IBM SoftLayer launches servers with Nvidia Tesla K80 GPU accelerators

IBM today announced the availability of new servers packing Nvidia’s Tesla K80 GPU (graphic processing unit) accelerators in the IBM SoftLayer public cloud. IBM is the first cloud provider to start offering dedicated, “bare metal” servers featuring K80s, as a result of exclusive partnership with Nvidia, a SoftLayer spokesman told VentureBeat.

Until now, IBM SoftLayer has offered servers featuring Nvidia’s Tesla K10 and Grid K2 GPUs that run in conjunction with dual Intel Xeon chips that are not running virtualization software. While virtualization does offer economic benefits in the sense that companies can run multiple applications on a single physical server, it can inhibit performance, which is why IBM offers dedicated, non-virtualized servers, including those with GPU accelerators. The Nvidia Tesla K80s are merely the latest-generation Tesla GPUs that provide as much as 2.91 teraflops of performance.

IBM is charging $1,359 per month for the new servers, which include dual Intel Xeon E5-2620 chips, 64 GB of RAM, and 10 gigabit per second public and private uplinks, the spokesman told VentureBeat.

The new servers could make IBM SoftLayer look like a great place to run certain types of niche applications. A trendy type of artificial intelligence called deep learning is now commonly performed on GPUs, and high-performance computing (HPC) has long relied on GPUs, for example.

Other public clouds do offer rentable GPUs. Market leader Amazon Web Services introduced the G2 GPU instance, relying on Nvidia Grid K520 GPUs, in November 2013. Amazon also offers CG1 instances that run Nvidia Tesla M2050 GPUs. But these GPUs are in virtualized servers. So the new SoftLayer servers could give IBM a bit more strength in its quest to take on the most prominent public clouds available today, including Microsoft Azure and Google Compute Engine — neither of which currently offers GPUs.