[Raion]

The most common myths I've seen thrown around are in regards to its performance.

Let's get one thing straight:

Itanium was neither dogshit slow, nor supercomputer-level fast.

Itanium held its own on technical computing loads, but was often very, very temperamental, especially for branch-heavy code streams with a lot of dynamic memory access. Post-Montecito, no Itanium part was performance-competitive with contemporary Power (Power7 for Tukwila, Power7+ for Poulson) or even really contemporary Xeon (though the Itanium parts had more exposed QPI.)

Itanium was designed under two premises which were true during its development in the 1990s:
Out-of-order cores will never be as wide as in-order cores (Itanium was strict in-order)
The ability to schedule accurately for in-order parts will improve, and hardware hooks like Advanced Loads will mitigate what remains.
However, in 2021 Apple is pushing a very wide, out-of-order core (the M1). The premise has changed, and rapidly obsoleted any general-purpose, high performance in-order core.

The second major myth was surrounding Itanium's market penetration/performance.

For what it's worth, it sold nearly 4 billion, was at one point number 2 for RISCy UNIX servers, and was definitely profitable for HPE, with a product line lasting for over a decade. On the other hand, outside of HPE, it never achieved widespread success due to missteps in the launch years, and by the time of Kittson (really, just a smol clock bump to Poulson) it really had slowed to a trickle.

The compiler isn't necessarily the full story either. While HPE's aCC and Intel's ICC did much better than GCC ever did with IA-64 binaries, the elephant in the room is that the ISA had poor code density, undersized L1I cache, and expensive memory access compared to other architectures.

The last myth is that Itanium killed off superior RISCs... this is basically bullshit. POWER survived through raw performance and daddy IBM's massive R&D into it. SPARC had a massive install base. PA-RISC was dead-end and intended from the beginning to be replaced. For those who have rose-tinted glasses, PA-RISC had many squirrely/weird design decisions, especially regarding addressing modes. MIPS was in a rough place with the cancellation of the R18000 design, and while SGI planned to port IRIX to Itanium, that got ditched because they squandered their money on projects that basically gave an impression to investors that SGI lacked any form of real direction. Alpha was killed off, but that's not due to Itanium. When Compaq bought DEC, they sold the IP off and basically put it on life support, halting development because their bread and butter was Intel. Alpha was legitimately competitive and performant at the time, for sure, but it has aged like milk, much like other dead RISCs.


Also, fun fact. While Intel became the face of Itanium, it mostly is an HPE design, a result of a program to replace HP's PA-RISC with a superior design. By the time Intel joined the alliance in 1994, most of the design work was basically baked. It killed IAX, Intel's RISC project, but that's about it. The Fort Collins Design Center had its origins in HPE.

[lunatic]

When it comes to performance, you should probably consider two perspectives: how was how compared to x86 and how was ist compared to RISC chips? BTW: Itanium was not RISC. Itanium came out as a world domination plan and was supposed to beat all RISCs and also higher end PCs. This plan just failed.

From the perspective of old RISCs, Itanium had no existing application base. Vendors were largely pushing Linux. Existing traditional systems offered compatibility layers, which worked fine (like on Windows or HP-UX), but they were just not as fast as the old platform. To really get any benefit, you needed (obviously) native IA64 applications. Additionally, many old RISC users hated Itanium with a passion, because they thought Itanium killed their old favorite chip/OS. That's not great marketing when you have to fight your most loyal customers.

Form the perspective of x86, Itanium was fast, but expensive. Windows for IA64 never got off the ground, because nobody really likes Windows. People want a certain set of applications and they all run on Windows. With Itanium, all this was lost. Sure, you got the convenience of having the Windows you know, but what can you do with it that doesn't involve running x86 software in compatibility mode? It's like back in the 90s when Microsoft sold Windows for Alpha, MIPS, and PowerPC. Here in Europe, there were a number of MIPS machines from Siemens, which ran NT. Their proprietary Unix was big endian while NT was little endian and needed a different firmware. So users were stuck with NT and couldn't run anything else. Years later you could see collectors showing off their Windows/MIPS and the only apps they used was Solitaire and the pinball game that came with NT4. Similarly, some government agency used Windows on PowerPC. When these machines reached the collectors market years later, luckily there was a working Linux port and you could also install a special version of AIX. Windows on Itanium was an attempt to make something work that failed three times already.
Vendors were pushing Linux on IA64, but the Linux people are notoriously not loyal to a CPU or a company. There was a sweet spot in 2003/2004 when Itanium was rather competitive, but soon after, people switched back to x86 and x86_64. There is substantial reason to assume that in effect AMD killed Itanium with the introduction of x86_64. After that, Itanium was put into the same corner as all the other proprietary RISC/Unix platforms and died like most of them. With the advent of x86_64, the world domination plan failed. Intel had no reason to spend all this money on Itanium just to battle itself.

[Raion]

Hey Lunatic!

> When it comes to performance, you should probably consider two perspectives: how was how compared to x86 and how was ist compared to RISC chips? BTW: Itanium was not RISC. Itanium came out as a world domination plan and was supposed to beat all RISCs and also higher end PCs. This plan just failed.

It simply wasn't in the same ballpark as x86, that's why posting that is irrelevant. Compared to RISC chips it replaced (MIPS, PA-RISC) and those it outperformed, like SPARC, it was a better plan.

As for Itanium not being RISC, that's a bit fallacious IMHO. The ISA has fixed length instructions with a mostly load/store architecture and was designed to success PA-RISC. Under what viewpoint is EPIC not a variant of RISC?

> From the perspective of old RISCs, Itanium had no existing application base.

Irrelevant when you consider that SGI offered Quicktransit for IRIX to Linux users, HP-UX had compatibility, etc. They were plenty fast, to be objectively honest. But yes, to get full potential, native IA-64 apps. And those came! Oracle, for instance.

> Additionally, many old RISC users hated Itanium with a passion, because they thought Itanium killed their old favorite chip/OS. That's not great marketing when you have to fight your most loyal customers.

Other than some stubborn stragglers, most people didn't care with HP-UX, or the various Japanese companies that packaged Itanium, or even SGI -- most of their customerbase loyally migrated.

Also, lemme correct a misunderstanding. Windows on IA-64 was never intended for the desktop market. It was a product of Microsoft being part of the Itanium alliance, as far as I can tell, and looking to see how it would perform for the Windows Server customerbase, which has far less particular requirements than the desktop crowd. Outside of the early Merced systems, very few vendors targeted any sort of desktop platform for long, and it was always ghetto rigged when it existed. Neither was IA-64 intended to replace x86, it was a much higher end product to replace various RISCs.

The advent of x86_64/AMD64 didn't really change the IA-64 fate, fwiw. The premise that out-of-order cores will never be as wide as in-order proved false. That was true when it was fighting POWER5 and such.

[lunatic]

It simply wasn't in the same ballpark as x86, that's why posting that is irrelevant. Compared to RISC chips it replaced (MIPS, PA-RISC) and those it outperformed, like SPARC, it was a better plan.

Sure it was faster, but it couldn't keep the market of the old platforms.

As for Itanium not being RISC, that's a bit fallacious IMHO. The ISA has fixed length instructions with a mostly load/store architecture and was designed to success PA-RISC. Under what viewpoint is EPIC not a variant of RISC?

I guess one generally considers the complexity of VLIW to be in contrast to the simplicity of classic RISC architectures. I mostly stick to the classifications that other people have thought of. Otherwise you could get into serious trouble. Even later x86 CPUs did not stick to the CISC principle completely, but implemented RISC principles internally. This is however not a categorization that is close to my heart. There aren't that many CISC platforms left either (except for that one big one and maybe some embedded 68k?). Due to the embedded and mobile market, almost all the world uses RISC. That's why I would probably stick to traditional classifications.

But yes, to get full potential, native IA-64 apps. And those came! Oracle, for instance.

Would you want an Itanium machine to run Oracle? I know some people did, but not that many...
There were some announcements that some Itanium machine was now the fastest x-way system for Oracle or SAP and that's fine. Itanium had a good run in 2003 to maybe 2006. That's just not enough time to establish a platform when you only sell a few supercomputers or data base servers.

Other than some stubborn stragglers, most people didn't care with HP-UX, or the various Japanese companies that packaged Itanium, or even SGI -- most of their customerbase loyally migrated.

You mean like NEC and so on? I have only seen a NEC Itanium machine once as a frontend for an SX supercomputer. The lion share of all Itaniums was sold by HP. The HP-UX and VMS people kept it alive in the end.

Also, lemme correct a misunderstanding. Windows on IA-64 was never intended for the desktop market. It was a product of Microsoft being part of the Itanium alliance, as far as I can tell, and looking to see how it would perform for the Windows Server customerbase, which has far less particular requirements than the desktop crowd. Outside of the early Merced systems, very few vendors targeted any sort of desktop platform for long, and it was always ghetto rigged when it existed. Neither was IA-64 intended to replace x86, it was a much higher end product to replace various RISCs.

I had an early HP i2000 Merced machine. That was basically an Intel design. It could run Linux, Windows, HP-UX, a VMS beta, IBM ported AIX to it, and rumors have it that even Sun attempted a Solaris port. Windows actually was the only commercial thing that supported all the hardware including graphics. Windows XP on Itanium wasn't any different than XP on a PC. Well, the Itanium version was 64bit. At that point Windows on Itanium was a serious attempt at establishing a desktop system. Vendors quickly recognized that there was no market for such machines. Consider: the only Itanium2 workstations are the SGI Prism and the HP zx2000 and zx6000. The Prism is really rare. Hardly anyone bought it, even though technically, it's an Altix 350 with a graphics card. Similarly, the zx6000 is pretty much an rx2600 server with a graphics card and new fans. HP even sold the "office friendly conversion kit" for the rx2600. Hardly anyone bought a zx machine as a workstation. Most machines you will find are rack mount system. HP sold the zx6000 as a Linux server for a good price and many places ordered one to test the technology. HP for example sold the C8000 PA-RISC workstation in parallel, because they had many customers who wanted a PA-RISC workstation for CAD/CAM. After even this very small first generation of production quality Itanium workstations, the market was dead. The last batch of PA-RISC machines showed to many people that maybe switching to Itanium wasn't as great an idea as many thought. The PA-8800 und 8900 CPUs were pretty good and dual core, using the same socket as the Itanium2. The maximum you can get in a zx6000 is dual 1.5GHz Madison 6MB and 24GB RAM. In the C8000 you can get quad 1.1GHz PA-8900 and 32GB RAM. Essentially that makes the C8000 still the fastest HP-UX workstation. The same problem was present in servers. PA-RISC gave you twice the CPUs in any given form factor. It took Itanium another generation to go dual core.

Itanium eventually replaced PA-RISC (HP), MIPS (SGI), Alpha (also HP) and still failed in the end. Itanium failed to pick up essential market share that even the original RISC platforms had. Many customers didn't migrate to Itanium, but to commodity PC hardware.
I personally like Itanium and I have three machines. Truth is that Itanium couldn't establish a long term platform. If they had brought it to the market 10 years earlier, it might have had a chance.

If you just consider technical merits, the Itanium 2 machines were really good and solid designs. The Itanium 1 (Merced) machines were pretty bad and unreliable, but everyone said that this is rather a technology demonstrator you can use to port your apps. When you consider the Itanium2 machines, vendors put the CPU in their own proprietary designs. The HP Itanium and late PA-RISC systems used the same zx1 chipset. An Altix is essentially an Origin with IA64 instead of MIPS. There wasn't too much innovation on this front. I even have one Itanium machine that is really an in-box-upgrade for an older generation PA-RISC server. Keep your disks and power supplies and I/O, just replace the mainboard, memory, and CPUs. All this wasn't a bad strategy, but people wouldn't buy those high end Unix system as they used to. It's not like people then board SPARCs and pseries like crazy. Now all the world uses x86_64 and might see some more arm in the future.
Do you still remember when Intel replaced the Pentium IV with the Pentium M? The M was a descendent of the even older Pentium III architecture. It's main advantage was that it didn't use power like it was free and it had room to scale up. Itanium was way too much like the Pentium IV: a huge chip, always scaled up to the maximum, huge power requirements. When I fire up my Altix 350 it sucks like 3000W (measured running...).

[pcar]

Back when I was at Disney/ESPN, we bought 2 HP Superdome's that had 128 procs each (256 cores) running Winblows Server Datacenter edition that we then ran SQL Server on to crunch a bunch of ad serving data, basically trying to run predictions on what ads to queue up to serve to users when they logged in. Those nodes were running at 99% cpu for months crunching data and we never realized the outcomes we expected due to poor perf, ended up scrapping the whole thing.

[Raion]

Sure it was faster, but it couldn't keep the market of the old platforms.

I think comparisons to x86 are poor, and flawed.

I guess one generally considers the complexity of VLIW to be in contrast to the simplicity of classic RISC architectures. I mostly stick to the classifications that other people have thought of. Otherwise you could get into serious trouble. Even later x86 CPUs did not stick to the CISC principle completely, but implemented RISC principles internally. This is however not a categorization that is close to my heart. There aren't that many CISC platforms left either (except for that one big one and maybe some embedded 68k?). Due to the embedded and mobile market, almost all the world uses RISC. That's why I would probably stick to traditional classifications.

Itanium is not VLIW. EPIC is not the same as VLIW and has important differences.

My view is that RISC vs CISC is only good in historical discussions, and nowadays is relatively immaterial.

Would you want an Itanium machine to run Oracle? I know some people did, but not that many...
There were some announcements that some Itanium machine was now the fastest x-way system for Oracle or SAP and that's fine. Itanium had a good run in 2003 to maybe 2006. That's just not enough time to establish a platform when you only sell a few supercomputers or data base servers.

Oracle was a major app for the HP-UX server market at least.

You mean like NEC and so on? I have only seen a NEC Itanium machine once as a frontend for an SX supercomputer. The lion share of all Itaniums was sold by HP. The HP-UX and VMS people kept it alive in the end.

Oh nice, vector computers. And yeah, NEC, and many other firms made up portions. The Windows install base was never particularly big for it, and the Linux install base was also tiny (SGI was tiny in terms of marketshare)

I had an early HP i2000 Merced machine. That was basically an Intel design. It could run Linux, Windows, HP-UX, a VMS beta, IBM ported AIX to it, and rumors have it that even Sun attempted a Solaris port. Windows actually was the only commercial thing that supported all the hardware including graphics. Windows XP on Itanium wasn't any different than XP on a PC. Well, the Itanium version was 64bit. At that point Windows on Itanium was a serious attempt at establishing a desktop system. Vendors quickly recognized that there was no market for such machines. Consider: the only Itanium2 workstations are the SGI Prism and the HP zx2000 and zx6000. The Prism is really rare. Hardly anyone bought it, even though technically, it's an Altix 350 with a graphics card. Similarly, the zx6000 is pretty much an rx2600 server with a graphics card and new fans. HP even sold the "office friendly conversion kit" for the rx2600. Hardly anyone bought a zx machine as a workstation. Most machines you will find are rack mount system. HP sold the zx6000 as a Linux server for a good price and many places ordered one to test the technology. HP for example sold the C8000 PA-RISC workstation in parallel, because they had many customers who wanted a PA-RISC workstation for CAD/CAM. After even this very small first generation of production quality Itanium workstations, the market was dead. The last batch of PA-RISC machines showed to many people that maybe switching to Itanium wasn't as great an idea as many thought. The PA-8800 und 8900 CPUs were pretty good and dual core, using the same socket as the Itanium2. The maximum you can get in a zx6000 is dual 1.5GHz Madison 6MB and 24GB RAM. In the C8000 you can get quad 1.1GHz PA-8900 and 32GB RAM. Essentially that makes the C8000 still the fastest HP-UX workstation. The same problem was present in servers. PA-RISC gave you twice the CPUs in any given form factor. It took Itanium another generation to go dual core.

Desktop usage was not the main goal of Itanium, and neither was it well-suited due to being much more expensive to produce. Desktop and workstation usage aren't exactly the same markets, even. But yeah, most workstations of Itanium origin were ghetto rigged.

Itanium eventually replaced PA-RISC (HP), MIPS (SGI), Alpha (also HP) and still failed in the end. Itanium failed to pick up essential market share that even the original RISC platforms had. Many customers didn't migrate to Itanium, but to commodity PC hardware.
I personally like Itanium and I have three machines. Truth is that Itanium couldn't establish a long term platform. If they had brought it to the market 10 years earlier, it might have had a chance.

I think Itanium's biggest issue was being delivered late. If it had been on time, things would have been a bit different. That being said, it didn't "Fail". it just got obsoleted quicker than expected by POWER and other high end players beating it.

If you just consider technical merits, the Itanium 2 machines were really good and solid designs. The Itanium 1 (Merced) machines were pretty bad and unreliable, but everyone said that this is rather a technology demonstrator you can use to port your apps. When you consider the Itanium2 machines, vendors put the CPU in their own proprietary designs. The HP Itanium and late PA-RISC systems used the same zx1 chipset. An Altix is essentially an Origin with IA64 instead of MIPS. There wasn't too much innovation on this front. I even have one Itanium machine that is really an in-box-upgrade for an older generation PA-RISC server. Keep your disks and power supplies and I/O, just replace the mainboard, memory, and CPUs. All this wasn't a bad strategy, but people wouldn't buy those high end Unix system as they used to. It's not like people then board SPARCs and pseries like crazy. Now all the world uses x86_64 and might see some more arm in the future.

ARM will likely replace x64 because of how the Apple M1 is doing, and other products like Nuvia and such. That all being said, if we're talking POWER, POWER will probably continue to be developed for a while. I will probably stick to POWER8, 9 and 10 machines due to better hardware openness -- unless something ala Talos-inspired comes to ARM.

Do you still remember when Intel replaced the Pentium IV with the Pentium M? The M was a descendent of the even older Pentium III architecture. It's main advantage was that it didn't use power like it was free and it had room to scale up. Itanium was way too much like the Pentium IV: a huge chip, always scaled up to the maximum, huge power requirements. When I fire up my Altix 350 it sucks like 3000W (measured running...).

yeah, the A350's power usage is pretty high, even higher than some of my MIPS machines it replaced. But Pentium IV's netburst isn't really closely related to Itanium 2.

[lunatic]

Itanium is not VLIW. EPIC is not the same as VLIW and has important differences.

where do you get that from, given that for example wikipedia essentially starts the respective article with a description that EPIC is an implementation of VLIW? Differences, sure. What has no differences?

ARM will likely replace x64 because of how the Apple M1 is doing, and other products like Nuvia and such. That all being said, if we're talking POWER, POWER will probably continue to be developed for a while. I will probably stick to POWER8, 9 and 10 machines due to better hardware openness -- unless something ala Talos-inspired comes to ARM.

okay, so you now have the Power craze. Fine, this platform is going to be around for a while, I guess. It's probably a good choice if you like AIX. I still remember this thing said about AIX:

"AIX looks like it was implemented by a pretty smart space alien who
heard Unix described to him by a different space alien, but they had
to gesture a lot because their universal translators were broken."
    --unknown, but often misattributed to Paul Tomblin

Still, I had IBM products in the past and traditionally this is not a particularly open company, whatever the product says in its name. IBM products often have this "it works" design. Not very fancy, but it works... mostly. Another classic:

"Nobody has ever been fired for buying IBM".

AIX machines can also feel like there may be an evil spirit inside the box that may require sacrificing a goat or something in order to make it work. I have literally seen grown up men spending hours trying to configure network interfaces on AIX, and finally giving up crying. I once had a job where the task really was that they wanted some network service and had an RS/6000 machine. It could not be done. My job and the solution they finally chose really was to set up a Linux server right next to the RS/6000 and set up this service on Linux. The box was 99.9% idle, so in the end they kept looking for more services to put on this new Linux machine.

So far I have been a loyal Apple user for many years through all the ups and downs, but I admit I only got onboard for MacOS X. There is a trend to make everything into an iOS like child friendly system. My personal red line is the moment when they take the Unix away from me. That's when I'll leave this ship. I think the M1 chips are interesting, but I wouldn't count on it taking over x86_64. I certainly understand that Apple has some advantages when they design their own CPUs like they do on mobile devices.  Apple is really good with marketing. Like: "We haven't updated this computer model in a year and now you get a new one and it's faster! Aren't you excited?". Yeah, sure. It better be faster, because my mom can buy a PC that's twice as fast for half the money some place else.

yeah, the A350's power usage is pretty high, even higher than some of my MIPS machines it replaced. But Pentium IV's netburst isn't really closely related to Itanium 2.

No, it isn't. It just suffered from a similar design problem. Intel just couldn't find a way to make the chip fast without turning the machine into a noisy heat source.  People have been waiting for the end of Itanium for years, mostly because Intel could not keep up. Updated Itaniums were delayed over and over again and finally disappointed compared to the competition. It was clear that the end of the line was coming. Rumor was that only HP with its HP-UX and VMS commitment kept Itanium alive for so long. The disappointment was noticeable, because users thought they had just switched from one legacy platform, only to find themselves now on a new legacy platform.

and one EDIT:
I have heard for many years that users wished they could get a Proliant style server to run HP-UX or VMS. There was a well documented effort for an HP-UX port to x86_64 in order to keep the users happy, but eventually HP figured that users would be content with Linux. The VMS port that is now underway for a couple of years, follows the same logic. It's always a game about how much money you can milk out of the high end users. HP was also pretty good at it. Do you by any chance remember the HP3000? It's an old platform running MPE/ix, a proprietary non-Unix-but-sort-of-still-POSIX operating system. HP did a similar move like IBM did with its AS/400. They ported MPE/ix to PA-RISC and sold this as HP3000. In reality, this has been just a few HP9000 models that got a new front plate and a new firmware identifier. MPE/ix would refuse to boot unless it found a dedicated HP3000. HP-UX was less picky and you could easily run HP-UX on the HP3000. In fact, HP themselves used this to provide diagnostics. There were other issues. Some HP3000 were limited and ran underclocked with MPE/ix for no reason other than maintaining a price/performance difference. I once had a machine that had 200MHz CPUs in MPE/ix, but you could just as well boot up HP-UX from the usual CDs and it would detect and use a 650MHz CPU.
This is a side track of course, but I'm trying to say that a company can have diverse motivations to keep its customers locked into a proprietary platform. These days, every company claims that their system is open and that they love nothing more than your freedom, but experience often tells us (look at Apple) that it is more profitable if they can chain you to the system, so they can take all your money and you will still come back for more, because you have no choice.

[jwhat]

Hi Raion & Lunatic,

in server market RISC/CISC is no longer important.

Virtualisation is king and pretty much all enterprise workloads are running on VMs or via Containers nowadays.

I looked up VMWARE, they started in 1998 (20 years ago), Itanium Merced was released in market in 2001 (also around 20 years ago) and AWS cloud appeared in market in around 2004/2005.

While much is written about x86_64 vs IA64 and CISC vs RISC I am surprised that virtualisation is not part of discussion as this what has really changed the enterprise compute market.

I know that IBM laid the ground work for HW virtualisation and Intel had virtual 8086 mode in the 386 processor.

Did Itanium have any virtualisation support ?

While I love my SGI / MIPS machines, MIPS would need more that just a faster MIPS processor to survive now, it would have to also have virtualisation support.

I believe that virtualisation extensions were added to MIPS but not sure when.

I know that POWER and RISC-V have virtualisation capabilities.

My question is: "Can an CPU with virtualisation support, still be a RICS processor ?"

Does HW virtualisation require introduction of level of silicon complexity that means while the individual VM might see a "RICS machine" the provider of this is intrinsically a "CISC" machine ?

Cheers from Oz,


jwhat/John.

[robespierre]

Hello jwhat,
Hardware virtualization (HVM) was in use commercially since the 1970s on IBM mainframes. The reason it was used then is the same as "cloud computing" now: to bill clients for fractional machine time where the client has access to a complete operating system environment.

Architectures that support HVM must satisfy certain criteria that have been widely known since the 1970s: https://en.wikipedia.org/wiki/Popek_and_...quirements
There is nothing about CISC or RISC in them, only some rules about supervisor state. However, MIPS was designed for maximally performant workstations and other small machines, not for mainframes, so it was not designed with HVM as a use case. There are decisions in the MIPS design that frustrate virtualization, such as its exception and cache architectures. Later, after the demise of SGI, some virtualization extensions (VZ etc) were added to MIPS64.

Many of the applications that benefit from virtualization can also be paravirtualized (PVM) and this can be done on almost any architecture, as it relies on software changes.

[Raion]

where do you get that from, given that for example wikipedia essentially starts the respective article with a description that EPIC is an implementation of VLIW? Differences, sure. What has no differences?

I apologize for being late on reply, been very busy with my work. This has been answered dozens of times by people much smarter than me and who can word it better, i.e. here: https://www.quora.com/What-is-the-differ...W-and-EPIC

I believe the differences in implementation are unique enough that classing EPIC as VLIW or a type of VLIW is incorrect and ignorant. That's all.

okay, so you now have the Power craze. Fine, this platform is going to be around for a while, I guess. It's probably a good choice if you like AIX. I still remember this thing said about AIX:

"AIX looks like it was implemented by a pretty smart space alien who
heard Unix described to him by a different space alien, but they had
to gesture a lot because their universal translators were broken."
    --unknown, but often misattributed to Paul Tomblin

I'm gonna run POWER Linux likely with KVM to develop my own POWER-based OS, and then when that OS is sufficiently stable, switch to that on bare metal.

AIX is big-endian and highly cost-ineffective and impractical, and requires specific POWER machines (IBM, Tyan and other vendors sell cheaper products only intended for FOSS, non-AIX applications)

So far I have been a loyal Apple user for many years through all the ups and downs, but I admit I only got onboard for MacOS X. There is a trend to make everything into an iOS like child friendly system. My personal red line is the moment when they take the Unix away from me. That's when I'll leave this ship. I think the M1 chips are interesting, but I wouldn't count on it taking over x86_64. I certainly understand that Apple has some advantages when they design their own CPUs like they do on mobile devices.  Apple is really good with marketing. Like: "We haven't updated this computer model in a year and now you get a new one and it's faster! Aren't you excited?". Yeah, sure. It better be faster, because my mom can buy a PC that's twice as fast for half the money some place else.

I used to be an Apple user, but over time found the anti-power user interface and design frustrating, I hate the fisher-price scheme it has, and find the overall architecture subpar, especially in its filesystems (I HATE RESOURCE FORKS AND .DS_STORE BS) and interoperability.

This is a side track of course, but I'm trying to say that a company can have diverse motivations to keep its customers locked into a proprietary platform. These days, every company claims that their system is open and that they love nothing more than your freedom, but experience often tells us (look at Apple) that it is more profitable if they can chain you to the system, so they can take all your money and you will still come back for more, because you have no choice.

Apple's business practices sicken me, plain and simple. That's why I would almost never give them any money.

Jwhat, as for RISC/CISC, this matters not in the current year and only matters in historical contexts really. There's no reason to try putting computers in these discrete, rigid boxes. MIPS is dead, it'll never have another day to shine. SPARC had some virtualization, as does ARM too -- there's nothing inherent to VT support.