Ampere teams with Scaleway for cloud-based Arm instances

Arm cloud servers are making moves in the fight for AI in the cloud.

First came Amazon Web Services with its homebrewed Graviton 4 chip, a custom Arm processor used in AWS AI services. The latest move is from Ampere Computing, the startup led by former Intel executive Renee James. Ampere has partnered with French cloud services provider Scaleway to offer cost efficient, cloud-based AI.

The two companies a few weeks ago announced availability of cost-optimized COP-Arm instances based on Ampere’s Altra family of data center processors. Scaleway is one of the largest cloud service providers in Europe and serves more than 160 countries in total.

The Ampere Altra processor is a power-efficient and high-performance alternative to traditional x86 processors designed for the cloud. Unlike x86 processors, the Altra does not use multi-threading. Instead, it opts for many cores in the package – up to 192 in the top-of-the-line product. Ampere believes that cores are more efficient than threads and better performing.

Scaleway instances are designed to support real-time AI, such as chatbots, real-time analytics, or video content analysis. The Altra is designed to excel in AI inference applications with much lower power draws than a GPU.

“Our ARM instances are meticulously crafted for energy optimization, allowing users to address computational tasks with a focus on eco-efficiency. Ideal for research simulations or financial modeling, for example, ARM instances minimize environmental impact without sacrificing the computational power essential for demanding High-Performance Computing workloads,” the company said in a statement.

Jeff Wittich, chief product officer at Ampere, said Altra has four times the performance per watt of Intel’s Ice Lake generation of Xeons, 10 times the performance per watt of AMD’s Milan, and five times the performance of Graviton 3 from AWS.

“So, for something like inference, that is a really, really massive advantage,” Wittich said, adding that cloud service providers pass on the savings in performance per dollar and per watt to the customers and even their greenhouse gas emissions.

“We do have some CSPs that actually pass along with real greenhouse gas emissions from the instances that people are running, which has started to matter more and more that they’re able to bring down their Scope III emissions by utilizing services that have a smaller carbon footprint,” he said.

Wittich says a big point of collaboration with Scaleway is ensuring that the inference performance is most efficient. “The software libraries that we’ve developed are easy for the customers to adopt, and that they get as many of their customers as possible running inferencing on Ampere processors because that inherently gives Scaleway the scale that they need,” he said.

Ampere instances are available from Scaleway now.