’s fastest CPU.

“GPU accelerators have gone mainstream in the HPC and supercomputing industries, enabling engineers and researchers to consistently drive innovation and scientific discovery,” said Sumit Gupta, general manager of Tesla Accelerated Computing products at NVIDIA. “With the breakthrough performance and higher memory capacity of the Tesla K40 GPU, enterprise customers can quickly crunch through massive volumes of data generated by their big data analytics applications.”

Key features of the Tesla K40 GPU include 12GB of GDDR5 memory, 2,880 CUDA parallel processing cores, dynamic parallelism, and PCIe Gen-3 interconnect support. The Tesla K40 GPU accelerates the broadest range of scientific, engineering, commercial and enterprise HPC and data center applications. The NVIDIA Tesla K40 GPU accelerator is available immediately.

TACC, HP, NVIDIA partner for remote visualization project

The Texas Advanced Computer Center (TACC) at The University of Texas at Austin, along with technology partners HP and NVIDIA,  announced that they will deploy Maverick in January 2014, a unique, powerful, high performance visualization and data analytics resource for the open science and engineering community. In addition to launching Maverick in January 2014, TACC this month is deploying Stockyard, a 20 petabyte large-scale global file system. Other systems for storing and analyzing data sets and for hosting web portals and gateways that provide access to scientific data will be announced in 2014.

The Maverick system will be comprised of five racks containing 132 HP ProLiant SL250s Gen 8 compute nodes and 14 HP ProLiant management, login, and Lustre router servers. Each of the 132 compute nodes will include two ten-core Intel Xeon E5-2680 V2 processors with 256GB of DDR3 1866MHz memory each, a Mellanox Connect-X3 FDR InfiniBand FlexibleLOM adaptor, and one NVIDIA Tesla K40 GPU accelerator. A Mellanox FDR InfiniBand interconnect will provide a high-performance communication platform.

“This system will be great for Big Data analysis — every node in Maverick will have large memory, a state-of-the-art GPU accelerator, and be connected to massive data storage,” said Niall Gaffney, TACC’s director of Data Intensive Computing. ”Data scientists and all researchers will be able to use visual analysis techniques to explore data.”

Partner support

The Tesla K40 GPU will be available in the coming months from a variety of server manufacturers, including Appro, ASUS, Bull, Cray, Dell, Eurotech, HP, IBM, Inspur, SGI, Sugon, Supermicro and Tyan, as well as from NVIDIA reseller partners.

• SGI announced the availability of NVIDIA Tesla K40 GPU accelerators in fully managed and integrated solutions across its entire server product line. Tesla K40 GPU accelerators are available in UV 2000, Dense Rackable servers, the Rackable C2108 and UV 20 servers, and they can be hosted in an SGI ICE X fabric via Rackable service nodes. ”NVIDIA’s accelerators enable our customers to realize significant improvements in processing performance,” said Bill Mannel, general manager, Compute at SGI. “Accelerator-based HPC solutions feature intelligent NVIDIA GPU Boost technology, which converts power headroom into a user-controlled performance boost, enabling our customers to unlock the untapped performance of a broad range of applications to address compute and Big Data challenges.”
• Cray announced the Cray CS300 line of cluster supercomputers and the Cray XC30 supercomputers are now available with the NVIDIA Tesla K40 GPU accelerators. ”The addition of the NVIDIA K40 GPUs furthers our vision for Adaptive Supercomputing, which provides outstanding performance with a computing architecture that accommodates powerful CPUs and highly-advanced accelerators from leading technology companies like NVIDIA,” said Barry Bolding, vice president of marketing at Cray. “We have proven that acceleration can be productive at high scalability with Cray systems such as ‘Titan’, ‘Blue Waters’, and most recently with the delivery of a Cray XC30 system at the Swiss National Supercomputing Centre (CSCS). Together with Cray’s latest OpenACC 2.0 compiler, the new NVIDIA K40 GPUs can process larger datasets, reach higher levels of acceleration and provide more efficient compute performance, and we are pleased these features are now available to customers across our complete portfolio of supercomputing solutions.”
• Supermicro debuted new 4U 8x GPU SuperServer that supports the new and existing active or passive GPUs (up to 300W) with an advanced cooling architecture that splits the CPU (up to 150W x2) and GPU (up to 300W x8) cooling zones on separate levels for maximum performance and reliability. In addition, Supermicro has 1U, 2U, 3U SuperServers, FatTwin, SuperWorkstations and SuperBlade platforms ready to support the new K40 GPU accelerator.
• Cirrascale announced it will offer the NVIDIA Tesla K40 GPU accelerator throughout its GPU-enabled blade server and high-performance workstation product lines. Utilizing a pair of the company’s latest proprietary 80-lane Gen3 PCIe switch-enabled risers, the GB5400 supports up to eight discrete NVIDIA Tesla K40 Accelerator cards in a single blade. “As always, NVIDIA is pushing the performance envelope with its latest GPU accelerator,” said David Driggers, CEO, Cirrascale Corporation. “Our customers and licensed partners in HPC are moving rapidly to take advantage of this increased performance, and want to ensure they can scale the solutions they choose. We’re confident the Tesla K40 GPU with our latest Gen3 switch-enabled riser meets these needs.”