This article provides a subjective history of POWER and open source from the viewpoint of an open source developer, outlines a few trends and conclusions, and previews what the future will bring. It is based on my talk at the annual OpenPOWER North America Summit, in which I aimed to show the importance of desktop/workstation-class hardware available to developers. In this article, I will cover a few additional topics, including cloud resources available to POWER developers, as well as a glimpse into the products and technologies under development.
The biggest problem for POWER that I can see at the moment is that the kind of POWER processors you want – little endian – are expensive. This precludes more affordable desktops from entering the market, let alone even laptops. Big endian POWER processors aren’t exactly future-proof, as Linux distributions are dropping support for them.
It’s a difficult situation, but I don’t think there’s much that can be done about it.
There is a solution, as there always has been.
Those who want big endian POWER systems will have to maintain the code. It will be a never ending series of tiny annoying mosquito bites, but they can do it, if there’s the will.
Wow, I thought I was the only one using OoenPower architecture. I have a dual CPU Power 9 system with; 128GB RAM, 1TB SSD/10TB SATA and a AMD Radeon VII Pro. I had purchased the Radeon VII Pro for my Threadripper system but after the Nvidia 3090 RTX came out, I put the Radeon VII Pro into my Power 9 system and it worked flawlessly, though it did come with a AMD WX-7100 so it wasn’t that much of a surprise. The OS is Cent OS.
The system has been a fantastic workstation, stable, fast and secure, which is why I moved to Power 9. I’m also a freak when it comes to unusual hardware, an obsession actually. Which started back in the 90’s when my uncle, who worked for SGI at the time, gave me a SGI Indy.to play with. Seeing how much I liked and used it he would periodically bring a new SGI machine for me to use for a few months and than replace it with a new.
My other computer was an Amiga that I shared with my brother so we started making movies with 3D effects using Aftereffects and LightWave on the SGI and than using Video Toaster to. bring it all together. In University I had an SGI Fuel and a Powerbook. After that I started using Sun Sparcstations, than Apple OSX and now Power 9 workstation with Linux and a Nvidia Xavier Dev board connected to 5 Nvidia Jetson X2 boards in a cluster.
I also plan on getting an Apple Macbook once they start using ARM64.
later I did manage to buy a used SGI Tezro, which is absolutely loaded to bare with quad 1Ghz R16000 CPU’s, V12 GPU, 8GB RAM, advanced audio and video capture cards. In fact, I’m using it right now.
@Zan Lynz
Isn’t there already a high end PowerPC segment doing this? I recall reading an article somewhere that talked about the reliability of systems in critical environments, BE versus LE, and why one if preferred for say Space based hardware over the other.
I gather there would also be reasons to argue for one or the other on the die level as well, but I’m not nearly well enough informed to enter that discussion. I expect very very few are qualified to discuss it!
I highly recommend the Raptor Blackbird system. It’s been serving me well for nearly a year. The best part is that I have access to the abundance of softwares, thanks to little endianess which makes porting app from x86 easy.
I have to agree that POWER9 CPU can be way over the budget for students and enthusiasts. I could only hope that the pricing would go down further once POWER10 is out.
I’m confused about wanting a little endian Power cpu… All Power cpus since Power3 can run little or big endian. I run little endian Linux on Power7, 8, and 9 cpus regularly at work.
The issue is the rest of the system, in particular firmware and similar aspects that users often have limited control over. POWER chips _can_ run BE or LE, but the systems they are put in are almost always designed under the assumption that they will only be run BE or LE, and support for running cross-endian relative to what the system was designed for is spotty at best. Because of this and switch-endian functionality being uncommon, many people tend to assume that the CPU is at fault (partially influenced by MIPS and ARM, both of which support either endianess at the ISA level but have the selection baked into the chip at the hardware level).
And, quite simply, POWER systems built to run LE tend to be far more expensive than those built to run BE.
Source: https://catfox.life/2018/11/03/clearing-confusion-regarding-modern-powerpc-endianness/
So with Linux at least you can run Linux in Little Endian even if the system is made for Big Endian
Hello,
On Arm processors, endianness isn’t hardcoded and can be switched at runtime (applies to both 32-bit and 64-bit).
On 32-bit Arm, it’s even at beyond that, with the SETEND instruction being available in user-space too, so that a thread can just set endianness for it without going through the kernel at all.
The reason for the distros favoring little endian makes sense but it’s unfortunate. It’s a lot easier for humans to examine raw memory dumps when the platform is big endian. Once again, spur of the moment decisions made in the 1970s are baked into what the world has to deal many tech generations later, and in this case it affects use of a newer chip architecture that was designed to accommodate either endian.
astro,
You are not wrong, but IMHO it’s one of the more nuanced holdovers from the past. Some legacy holdovers have been far more detrimental for progress than big/little endian architectures. Many protocols were not designed with security in mind and over the years their limitations have not aged well. Things like SMTP, DNS, IPv4 are notoriously problematic building blocks. Although we do our best to fix the limitations with a patchwork of extensions and hacks, our protocols have agglomerated into a huge mess because of it. They would benefit from a major overhaul. Other decisions like 1500 byte packets were made in an era of 300 baud modems made more sense then, but are highly inefficient today. Yet they remain the standard across the internet for legacy’s sake. Then you’ve got the “if it works, don’t fix it crowd”, meanwhile our infrastructure is hanging by a thread, haha.
In the early years engineers didn’t have as much experience and I doubt they would have expected that their work, which was designed to solve their specific problems at the time, would continue to be used so many decades later. If they they had known, maybe they could have engineered more forward compatibility into the mix.
A bunch of small backyard operators rarely have the pull of a few big commercial users, despite the backyard users often far far outnumbering the commercial users. The marketing/salespeople see the single big value order ahead of a billion nickel and dime orders.
I was exposed to this in a security application many years ago. A piece of high end($M) scanning equipment used in medical and scientific research was sought after by a Fed authority for a security application. They had one demand, the device had to have tenfold resolution improvement over the normal gadget. It was technically possible but never done because it made the device uneconomical for the wider commercial market. So the Feds wrote an order for a single machine to the value of about three full years worth of producing several dozen of the standard machines, and took delivery of exactly one and one only about twelve months later! btw., ten or fifteen years from now we will all benefit from this,
What happens with R&D around architectures like POWER is very similar, if someone big enough finds a serious use for it, enough that they can justify a whatever the cost situation it will happen.
I don’t think presence of the large POWER consumer is the problem – IBM is as large as they get. What we lack is affordable hardware and as result wide adoption by non-corporate consumers.
We also lack a broad dissatisfaction with x64. ARM has low power consumption, Intel/AMD/[insert minor players here] have the broadest compatibility on the desktop, and RISC-V has the most momentum behind a free ISA.
What is the compelling reason for the majority of people to switch to POWER? Why POWER specifically? Until there are good answers to that that pull people in there won’t be the economies of scale to get the price down.