Arm sees support path to heterogeneous compute
Toward making software development 'frictionless' over multiple processors, accelerators
Arm says heterogeneous compute architectures – those with a mix of CPUs, GPUs, DPUs, and other processor types – pose a challenge for software developers, and greater multi-architecture support is needed to address this.
Specialized processing, as the chip designer refers to it, will likely succeed Moore's Law for driving innovation. System builders will focus on attributes such as performance, efficiency, and optimization for the task at hand rather than clock speed when combining CPUs, GPUs, DPUs, and other devices, Arm said.
But this specialized processing model upends business as usual for software developers, according to Bhumik Patel, Arm's Director for Software Ecosystem Development. He says in a blog post that the answer is to enable a frictionless experience for developers to achieve multi-architecture support for the software they code.
Arm has a vested interest in this, of course, as many of the DPUs and other accelerators that appear in heterogeneous compute architectures tend to be based around Arm processor cores.
A good example is Nvidia's Bluefield DPU, aimed at applications such as SmartNICs that can offload network processing tasks from the host CPU.
Patel said that frictionless development requires the availability of developer tools across the software stack, and for cloud and edge deployments, this must include the ability to develop applications with cloud-native practices. "Over the past few years, in collaboration with our partner ecosystem, we have enabled the majority of projects across the Cloud Native Computing Foundation landscape, and we continue to drive further adoption of multi-architecture support," he claimed.
Arm said it has also invested in efforts such as Project Cassini and Project Centauri to simplify the process of bringing cloud-native software experiences to edge deployments such as 5G base stations and IoT gateways, kit that is often based on Arm technologies.
- Goldman Sachs reportedly set to head up $60bn Arm IPO
- Nvidia reveals 144-core Arm-based Grace 'CPU Superchip'
- UK's largest union to Arm: Freeze job cuts now
- SiFive bags $175m to further challenge Arm with RISC-V
Arm said both initiatives have three elements: a product certification process outlining hardware and firmware specifications; a security certification program; and reference implementation guides for software developers.
Hardware and security certifications ensure developers know what to expect from devices that meet Cassini or Centauri standards, while the reference designs remove any time, cost, and effort required to develop on Arm.
VMware is one partner that has worked closely with Arm via Project Cassini to get ESXi-Arm up and running. This is the version of VMware's ESXi hypervisor designed to run on 64-bit Arm silicon, and forms a key part of its Project Monterey effort to enable Arm-based SmartNICs to handle network processing, zero-trust security, and storage acceleration features in VMware-based infrastructure.
Arm may not be the only firm with an interest in cross-platform development, but it is perhaps in a unique position thanks to the diverse range of areas that the Arm architecture is deployed in. And the firm may have yet another architecture to support when its upcoming GPU is introduced.
Patel said Arm is dedicated to accelerating deployments by enabling frictionless development. As long as such initiatives don't get axed as a part of Arm's efforts to make itself leaner and more appealing to investors for its upcoming IPO, of course. ®