I’m not a POWER (or recently: Power) expert, only an enthusiastic user and advocate. Still, in the past couple of weeks a number of people from around the world asked my opinion how the POWER architecture could be kept relevant. This blog is really just an opinion, as I do not have the financial means to go ahead. It is full of compromises some people are not willing to make. However, I think this is the safest and fastest way forward.
Why? Is there a problem?
Power 10 was just released, and used in some of the most powerful servers ever. Power became an officially supported architecture in major Linux distributions. Why do I talk about becoming irrelevant? Is there really a problem?
Well, it all depends on the perspective. IBM treats Power as an enterprise platform, just like mainframes. And as long as they run AIX and IBMi with a couple of proprietary commercial applications, they are right. However, as far as I know, a good part of Power boxes run Linux. And Linux is a volume play. The more users and developers work on a platform the better chance it has for survival. This is how 32 bit Power support was dropped many years ago from most distributions, even if some people still have Apple Macs and Genesi Pegasos boxes running. And this is how 64bit big-endian support was removed from mainstream distributions as well.
Power 9 had a huge momentum, in most parts software support for Power 9 is now in par with x86 and ARM. Unfortunately it is not enough to reach a momentum, it needs to be maintained as well. Raptor Computing did a fantastic job making Power more affordable. Those machines reached key developers in major projects. However their prices are going up due to supply chain issues and they do not plan on Power 10 any time soon (as it would require to use some closed source software components in the firmware).
The OpenPower foundation is planing to solve the volume play in its Power Pi project, but it is still years away. Currently there is no CPU that could be used on the planned $250 board and normally it takes 1.5 years or more to go from planning to a mass produced CPU.
You might say, that there are free resources available for open source developers. There is GitHub CI support for Power and various universities provide remote access to interested open source developers to Power servers. However most developers consider having a system on their desk locally as the best way to develop software. ARM and even RiscV have a huge advantage with the average developer now.
The Power architecture is handled as first class citizen in most major Linux distributions and even in FreeBSD, but by the time we have affordable Power hardware to grow the number of Power users and developers, many of them might already drop this level of support for Power.
Affordable hardware quickly
The previous section of my blog can be easily summarized in a single sentence in a TL;DR; style: we need affordable Power hardware quickly to keep and expand the momentum. Obviously it needs compromises as well.
Keep the dream alive
Old Macs are big-endian, just as network processors from NXP. Some Power developers still want big-endian systems to keep the dream alive. But support for big-endian systems is mostly gone from Linux distributions, and when it comes to developing common utilities or even programming languages, most developers are no more even aware that a world exists outside of little-endian. As much as I love the PowerPC laptop project, I see it now as a dead end: producing hardware for an ever shrinking software ecosystem.
As much as I’d love to see a Power 10 desktop, I do not expect it to be affordable any time soon. Right now only 15 and 30 core variants are available for high end servers. Even if Power 9 is not so power-efficient and can be outperformed in some cases by some of the latest x86 CPUs, it is already available and at a relatively good price.
Not fully open source desktop board
Raptor Computing did a fantastic job at creating fully owner controlled boards where even the smallest bit of software controlling the board is open source. However even their smaller board is a full server board. Removing server components, like remote management capabilities, could bring costs down, just like components with closed firmware. My experience with firmware is that open source does not mean necessarily better, rather the opposite (yes, I am aware that this statement contradicts my title: open source evangelist).
I would not want to compete with IBM or Raptor Computing with server boards. Both have done their optimizations in enterprise manageability or having a fully open source stack down to the lowest level. On the other hand, while using a server board in the desktop technically works, simplifying it down to the desktop both on the hardware and software side can help to make it more affordable and thus reach more users. Hopefully a lot more users.
Obviously, creating a more affordable Power 9 board quickly is just a first step. It helps to reach more users and developers than the current IBM and Raptor Computing offerings. It also helps to make sure that efforts of the OpenPower Foundation are not wasted and Power support stays as first class citizen in major Linux distributions.
I do not know the Power 10 CPU roadmap and if there will be any smaller versions of Power 10, but I really hope so. Those could be used in desktop systems once available.
Of course the ultimate target is a board that anyone can afford without thinking twice. Just like a Raspberry Pi. The Power Pi is planing to fulfill this idea. It might be here sooner or later than lower end Power 10 systems.
The Libre-soc project is also building a Power CPU with many ground breaking ideas. Unfortunately a generally available version is expected to arrive even later than the Power 10 based desktop or Power Pi.
Power itself is probably not in direct danger, but Linux and open source are definitely becoming an endangered species on Power due to the lack of a large active community. This situation could be improved with more affordable Power hardware. In short term a Power 9 board could be used for this purpose, on a longer term there are many open possibilities ranging from SoC to Power 10 (or later).
Obviously, my post only covers one aspect of a problem: keeping the open source community around POWER healthy. I have no idea about the engineering or financial side. I wonder about your opinion and if anyone will step up and implement something along these lines.