There is an arms race in the cloud. In years past, the hot topics were new CPUs, more memory, and faster SSD storage. Today, there is a new battlegroud: GPUs. The reason for this is that cloud providers can provide on-demand GPU compute clusters to customers who are occasional users of AI. While most major AI shops will build their own infrastructure that has payback periods in a quarter or two, the cloud is attractive for some models. We have seen Amazon AWS, Microsoft Azure, and Google’s GCP all announce support for the newest NVIDIA Tesla V100 “Volta” generation GPUs. Now IBM is getting into the mix.
NVIDIA Volta Enters the IBM Cloud
IBM is offering the ability to attach up to two NVIDIA Tesla V100 PCIe GPUs to cloud instances. The company is not highlighting NV-LINK in its cloud offering. Perhaps one of the standout features of the IBM Cloud and NVIDIA announcement is that it is attempting to bridge a hybrid-cloud model. IBM also offers IBM POWER9 server infrastructure for those who want something different for their on-prem solutions. Here is an excerpt from the release:
At IBM, we’re focused on delivering new AI capabilities both in the cloud and on premises to help enterprises not only gain critical insights from their data, but also create new value with that data. We’ve been working closely with NVIDIA to bring their latest GPU (graphics processing unit) technology, NVIDIA Tesla V100, to the cloud and were the first to offer a comprehensive suite of GPUs including the P100, K80 and M60 on IBM Cloud bare metal and virtual servers. To power on-premises workloads, IBM also offers the industry’s only CPU-to-GPU NVIDIA NVLink connection on our latest POWER9 servers.
Building on this momentum, we’re excited to share that IBM is introducing the NVIDIA Tesla V100 GPU to support mission-critical AI, deep learning and HPC workloads on the cloud.
Starting today, you can equip individual IBM Cloud bare metal servers with up to two NVIDIA Tesla V100 PCIe GPU accelerators — NVIDIA’s latest, fastest and most advanced GPU architecture. The combination of IBM high-speed network connectivity and bare metal servers with the Tesla V100 GPUs provides the performance and speed that enterprises need. That means AI models that once needed weeks of computing resources can now be trained in just a few hours.
Performance and rapid provisioning are critical for AI and HPC workloads. Building on IBM bare metal support for the NVIDIA Tesla P100 GPU, IBM will also make the P100 GPU available on IBM Cloud virtual servers. This provides power and high performance for AI and deep learning workloads with the scalability and flexibility of IBM’s virtual servers. With the Tesla P100 GPU accelerator, you can leverage up to 65 percent more deep learning capabilities and 50 times the performance than its predecessor.
(Source NVIDIA GPUs on the IBM Cloud)