The company said the Tesla S1070 is a so-called 'GPU computer' and claimed that it offers higher performance with lower power consumption than CPU-only systems. Each Tesla box contains up to four GPUs and multiple units can be configured into computing clusters incorporating onboard communications links and data storage, Infiniband switches and cabling.
Tesla offers a preconfigured cluster of four S1070 boxes that Nvidia said is capable of delivering up to 16 teraflops, or one teraflops per GPU in the cluster.
Nvidia said that the French bank BNP Paribas has installed two Tesla clusters.
We're sceptical that much useful programming infrastructure exists to take advantage of Nvidia's GPU-based compute resources, and suspect that initial applications that might be using such GPU HPC systems are very likely hand-coded, in a FPGA assembly language dialect, by elves.