Will anybody save Linux on Itanium? Absolutely not
It's doomed to sink... but the how and why is interesting
Analysis No sooner than Intel IA64 support is removed from the Linux kernel, complaining about it begins… but the discussions are fascinating.
Although the proposal to remove support for Intel's infamous Itanium architecture - aka Itanic - from Linux was rebuffed in February, just weeks ago, in October, the move was approved for kernel 6.7. Just as one might expect, this has made a few people very angry and been narrowly regarded as a bad move. LWN has an excellent analysis about this under the title of the Push to save Itanium.
Linus Torvalds, meanwhile, has responded, saying:
I'd be willing to resurrect Itanium support, even though I personally despise the architecture with a passion for being fundamentally based on faulty design premises, and an implementation based on politics rather than good technical design.
He offers a perhaps atypically modest and reasonable proposal, neatly summarized by @mewse on Lobsters:
You people who still want the architecture, maintain an out-of-branch patch set for the architecture for a year and we'll consider making it mainline again.
Translation: if the people who are complaining about needing more time for this change, suddenly find the time to modernize the code they don't want removed, it can be re-added. It will not happen and I imagine everyone knows it won't happen.
We couldn't put it better ourselves. We especially admire this summary of how and why the Itanic set sail at all. We highly recommend it, but in case some of the terminology is unfamiliar, we will also attempt to unpack it.
To summarize the summary, when Intel began its EPIC project – no, really, it stands for Explicitly Parallel Instruction Computing [PDF] – the idea of out of order execution (OoOE) in microprocessors was new, and for x86, unproven. In brief, the concept of OoOE is that processors can break down complex x86 instructions into smaller, RISC-like chunks, resequence them on the fly to run them as fast as possible, and then reassemble the results into the order that the software originally expected. In 1994, it wasn't certain that future x86 processors would be able to effectively exploit the instruction level parallelism, or ILP as HP called it [PDF], in machine code. So rather than bank on major improvements in the design of CPUs to execute existing code, it looked like a safer bet to get the compiler to do that in advance, and to design a whole new instruction set with Very Long Instruction Words to make that possible.
The meta-analysis on Lobsters is also an interesting read.
The Itanic isn't the only Intel CPU architecture that disappeared under the waves, of course. We recently talked about iAPX432, the company's late-1970s attempted mainframe on a chip. The very complex design of the iAPX432 was replaced with a simpler RISC one to create the i960 RISC chips, which were just fading away as The Register arrived on the scene in the late 1990s.
The i960 was also Intel's first OoOE processor, and its lead architect, Intel superscalar boffin Fred Pollack, also worked on the Pentium Pro, a design so good that it sank the Itanic. The i960 was later succeeded by another Intel RISC chip, the i860, the chip on which Windows NT was developed. The i860 went nowhere too.
- Linux will soon offer switchable x86-32 binary support
- Intel's PC chip ship is sinking with Arm-ada on the horizon
- Want to live dangerously? Try running Windows XP in 2023
- Debian dev to the rescue after proposal to remove Itanium from Linux kernel
Even within the world of x86, it has had to climb down and backtrack. The Pentium Pro provided the design of the CPU core in the Centrino family of chips from Intel Haifa in Israel, which saved the company from the big, hot, and uncompetitive Netburst architecture of the P4. Netburst never delivered its promised 10GHz, and instead, Intel switched back to the Pentium-III-derived core from the Pentium M. ®
"Itanic" is of course the Register's very own, academically approved sobriquet for Intel's doomed VLIW architecture. The only other VLIW machine that the FOSS desk knows reached the market was Transmeta's Crusoe – one of our articles about which, ironically, also mentioned Intel's Fred Pollack.