Linux Developers Still Working To Retire Intel Itanium/IA-64 Support

[phoronix]

Phoronix: Linux Developers Still Working To Retire Intel Itanium/IA-64 Support

Back in February was a patch series proposed retiring the Intel Itanium (IA-64) architecture support from the Linux kernel. That removal has yet to take place in Linux Git but it's still being talked about and user-space developers are also eager as it would mean being able to clear out Itanium user-space code too...

https://www.phoronix.com/news/Intel-IA-64-Retire-Still-WIP

[uid313]

Was Itanium a technically retarded architecture, or was it ahead of its time, or was it just poor project management, bad rollout, stupid marketing, and stuff like that?

Did Itanium have great ideas? Did Intel learn anything from it and applied it to x86?

[Jonjolt]

Well they don't call it Itanic for nothing.

[rmfx]

One day is gonna be x86 turn.
This crap also lasted for way too long.

[rene]

One day is gonna be x86 turn.
This crap also lasted for way too long.

yeah, they should rather finally remove this 40 year old complex x86 mess instead of this epyc, and big enterprise ready, modern IA64 code, ... https://www.youtube.com/watch?v=UNob4G2Zr8s

[jacob]

One day is gonna be x86 turn.
This crap also lasted for way too long.

In its latest incarnation (amd64 + avx512), x86 is actually a pretty damn good and efficient instruction set, very much NOT like the 8086 of old. But I agree that sooner or later it's going to hit a wall: namely, the cost and overhead of instruction decoding will become prohibitive and create an intractable performance bottleneck. However I hope that this will happen as late as possible as there's no serious alternative on the horizon. ARM is proprietary as hell, seems to be heading towards a strictly locked down ecosystem and despite all the talk, CPUs even remotely competitive in terms of absolute performance (not merely per-watt) with Intel or AMD seem to only ever exist on paper with the partial exception of M1/M2. POWER is open and performant, but very expensive and its ISA complexity problem is at least as bad as x86's. RISC-V... well I suppose it's an option for those who fancy paying $800 for an ultra low end microcontroller.

[Dawn]

In its latest incarnation (amd64 + avx512), x86 is actually a pretty damn good and efficient instruction set, very much NOT like the 8086 of old. But I agree that sooner or later it's going to hit a wall: namely, the cost and overhead of instruction decoding will become prohibitive and create an intractable performance bottleneck. However I hope that this will happen as late as possible as there's no serious alternative on the horizon. ARM is proprietary as hell, seems to be heading towards a strictly locked down ecosystem and despite all the talk, CPUs even remotely competitive in terms of absolute performance (not merely per-watt) with Intel or AMD seem to only ever exist on paper with the partial exception of M1/M2. POWER is open and performant, but very expensive and its ISA complexity problem is at least as bad as x86's. RISC-V... well I suppose it's an option for those who fancy paying $800 for an ultra low end microcontroller.

Re: Power: How is it "almost as bad as x86"? Power isn't variable length (except in the e200, which only automotive users care about, and the limited handful of prefix ops introduced in Power10.) Building and validating a conformant PPC core is not really any worse than building and validating a conformant ARM. It's notable that PPC4xx has typically compared very well to contemporary ARM on both area and active power.

Re: ARM: Altra and Grav2/Grav3 were all competitive at release. So was A64FX for its target workloads. PCs aren't everything.

[Eirikr1848]

Bought a brand new HP Itanium rack mounted workstation in 2021. Getting Arch to work on it has been a biotch, and then basically living off compiling everything else locally.

Did not realize it was hanging on worse than powerpc in terms of kernel support.——-

As I am apparently the 1/8,000,000,000 odds human of having an Itanium rig in my rack that I actually use with an Instinct card… I’m sad I have to now learn how to maintain kernel, gcc, and grub support

Well. Crap. That sucks. 2 years, few thousand $$ wasted. Glad I also bought a POWER system.

[Estranged1906]

Bought a brand new HP Itanium rack mounted workstation in 2021. Getting Arch to work on it has been a biotch, and then basically living off compiling everything else locally.

Did not realize it was hanging on worse than powerpc in terms of kernel support.——-

Don't worry, I'm sure you can slap NetBSD on it, it even runs on toasters, so surely Itanium won't be a pro- oh.

[Eirikr1848]

Don't worry, I'm sure you can slap NetBSD on it, it even runs on toasters, so surely Itanium won't be a pro- oh.

Catching up on notifications and noticed this. Thanks, actually. I’m still using Arch on it for now but maybe a *BSD is the way to go longer-term