14 DAY TRIAL // NVIDIA launched the Tesla K40 GPU-based PCI Express compute accelerator not too long ago, so, naturally, makers of supercomputers are adopting it. In this case, it is Cray that chose to avail itself of its services.
Cray is one of the oldest names in the high-performance computing industry, having set up many supercomputers over the years.
Whenever the Top 500 list is published, Cray either has the top of the list or one (or more) of the top five spots under its belt.
Ever since CPU computing took off, Cray has been using NVIDIA Tesla adapters to carry out parallel computing tasks.
The Tesla K20 was the best one available (and the only one used really) in such systems, until recently. Now, though, the K40 has risen, and it is getting contracts quickly.
Cray will use it in the CS300 series of scalable cluster supercomputers, which act like building blocks for large conglomerates, both air- and water-cooled. The Cray Advanced Cluster Engine ensures smooth network, server, cluster, and storage management for them.
With the addition of Tesla K40, parallel compute performance will rise drastically. A single Tesla K40 is 40% better than a K20 after all, so the overall performance advantage should be similar.
"The NVIDIA Tesla GPU accelerators are the highest performance, most energy-efficient accelerators we have ever built," said Sumit Gupta, general manager of Tesla Accelerated Computing products at NVIDIA.
"With Tesla accelerators added to the Cray CS300 cluster systems and Cray XC30 supercomputers, Cray customers now have additional options to accelerate a wide range of compute- and data-intensive applications and workloads."
For those who want to know about the specs of the Tesla K40, here they are: 2,880 CUDA cores (Kepler), 12 GB GDDR5 VRAM, and 745 MHz GPU clock (810 – 875MHz Boost Clocks).
The Cray CX30 “Cascade” supercomputers will integrate Tesla K40, since they are the company's most advanced systems and need the best hardware, always.