Ampere to buy OnSpecta for AI inference workloads acceleration

Get a free Techzine subscription!

Chipmaker Ampere announced on Wednesday that it plans to acquire OnSpecta, a startup that makes software used to accelerate AI inference workloads at the edge and in the cloud. The terms of the deal have not been made public.

OnSpecta was founded in 2017 and is headquartered in Redwood City, California. It has collaborated with Ampere before.

The startup’s Deep Learning Software (DLS) has proven to accelerate Ampere-based instances running popular AI-inference workloads by up to 4 times. Ampere started shipping its Altra processor, an Arm-based server chip for cloud computing and hyperscale data centres in 2020.

Ampere’s ambitions

In early 2021, Ampere launched its Altra Max 128-core server processor. At the same time, the company outlined its roadmap for the next generation of chips, saying it would release the 5nm processor Ampere chips next year and follow them up with another generation in 2023.

The chipmaker is determined to take on AMD and Intel in the chips market and makes the case for cloud deployments that follow this format:

  • Consistent scaling as more cores are added to the chips
  • Reduce power consumption
  • Increase performance for workloads

Adding deep learning expertise to these releases will enable the company to deliver on many of these promises.

Doubling down

The acquisition will give Ampere an optimized model zoo with object detection, recommendation engines, and video processing capabilities.

Ampere’s processors are gaining traction and seeing adoption by server makers that supply hyperscale cloud providers. Some of the server makers using Ampere processors include Gigabyte. Foxconn, and Inspur.

Ampere Altra has also seen adoption by big-name players like Equinix, Oracle Cloud, Microsoft Azure, Tencent Cloud, Cloudflare, and Bytedance. With the AI capabilities added, Ampere’s cloud deployment ambitions could very well work out.