Supermicro and Nvidia took the wraps off a class of server that's turbo charged by graphics processors.
At the Computex trade show in Taiwan, Supermicro is demonstrating a server that features Nvidia's Tesla GPUs with multi-core processors in a single 1U rack server.
Supermicro is the first to market with its Tesla-based server that reportedly delivers 12-times the performance over a typical quad-core server. Supermicro's Nvidia Tesla-based hardware is dubbed the SuperServer 6016T-GF-TM2.
Here's the math behind the 12-times performance claim. A traditional 1U CPU server with two CPU sockets and four cores per CPU has 170 Gigaflops of computing capability. The peak single precision Flops for a quad-core Xeon R5400 CPU is 85 Gigaflops. The standard 1U CPU-only server has 2 x 85 Gigaflops, or 170 Gigaflops.
A Supermicro SuperServer with two Tesla M1060 GPUs and two quad-core CPUs has 2.0 Teraflops of computing capability. Two Tesla GPUs = 2 x 932 Gigaflops = 1.86 Teraflops. In this configuration, three of four cores in each quad core can be used for compute. 2 x .75 x 85 Gigaflops = 127.5 Gigaflops. GPU + CPU = 1.86 Teraflops + .127 Teraflops = 2.0 Teraflops.
These servers are used in corporate datacenters that support graphics-intensive operations like oil and gas exploration. In a statement, the two companies cited Petrobras a customer.