Former AMD Chip Architect Says It Was Wrong To Can Arm Project

Amid the renewed interest in Arm-based servers, it is easy to forget that one company with experience in building server platforms actually brought to market its own Arm-based processor before apparently losing interest: AMD.

Now it has emerged that Jim Keller, a key architect who worked on Arm development at AMD, reckons the chipmaker was wrong to halt the project after he left the company in 2016.

Keller was speaking at an event in April, and gave a talk on the "Future of Compute", but the remarks were unreported until picked up by WCCF TECH.

In the talk, available online via YouTube, Keller discusses how when planning the Zen 3 core – now at the heart of AMD's "Milan" Epyc processor chips – he and other engineers realized that much of the architecture was very similar for Arm and X86 "because all modern computers are actually RISC machines inside," and hence according to Keller, "the only blocks you have to change are the [instruction] decoders, so we were looking to build a computer that could do either, although they stupidly cancelled that project."

That project was apparently the K12, which was planned to be AMD's first custom microarchitecture based on the 64-bit ARMv8-A instruction set, and would have led to chips that would follow on after the Opteron A1100 series chips, which were based on Arm's Cortex-A57 core designs.

Keller's involvement at AMD goes back a long way, as he worked on the Athlon (K7) processor and the K8 microarchitecture, which introduced a 64-bit instruction set for X86 processors and was thus the foundation of the first generation of Opteron chips, which saw AMD claim a sizable fraction of the X86 server market in the first decade of this century.

After working at SiByte (later acquired by Broadcom), PA Semiconductor, and then Apple, Keller rejoined AMD in 2012 to lead development of the first Zen microarchitecture and on Arm-based chips. He left again, to join Tesla in 2016, before the first Zen-based Epyc chips were launched.

Keller went to Intel in 2018 and ejected in 2020. He then rocked up at Tenstorrent.

• Nvidia wants to lure you to the Arm side with fresh server bait
• AMD beds Arm in 'Project SkyBridge' and home-grown 'K12' ARM architecture chips
• AMD bests Intel in cloud CPU performance study
• US to help Japan make leading-edge 2nm chips, possibly by 2025
• Who's growing faster than Nvidia and AMD? Rising datacenter star Marvell

The Opteron A1100, codenamed Seattle, was launched in early 2016 with up to eight cores and clock speeds of 1.7–2.0GHz. AMD had originally planned to follow it up with a sixteen core version, but by the end of 2016 the company had apparently changed its mind and decided to focus all its efforts on the X86 architecture and the Zen core designs.

Since then, things have changed, with some of the cloud providers introducing Arm-based servers, such as AWS operating EC2 instances based on its own Graviton chips and Microsoft offering Azure virtual machines powered by Ampere's Altra Arm-based processors.

Who knows, perhaps AMD might be tempted to revive its K12 project and bring it up to date if interest in Arm servers continues to blossom? ®