An AI battle royale is shaping up between cloud giants like Google, Microsoft, and AWS–and do not forget about IBM, whose increasing bets on AI involve resurrecting a familiar name with a new aim.
Announcements this week coming out of separate conferences hosted by Google and IBM show just how much these giants are ramping up their system and product arsenals for that battle.
At Google’s I/O conference, the company announced that its new A3 GPU supercomputer, built on Nvidia H100 Tensor Core GPUs, is launching in a “private preview” version to support GPU-powered training and inference of AI and ML models. A3 virtual machines (VMs) are powered by eight H100 GPUs.
“Google Compute Engine A3 supercomputers are purpose-built to train and serve the most demanding AI models that power today’s generative AI and large language model innovation,” according to a blog post from Roy Kim, Director, Product Management, Google Cloud, and Chris Kleban, Group Product Manager, Google Cloud.
With 26 exaflops of AI performance, the new system out-performs Google A2 supercomputer built using Nvidia A100 GPUs, the company said, adding that the A3 also is the first GPU instance to use custom-designed 200 Gbps infrastructure processing units co-developed with Intel. This allows “GPU-to-GPU data transfers bypassing the CPU host and flowing over separate interfaces from other VM networks and data traffic,” the blog posts stated. “This enables up to 10x more network bandwidth compared to our A2 VMs, with low tail latencies and high bandwidth stability.”
Google also touted intelligent Jupiter data center networking fabric that it said can scale up to “tens of thousands of highly interconnected GPUs” for adjustable “topology on demand.”
As companies eventually migrate from AI training to AI inference as their strategies progress, Google also said the A3 virtual machines support up to “a 30x inference performance boost when compared to our A2 VM’s that are powered by NVIDIA A100 Tensor Core GPU.”
Google’s announcement follows Microsoft’s move late last year to build an AI supercomputer also leveraging the H100 from Nvidia, so as Google and Microsoft exchange competitive blows in the AI battle, Nvidia stands to win either way.
While Google I/O was going on this week, IBM was making an AI splash of its own at its IBM Think conference, where the company made several AI announcements, including the introduction of WatsonX, which the company described as a platform supporting “foundation models and generative AI, offering a studio, data store, and governance toolkit” to help enterprises as they work to leverage AI capabilities.
If the name “Watson” sounds familiar, there is good reason for that, as WatsonX evokes the name of IBM’s Watson, which might be thought of as the AI O.G. in comparison to the young buck known as ChatGPT. IBM was never able to leverage Watson’s famous “Jeopardy!” skills into much of a revenue generator, but years later there is hope it can be a winning brand for a platform to enable enterprise AI applications.
"With the development of foundation models, AI for business is more powerful than ever," said Arvind Krishna, IBM Chairman and CEO. "Foundation models make deploying AI significantly more scalable, affordable, and efficient. We built IBM watsonx for the needs of enterprises, so that clients can be more than just users, they can become AI advantaged. With IBM watsonx, clients can quickly train and deploy custom AI capabilities across their entire business, all while retaining full control of their data."
WatsonX is expected to be available in July. Meanwhile, IBM also announced other AI plans at Think as it looks to up its game amid the growing AI excitement. These plans include a GPU-as-a-service infrastructure offering designed to support AI-intensive workloads, an AI-powered dashboard to measure, track, manage, and help report on cloud carbon emissions, and a new practice for watsonx and generative AI from IBM Consulting that will support client deployment of AI.
That last offering is a reminder that this is a market that is still sorting itself out. Enterprises that were not even thinking about AI five years ago are now working to knit it into their corporate strategies as they quickly attempt to become conversant in topics like “generative AI” and “large language models” and the whole new lingua franca around AI that seemingly has emerged overnight. Ultimately, this excitement translates to huge demand for systems to train AI and large language models and eventually perform AI inference, as well as other platforms and tools for helping companies figure out exactly how and where they should be adopting AI capabilities.