Until now, Amazon Web Services (AWS) was the only hyperscaler investing heavily in ARM workloads. At Google and Microsoft, they leaned on generic Ampere Altra chips. Until now, as Microsoft adds to Azure the Maia and Cobalt chips, for AI and generic CPU workloads. To top it off, Microsoft is also adding new Nvidia and AMD hardware. The Azure infrastructure is getting a major upgrade.
With Microsoft Azure Maia and Microsoft Azure Cobalt, the Redmond-based company will compete directly with the AWS Inferentia and the AWS Graviton 3 chip.
The Azure Maia chip will be used for inferencing. With inferencing, it can provide support for AI workloads such as OpenAI, Bing chat and CoPilot. Whether this chip will be available in instances for customers is not yet entirely clear. It is at AWS, so we suspect Microsoft will also start offering it.
The Azure Cobalt is the direct counterpart to the AWS Graviton 3 chip and all Intel Xeon chips. Namely, it is optimized for generic workloads (CPU), but at a much lower power consumption and cost. Something that is also the case at AWS with their ARM-based chips. With this chip, we can safely assume it will soon become available to customers in many regions, probably leading to the necessary optimizations in the Azure Cloud. It will allow organizations to bring down their Azure costs.
Of course, we still have to wait on benchmarks of the Cobalt’s performance, as well as its power consumption and price tag. Until then, the question remains if Microsoft’s first proprietary chip can compete with Intel Xeon chips and the Graviton 3.
What we do know for sure is that Intel will once again have competition in hyperscaler data centers. In addition to AWS, Microsoft is now developing its own chips. That may result in fewer sales from Intel to Microsoft.
Of course, the Azure Maia also has yet to prove itself in terms of performance and cost. But should that chip not be liked for whatever reason, Azure has some additional choices to offer. It is also adding the AMD MI300X and Nvidia H100 Tensor Core GPUs to its portfolio for AI workloads. New VMs will be available with these chips on board.
Azure Boost available
Finally, Microsoft is now presenting the availability Azure Boost. This is a system that makes storage (storage) and networking much faster. Azure Boost does this by taking these (virtualized) tasks away from the host machines and hypervisor, but performing them on purpose-built hardware. It seems Microsoft has added a DPU to certain instances to make this possible. The company claims that this allows network speeds of 200 Gbps and for remote storage 10GBps with some 400K IOPS, giving organizations access to extremely fast storage.
All in all, Microsoft is giving its Azure infrastructure hefty hardware upgrades to make a bigger stance against competition from AWS and Google Cloud.
Also read: H200: Nvidia’s latest AI chip with 141GB of super-fast memory
1 day ago
