Super Micro Computer Inc. has launched the second generation of its GPU-optimized server solutions.
Optimized for the new NVIDIA Tesla 20-series GPUs (based on “Fermi” architecture), the product line includes a very fast 1U server, a 4U/tower that supports four NVIDIA Tesla GPUs (and three additional PCI-e cards), and a 2U Twin server, which supports two hot-pluggable GPU nodes with redundant power.
Supermicro president and CEO Charles Liang says the company has made “the world’s fastest servers even faster” by optimizing the HPC product line to support the latest Tesla 20-series GPUs.
“Our dual-GPU 1U SuperServer now delivers up to 1 TeraFLOPS double-precision performance,” Liang explains, “making it suitable for a far wider range of high-performance computing applications, where more complex data calculations can be performed faster.”
NVIDIA general manager, Tesla business, Andy Keane says the optimized products from Supermicro improves integration of GPUs into large, scale-out deployments in commercial and scientific computing data centers. “Servers with Tesla GPUs provide maximum compute-density for the data center,” Keane says, “delivering the same compute performance as a CPU-only cluster at one-tenth the cost and one-twentieth the power consumption with the ability to scale to thousands of nodes.”
Supermicro’s new GPU product line features both Intel Xeon 5600 Series (Westmere) processors as well as AMD Opteron 6100 Series (Socket G34) processors.
With multiple x16 non-blocking native PCI-Express 2.0 connectivity, 1400-watt Gold Level (93% efficiency) power supplies, and Supermicro’s advanced thermal design, these parallel, multi-GPU systems have been optimized for a wide range of graphics and computationally intensive applications.
Supermicro calls its 6016GT-TF-FM205 Tesla-based server “the fastest 1U server in the world.” It supports dual Intel Xeon 5600 series processors and features two NVIDIA Tesla M2050 GPUs via two Gen2 PCI-Express x16 connections.
The 7046GT-TRF-FC405, on the other hand is suitable for cluster configurations and personal supercomputing. It is housed in the SC747TQ-R1400, Supermicro’s 4U rackmount convertible tower chassis. This chassis supports up to 11 full-height, full-length expansion cards, eight hot-swappable 3.5-inch SAS/SATA drives, and special design features that bolster graphics and computationally intensive applications.
The 2U Twin GPU server introduces an innovative architecture with hot-swappable devices. Each computing node features onboard QDR InfiniBand for 40 Gb/second high-bandwidth connectivity and supports six hot-swap 3.5-inch SATA drives for I/O performance.