News · May 19, 2020
Microsoft builds an Azure supercomputer to train OpenAI's models
At its online Build conference, Microsoft described a cloud supercomputer made for training very large AI models, and said parts of the same stack would reach Azure customers. The race to large models is now a cloud infrastructure story.
What happened
- Microsoft said it built, with OpenAI, a single system in Azure with more than 285,000 CPU cores, 10,000 GPUs and 400 Gbps of network connectivity per GPU server, which it ranked among the top five publicly disclosed supercomputers, according to its May 19, 2020 announcement.
- The company said its own Turing models improve language understanding in Bing, Office and Dynamics, and that it planned to open source Turing models and training recipes in Azure Machine Learning.
- Microsoft also announced a new DeepSpeed version for training much larger models and distributed training support in ONNX Runtime.
- The Register reported that the cluster is reserved for OpenAI and that Microsoft declined to share benchmark figures or chip details.
Why it matters
Few companies will train models at this scale, but many will consume them. The practical shift is that state-of-the-art language models are moving behind cloud platforms and APIs, so the important engineering choices are about integration, cost per call and data handling rather than training hardware.
For CTOs, open tools such as DeepSpeed and ONNX Runtime are the more immediate news: they can cut training and inference cost on models you already run. Benchmark them on your own workloads before committing to bigger GPU budgets.
Sources
- Microsoft Source: Microsoft announces new supercomputer, lays out vision for future AI work
- The Register: Microsoft claims it has spun up a top-five AI supercomputer for its pals at OpenAI – but won't reveal the full specs
This post is CMM’s own summary and commentary on publicly reported news. Facts are attributed to the sources linked above; please read them for full details. CMM is not affiliated with these publishers or with the companies mentioned, and trademarks belong to their owners. Prepared with AI assistance. Corrections or requests: [email protected]. See our editorial policy.
Want help with this?
Tell us what you need, and we will suggest the right team model on a short call.
Talk to us