Fresh evidence Amazon is ARMing its huge cloud against Intel et al
Attention, CPU gurus: Web bazaar wants to pour your brains into its servers
In early April, Amazon Web Services' chief technology officer Werner Vogels told The Reg that “there is absolutely room for ARM in the data centre” because “power management for ARM is considered state of the art” and AWS is “always looking for efficiency”.
It now transpires that AWS is also looking for a “CPU architect / micro-architect.” The job posting for that gig has vanished from the web, but this one for a ”CPU and System Architect” to join a “Silicon Optimization Engineering Team” remains.
The role needs someone with “deep knowledge in the following areas:”
- Server platform processor architecture
- Micro-architecture performance and workload characterization
- Memory subsystem performance optimization
- Server platform power optimization
- Virtualization technology
The Silicon Optimization Engineering team has already hired Director David Borland, whose work history includes stints at Intel, Marvell and Calxeda, suggesting he knows his way around just about every nook and cranny of a server.
Why would Amazon need those skills in-house? The company is extremely secretive about its operations, but the "CPU architect" job description – as noticed by Gigaom – suggests it is developing new server hardware at the very least. If that's the case, it is bad news for the likes of Quanta, as AWS would be treating them as dumb manufacturers rather than relying on them for innovation.
If the team and the hires are a signal that AWS wants to design custom silicon to power those servers, it's terrible news for Intel because any challenge to its dominance of high-end silicon will hit it in the place it hurts most: high-margin products. There is upside in the fact that, unlike Facebook, AWS doesn't like to open source its innovations. So even if AWS adopted ARM, it would likely not start to create the ecosystem of drivers and other code that VMware CEO Pat Gelsinger, among others, feels ARM needs before it can challenge x86 in the data centre. Intel will therefore be hoping AWS doesn't find open source religion at the bottom of a motherboard.
Of course Gelsinger and Vogels are talking about very different data centres. But both are keen on hypervisors, Gelsinger because it is his core product. During his chat with The Reg earlier this month, Vogels opined that once a workload sits inside a hypervisor, the underlying silicon matters little.
But he also said that some AWS customers have figured out the silicon underpinning some AWS instances and like the fact they're working on a particular CPU. How AWS would explain a move to ARM to those customers is anyone's guess. ®