I wonder why these haven’t gotten much attention yet? This is a better IMHO solution to the “more core wars” that the ARM side is about to get hip deep in.
Alas, great as they are in theory, they are once again let down by Windows and only perform to their true potential under custom Linux builds.
AMD starts bringing its own tiny CPU cores to new Ryzen 7040 laptop chips | Ars Technica
You just reminded me of these screen caps from a Broken Silicon interview where they discuss a lot of interesting tidbits about AMD’s upcoming Phoenix 2 architecture:
The first interesting thing is how smaller Zen 4c “Dionysus” cores are almost exactly scaled down versions of the full “Persephone” cores, down to the silicon layout.
They crammed in all the functions into a smaller die space, which is testament to AMD scalable architectural design (something both Intel, Apple, and Qualcomm have struggled with lately).
You can see how the full heterogenous SOC has room to be extended on the left side and perhaps even “mirrored” vertically in a chiplet config using their Infinity Fabric interconnect.
Even looking at their micro-op architecture, I see a lot of emphasis on wide registers, deep re-order buffers (note the multiple queues and schedulers), plus lots of distributed cache. These things are looking to be little IPC demons.
Are productivity apps really multi-core?
Well no, but even though most apps remain single-threaded primarily, the OS farms out a lot of system API calls to other threads. A simple example, is saving/syncing your word doc runs on a seperate thread than the main window.
The OS often has hundreds, even thousands, of background worker threads than happen on the other cores. When intelligently scheduled, this is the difference between an ultra-responsive OS, or a stuttery mess.
