After the announcement of POWER-based systems and in line with its strategy to further emancipate itself from Intel, NVIDIA took advantage of ISC this week to announce the integration of its Tesla K20 accelerator with the first ARMv8-64 based servers.
Initially available in July as development platforms from three manufacturers – Cirrascale, E4 and Eurotech – production servers should be out later this year. These machines will be based on Applied Micro’s X-GeneARM64 consisting of eight cores clocked at 2.4 GHz. Given the power of K20, the alliance offers up to 1 Tflops performance on a reduced footprint for a completely Intel-free GPU-accelerated computing solution.
Now, buyers and planners beware. Due in 2015, NVIDIA’s own Tegra SoC, although not dedicated to supercomputing, will also feature a CUDA-compatible, 192-core Maxwell GPU and 32-bit ARM cores within the same chip, with 64-bit versions to follow probably very soon. We haven’t reached the stage of a fully integrated HPC solution yet, but the path is now clearly set.
© HPC Today 2020 - All rights reserved.
Thank you for reading HPC Today.