Nvidia combines Tesla GPU with 64bit ARM chips for HPC systems

Nvidia has made it possible to use its Tesla video cards in combination with 64-bit SOCs based on the ARM architecture. Next month, the first high-performance computing systems with the arm chips and Tesla K20 accelerators should appear.

Until now, Tesla business cards with x86 and IBM Power chips could be combined: ARM chips were not that interesting for HPC applications. However, now that 64-bit models are appearing, which offer more performance but, above all, can use more memory, ARM is also becoming interesting for HPCs.

Nvidia brings support for the 64-bit ARMv8 architecture to the Tesla accelerators when using the Cuda 6.5 platform for parallel compute programming. However, it is not yet known when Cuda 6.5 will be released exactly. In any case, the first test systems for development purposes should appear in July, followed by production systems later this year.

Nvidia lists three models: the Cirrascale RM1905D with two Tesla K20 cards in a single 1u server, the E4 EK003, which also offers two cards in a 3U form factor with two motherboards, and a system from the Eurotech Group, which includes provides liquid cooling.

At the same time, Nvidia has abandoned its plans to develop its own 64-bit processor for servers, PCWorld writes. The company says it focuses on Tesla chips for mobile and embedded applications and on the Tesla accelerators for hpc.