IBM will soon offer Nvidia Tesla P100 GPUs on its Bluemix bare metal cloud servers, becoming what it says is the first major cloud company to do so (Tencent is also set to offer it soon).

Competitors such as AWS, Azure and Google Cloud currently offer the older K80 GPUs.

IBM pilgrimage
– Sebastian Moss

Rise of the GPU

Customers will be able to equip individual IBM Bluemix bare metal cloud servers with two P100 accelerator cards. This provides 4.7 teraflops of double-precision performance and 16 gigabytes of GPU memory in a single server. 

IBM will be using PCIe form factor cards within an Intel x86 server, rather than the slightly more powerful NVLink-optimized version used by Nimbix for its cloud service.

“The latest Nvidia GPU technology delivered on the IBM Cloud is opening the door for enterprises of all sizes to use cognitive and AI to address complex big data challenges,” John Considine, GM of cloud infrastructure at IBM, said.

“IBM’s global network of cloud data centers along with its advanced cognitive and GPU capabilities is helping to accelerate the pace of client innovation.”

Ian Buck, GM of accelerated computing at Nvidia, added: “As the AI era takes hold, demand continues to surge for our GPU-accelerated computing platform in the cloud.

“These new IBM Cloud offerings will provide users with near-instant access to the most powerful GPU technologies to date – enabling them to create applications to address complex problems that were once unsolvable.”

It is likely we will soon be hearing more about the P100 finding its way into the data center, with Nvidia and Microsoft revealing the HGX-1 at the OCP Summit earlier this year.

The open source hardware design consists of eight Tesla P100 GPUs connected via NVLink. When used with Microsoft’s Project Olympus design, allows for four HGX-1s to be connected at the same time, giving a total of 32 GPUs.