Fighting a software war
Also at Supercomputing '09, NVIDIA announced it had teamed up with Mellanox Technologies and unveiled software it claimed increases cluster application performance by as much as 30 per cent by reducing latencies which occur when communicating over Mellanox InfiniBand to servers equipped with NVIDIA Tesla GPUs.
NVIDIA said that whilst the system architecture of a GPU-CPU server requires the CPU to initiate and manage memory transfers between the GPU and the InfiniBand network, the new software lets Tesla GPUs transfer data to pinned system memory that Mellanox InfiniBand can read and transmit over the network, resulting in increased overall system performance.
"In GPU-based clusters, most of the compute intensive processing is running on the GPUs," explained Gilad Shainer, Mellanox's director of high performance computing and technical marketing.
"It's a natural evolution of the system architecture to enable GPUs to communicate more intelligently over InfiniBand. This helps create a computing platform that will enable future Exascale computing and dramatically increase performance for a broad spectrum of applications," he added.
In other words, "this is all software, so don't fight a hardware war," said NVIDIA's Keane in simpler terms.
NVIDIA says the software will be available in its CUDA architecture toolkit beginning in Q2 2010 and will work on existing Tesla S1070 1Us, Tesla M1060 module-based clusters and also with the new Tesla 20-series S2050 and S2070 1U systems.
