The Itanium processor was introduced in 2001 as the vanguard of Intel’s new instruction set. It was expected at the time that IA-64 architecture would displace x86 first in the server room and then personal computers, but things did not pan out that way. The history of IA-64 was reported on earlier this year by Abort Retry or Fail which stuck to the orthodox interpretation of it being a “titanic” failure. It is now abandoned. One generation on from its release, I suggest instead that it is time to look at the IA-64 legacy with fresh eyes. Could it become the next open ISA and in so doing, assist Intel’s Foundry venture?
Itanium would be the only processor line to use the so-called EPIC (Explicitly parallel instruction computing) paradigm, as opposed to RISC, CISC and the WLIW it borrowed from. EPIC was the result of over a decade of painstaking research from Hewlett Packard and then Intel. HP, who already had their successful PA-RISC architecture, envisioned that paradigm would run out of steam. Despite all the care taken in its design, and perhaps inevitably for a sharp departure from what went before, it underwhelmed at launch. Worse, AMD offered a 64 bit extension to the venerable x86 which was the boring, and therefore savvy, choice.
The tale is one of complacent management and a certain amount of bad luck rather than any fundamental flaw, followed by malign neglect. Software compilers were initially not up to the task, although researchers did make strides early on. There is no reason to believe they would not have continued to do so if interest had not waned in the architecture. There is no shortage of might-have-beens. If only Intel had been magnanimous enough to bring AMD onboard and rolled out the ISA together, we may have stepped boldly into the post-x86 age two decades ago. Instead it ended up being Intel who had to adopt AMD’s x86-64. This is on Intel who really should have known by then how difficult it would be to displace the entrenched architecture. x86 had escaped several other efforts to replace it, and if that taught them anything, you would think it would have been the need to kill x86 with fire.
Even without bringing onboard their main competitor, how things might have been different if the instruction set had enjoyed a big break as did others? MIPS and PowerPC had been used in consoles, and PowerPC enjoyed an illustrious career in the Apple Macintosh. Game consoles in particular have used chips that were notoriously hard to program. If anyone could figure out how to optimise software for Itanium it would have been game developers!
As it happened there was little reason for anyone outside specific niches to take notice of Itanium. As a professional product, would never be marketed directly to consumers over the heads of vendors as with their Pentium and Core sub-brands. In this vacuum many of us would only hear about the processor in connection with its nickname Itanic and assumed it had sunk without trace. In fact, until reading up about it recently, I had no idea it persisted so long! Around the time Itanium first came out I was moving from x86 to PowerPC mac, then about five years later back to x86 mac, all the while with ARM mobile phones, so like most people never had personal experience with which to challenge the narrative spread by internet wise-guys.
For younger readers who might be puzzled as to why Itanium got its sobriquet, in the late 1990s James Cameron directed the movie “Titanic” which pervaded deeply into popular culture at the time and became a lazy metaphor for almost any large failure; even the movie was notoriously over budget.
We might today compare attitudes towards IA-64 with those towards RISC-V, currently the darling of processor design. Could we accuse the name, referring to the fifth RISC design from Berkley, as being itself somewhat insular? One might also say RISC-V is merely a distillation of good design principles that already exist rather than breaking new ground. Fans point to RISC-V being open hardware, but RISC designs in the open domain are already a dime a dozen (Power, SPARC, SuperH and others). RISC-V cannot compete in performance with modern ARM designs, and there is little software support, yet unlike IA-64 at launch, RISC-V gets a free pass on these matters. I do not all say this to hate on RISC-V, just to illustrate the difference of public treatment, seemingly based on a perception of RISC-V being the plucky upstart. Could IA-64 pick up a few tricks from this newer architecture?
Clearly Intel was keen to jump on the RISC-V hype train and was at one point rumoured to be courting Si-Five, a design house specialising in the architecture, for acquisition. But why not have their own iron in the open ISA fire?
A good start would be for Intel to release IA-64 – for which there is no reason not to do so since it is no longer considered a living ISA – under an open licence so that its unique design may be studied. It would follow the footsteps of celebrated rivals such as IBM’s Power and Sun’s SPARC which entered the public domain and remain – to some extent – living architectures. As an EPIC rather than simply another RISC, and certainly not x86, it would be technically intriguing to the computer hobbyist! Intel strives not to be the same complacent firm as existed a couple of decades ago, and opening this ISA would be good public relations. I also think Intel owes us that in return for all the CHIPS act US taxpayer money.
It might also be a business opportunity for Intel. They are offering services now as a chip foundry who will build whatever customers want. If they did open IA-64 they stand to be the go-to fab because who better to design chips on any architecture than the original co-developer?
Might Intel promote their foundry capabilities and proud heritage – whilst reaching out to enthusiasts – by introducing a single board computer incorporating IA-64 in time for the 25th anniversary of that architecture? It could run on an improved version of IA-64, given that we don’t really need to worry anymore about backward compatibility. The Raspberry Pi and its imitators helped ease desktop linux onto ARM, and we have RISC-V single board computers serving to legitimise that ISA amongst hobbyists and give it the grassroots popularity we noted above. Many such hobbyists associate Intel with the x86 PC monoculture – a single board computer with novel architecture can help change that perception. I suspect a crowdfunding campaign would unearth considerable interest. Perhaps this Cinderella ISA might finally escape her like of drudgery in the server room and go to the ball?
Might I be so bold as to suggest a name for the SOC – Turbinia (or perhaps Turbinium in the style of Pentium, Itanium, etc)? What bolder act of self-effacing humility than to name it after another British steamship from the turn of the twentieth century? But perhaps not too humble to be naming your single board computer after the feisty little craft that steam turbine inventor Charles Parsons built to demonstrate the potential of his technology!
Now that a generation has passed since IA-64 was introduced, maybe it is time for us to revisit its legacy? Its botched introduction is now receding into memory, and on a technical level Intel has a lot to be proud of. Consider that ARM is only now gaining traction for high powered applications, and that was invented back in 1987, I do not think it is too late for IA-64. Could this story have a fairytale ending? And, finally, might we all stop calling it Itanic? It wasn’t especially funny in 2001; certainly not now in the mid 2020s and most of us are still stuck with x86!
Ecotechnia Radnorshire 