14 DAY TRIAL // We covered the release of the Tesla K40 not too long ago, and it's clear that the GPU compute accelerator is a very powerful piece of work. Supermicro agrees with us, and NVIDIA itself for that matter.
Supermicro has just released a super server featuring a 4U chassis and support for no fewer than eight of those new Tesla cards.
The server, 4U 8x GPU SuperServer, has two cooling zones, one for the CPUs (two Intel Xeon E5-2600 v2 "Ivy Bridge" processors at 150W TDP) and GPU areas (up to eight, for 300W thermal load).
Two rows of fans at the middle of the case separate the CPU/GPU cooling zones.
In addition to the 8 PCIe Telsa cards, though, there is room for full-length 2x PCI-E 3.0 x8 and 1x PCI-E 2.0 x4 slots.
There are 24 ECC DDR3 1600 MHz DIMM memory slots too (up to 728 GB capacity), a pair of 10GbaseT or GbE ports, one dedicated OPMI 2.0 port, and hot-swappable SAS/SATA/SSD bays (up to 25 drives can be installed at once).
One, two, three, or four redundant 80 Plus Platinum power supplies keep everything running (1600W power output for each). It ultimately depends on what hardware you install.
That said, SuperMicro also added support for Tesla k40 to the 1U, 2U, 3U SuperServers, FatTwin, SuperWorkstations, and SuperBlade platforms.
"The Tesla K40 GPU accelerator provides double the memory and 10 times higher performance than today's fastest CPUs, enabling enterprise data center and HPC customers to solve their most complex engineering and big data analytics computing challenges," said Sumit Gupta, general manager of Tesla Accelerated Computing Products at NVIDIA.
"When combined with Supermicro's high-density, scalable systems, the new Kepler-based accelerators deliver high performance computational horsepower with maximum energy efficiency."
Supermicro's complete line of GPU-enabled servers and supercomputing platforms can be found on the official website.