The French cloud computing infrastructure provider Scaleway SAS today announced the availability of what it calls Cost-Optimized instances based on the Arm central processing unit architecture, targeted at real-time artificial intelligence and high-performance computing workloads.

The cost-optimized COP-ARM instances are powered by Ampere Computing LLC’s Ampere Altra CPUs, which are said to be a more cost-effective and high-performing alternative to Intel Corp’s x86 processors.

Scaleway is a cloud infrastructure provider that targets developers and scaling businesses with more affordable services than those offered by giants such as Amazon Web Services Inc., Microsoft Corp. and Google Cloud. Among its services, it offers web hosting, cloud hosting, dedicated servers, private infrastructure and bare-metal servers. It serves businesses in more than 150 countries and has four data centers in France and one in the Netherlands.

The company said Ampere’s Altra processors can be a more affordable option for companies wanting to run AI training and large language model inference applications. Ampere is an agnostic, Arm-based processor developer that partners with cloud infrastructure providers, helping them to create their own data center processors. For instance, it previously collaborated with AWS, helping it to design its popular Graviton virtual machines.

It also provides chips to Oracle Corp., one of its main investors. The company is notable for succeeding with its Arm-based chips while other chipmakers, such as Qualcomm Inc., Marvell Technologies Inc., Advanced Micro Devices Inc. and Samsung Electronics Co. Ltd. have all exited the market for Arm-based data center chips.

Scaleway said the cost-optimized Ampere Altra instances are tailor-made for real-time AI workloads such as chatbots, data analytics and video content analysis. Their lightning-fast inference capabilities can, for example, ensure swift and efficient responses to questions posed by users to generative AI chatbots. It claims that the instances can be optimized for such workloads at a fraction of the cost of alternative solutions, such as graphics processing unit-based instances.

Ampere Chief Product Officer Jeff Wittich said AI training and inference are two different workloads, and while GPUs and accelerators make sense for the former, CPUs are perfectly suited to the latter. “Our CPUs are especially well-suited to inference because they are high-performance and balanced,” he said. “Plus you need predictable latency in these cases, and to keep processing close to the core, not have it bouncing around all over the place.”

Because AI is still in the hype and research phase thanks to the excitement around chatbots such as ChatGPT, many enterprises have instinctively sought to throw as much power as they possibly can at large language models. But Wittich said that as AI matures, common sense will prevail and companies will increasingly look for more cost-effective and sustainable ways to run it.

“Not everyone will be able to pay for a solution like ChatGPT, which features all of human knowledge,” he said. “We’ll see more specialization of models as well as refinement of existing models.”

These more specialized and refined models won’t need such enormous power, Wittich said, which is why Ampere’s Arm CPUs are such a credible option. “Overall, AI models will become smaller and more focused on specific tasks,” he said.

This is where the energy efficiency of Scaleway’s instances stands out, allowing customers to power AI workloads in a more sustainable way. The chips are also suitable for high-performance computing workloads such as research simulations, financial modeling and extensive data analysis, as well as running databases, cloud gaming and media encoding, minimizing the environmental impact without sacrificing on performance.

Scaleway said the new cost-optimized Ampere Altra instances are available to customers now, and represent a significant leap in cloud technology. “We are offering businesses a powerful and cost-effective alternative that can achieve high-performance results in the most efficient and sustainable way possible,” he said.

Image: Scaleway/Ampere