Talk about a big design win for Nvidia's GPU computing business. Earlier today, the company announced that IBM now offers Nvidia Tesla 20-series GPUs in its iDataPlex servers, making those "the first mainstream high performance computing . . . systems" to have Tesla inside.
Nvidia Tesla General Manager Andy Keane didn't mince words about the move. "IBM's adoption of Tesla for their HPC server line is the most significant milestone in Tesla history," he said. "Scientists worldwide can now access the power of Tesla and CUDA from the world leader in technical computing."
IBM says its iDataPlex dx360 M3 servers are "designed for data centers that require high performance, yet are constrained on floor space, power and cooling infrastructure." The systems support up to two six-core Intel Xeon 5600 processors clocked at 3.06GHz, up to two Tesla GPUs, up to 128GB of DDR3-1333 memory, and up to either eight 2.5" or 12 3.5" storage drives, depending on whether customers go for the 2U or 3U iDataPlex variant. Other options include 10-gigabit Ethernet, Fibre Channel, and 160-320GB PCI Express solid-state drives.
Nvidia's Tesla M2050, meanwhile, has a single Fermi GPU with 515 gigaFLOPS of peak double-precision number-crunching power, 1.03 teraFLOPS of peak single-precision power, and 3GB of GDDR5 memory pushing data through a 384-bit interface, for total bandwidth of 148GB/s. Each Tesla M2050 is rated for a peak power draw of 225W.