Linux Integrating AMD Platform QoS
Back in November, AMD debuted their new Zen 2 architecture. The new chips will offer a massive boost in performance, building on Zen and Zen+. To achieve the performance numbers, AMD has made a number of major changes to the underlying designs. Ahead of launch next year, OS support to starting to roll out. This week, we have the new Linux 4.21 kernel set to add some support.
The new kernel integrates new AMD Platform QoS support that is geared towards the 7nm EPYC and Zen 2 chips. Due in part to the shift to the chiplet design, resource allocation is quite different. The updates aim to provide monitoring of the resources as well as place limits on them. Initially, this will be for L3 cache monitoring, limiting, prioritisation and memory bandwidth. Due to the chiplet design, the L3 cache is shared on a core die, with challenges to latency and sharing.
New Chiplet Design Changes Resource Balancing
With up to 64 cores and 128 threads, Rome is a massive leap forward. Due to the massive increase in resources, the company is developing entirely new methods of core layout. It does look like AMD has been working towards this stage for quite a while. It combines elements we have seen before with Infinity Fabric, interposers, HBM and iGPU integration. The end result should be a highly scalable design.
As an open source OS, Linux is a great source of revelations for the chips. AMD has already introduced earlier patches that exposed a new compiler for Zen 2, with new instructions. Once we get closer to launch, we can expect even more matches to roll out in quick succession. AMD is also likely working with Microsoft to build support into Windows 10. Silicon is also likely in the hands of select customers so we might get leaks soon.
