> For Intel and AMD, the demarcation point between them and their customer is the CPU, so if they want to keep secrets and maintain a stable interface between them and their customers, they're effectively pushed to do it at the CPU boundary. With IBM, this historically hasn't been the case; traditionally, IBM has only sold servers containing their own CPUs, not the CPUs alone. This means that traditionally, the only customer for IBM CPUs has been IBM — which means there's far less motivation for IBM to lock things down. Moreover, IBM Power servers have traditionally shipped with a proprietary hypervisor (known as PowerVM) built into the firmware. On these servers, all OSes run under this hypervisor, and you can't run an OS directly on the bare metal. This means that the natural interface between IBM and its customers has naturally fallen at the hypervisor—OS interface, not at the CPU..
> When IBM suddenly decided to open up their POWER CPUs for use by third parties, they made available a platform which for its entire lifespan up until that point, hadn't evolved under the same pressures as the existing x86 CPU market, but instead in an environment in which there was really no natural motive for them to lock things down at the CPU level. Which is almost certainly why IBM POWER CPUs seem to so greatly lack any “curtains” — the curtain had always been shipped with the hypervisor, not the CPU.
HN discussion (2023): https://news.ycombinator.com/item?id=36127543
Why did POWER10 become less open, requiring binary blobs?