IBM officials for the past few years have been working with Nvidia to bring GPU accelerators to some of Big Blue’s Power-based servers used by enterprises and supercomputing organizations, and earlier this year said Nvidia’s Tesla K80 GPUs would be used in bare-metal server in its SoftLayer cloud environment.
GPU acceleration using Nvidia graphics products also are part of IBM’s vision for the OpenPower Foundation and the use of the system architecture in the high-performance computing (HPC) world.
Now, the company is again looking to leverage Nvidia’s GPUs, launching a portfolio of scale-out Linux-based systems running on its Power 8 processors that will offer Tesla GPUs to help push the performance of the servers without driving up the power consumption.
At the LinuxCon Europe 2015 show in Ireland Oct. 5, IBM officials unveiled the three new systems, part of the company’s vision of “waitless computing,” where workloads can be rapidly processed. The new servers are all preconfigured, scale-out offerings.
The S812LC server uses either an eight-core 3.32GHz Power 8 chip or a 10-core 2.92GHz processor, with up to 1TB of system memory and up to 14 3.5-inch internal storage drives. The S822LC offers similar specs but comes with two chips, for up to 20 cores, and two 2.5-inch drives. It’s aimed at commercial computing, according to IBM officials.
The S822LC for HPC includes the same features as the S822LC, but also includes the Nvidia GPUs, which IBM officials said helps create visual interpretations of data quickly and more easily than previous systems. With the LC systems, customers can take the servers out of the box and have them running within 30 minutes, officials said.
For several years, organizations in the HPC space have been turning to GPU accelerators and x86 co-processors to help improve the performance and energy efficiency of their systems. According to the organizers of the twice-yearly Top500 list of the world’s fastest servers, 90 of the systems on the most recent list released in June used accelerators, up from 75 on the list from November 2014. Fifty-two of them used Nvidia GPUs while another four used Radeon graphics from Advanced Micro Devices. Thirty-five used Xeon Phi co-processors from Intel. Four of them used a combination of Nivida and Xeon Phi chips.
The IBM announcement came days after Nvidia officials said Microsoft will use Nvidia’s Grid 2.0 virtualized graphics for applications and accelerated computing in systems for its Azure cloud environment.