In the spirit of speculative off topic wibble, I was discussing the Itanium processors on the MNT forum and wondered what people thought of the prospects of the underlying IA-64 architecture in the future?
I know Intel have stopped producing them, but they are the only processor to use so-called EPIC (Explicitly parallel instruction computing) paradigm whose research and development is summarised in the linked article. Others paradigms include RISC, CISC, WLIW etc.
There are so many RISC designs in the open domain they are a dime a dozen (POWER, SPARC, RISC-V, SuperH and others) yet no other EPIC designs. It just needs the collaborative refinement from being open to make it fulfil its promise. I wonder if Intel might open source this ISA since, IMO, they have no reason not to anymore. The architecture might gain a second wind due to this lack of competition within its field, since it would not be subject to the fragmentation that exists between myriad of open (not to mention proprietory such as ARM) RISC designs.
I would definitely love an Open Itanium processor powered laptop, as I think that the bones of this architecture are much better than x86 and most likely also the RISC designs. This might be a cut down version similar to PowerPC compared to IBM’s POWER server chips. What do you think?
I’m not familiar with intel stuff, but if that architecture was as great as you try to make us believe, why did it get discontinued?
Anyway, EPIC might have been of interest, as long as CPUs were still single core ones, but due to the current multi-core availabilty, this sounds obsolete to me.
I speculated on the other forum that if only Intel had been magnanimous enough to bring AMD onboard and rolled out the ISA together, they could have stamped out x86 altogether. x86 had outlived Intel’s 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 your main competitor, how things might have been different if they had got that architecture into a games console. Playstation has used some wild chips in the past, that were renowned as difficult to extract their potential. If anyone could figure out how to optimise software for Itanium it would have been game developers!
So I think that Itanium was the victim of poor management and corporate greed rather than any technical failing. I would argue it was technically ahead of its time: the middle of my three links in the OP show the care that went into the design from very experienced engineers who were looking at a future beyond RISC. And despite the wiseguys who kept calling it Itanic, it was in production for nearly two decades.
For single user computers you still want powerful cores, because not all applications lend themselves to being divided across several small cores. There is a reason that modern processors will have some power cores rather than just lots of efficiency cores.
Finally, consider those here (me included) who look forward to usable RISC-V mainboards. One could say RISC-V is nothing special, not even limiting yourself to open hardware, it just happens to be that currently most trendy. It was pointed out to me that several generation old ARM single board computers blow any RISC-V option out the water, for which there is not much native software.
For the daydream processor IA-64 seemingly has much more to offer. It would be easy to build something as strong as the best processor using that ISA (since it drives high power server chips), it might rather be necessary to throttle it to the power you need. Somebody else might be able to tell me whether legacy Itanium designs made with state-of-the-art technology and the feature size of its competitors might be competitive. But as an EPIC rather than simply another RISC, and certainly not x86, it would be technically more interesting too!
xd idk who’ll manufacture them. Unlikely there’s fab space. There’s reasons why Itanium failed too
EPIC was definetly really cool though! Not sure I’d like a comeback considering the fragmentation we’re going through in the industry, but interestingly nonetheless.
Sob. I worked for 30+ years on machines using HP’s previous processors (16-bit stack-oriented, then 32-bit PA-Risc) and their old proprietary operating system. PA-Risc was wonderful, IA-64… let’s just say… our community dubbed it “Itanic”. The mere mention makes me want to run and hide.
I always thought Itanium failed because it ran x86 software too slowly.
Back then I doubt anyone did the impressive just in time translators that I believe are used now for things like running x86 on ARM processors.
Also, the necessary compiler to get the best from the EPIC instruction set might have been something that would have been difficult to do well.
Now we have pipelines etc in our processors that attempt to do a lot of similar things automatically from normal code (but don’t always choose the optimal solution), the advantage from using EPIC instructions might be less. I suspect EPIC is likely to have some disadvantage due to the code being larger too – one of the same disadvantages that can make x64 code slower than x86 in a few cases.
Intel? This is not the same complacent firm as existed a couple of decades ago. They are offering services now as a chip foundry who will build whatever customers want. If they did open up IA-64 they would be the go-to fab because who better to design chips on any architecture than the original co-developer? Incidentally, Intel is now looking once again at what to do as the x86 desktop market dries up and ARM makes inroads into that and data-centre hardware.
Intel was a couple of years ago kicking the tyres of SiFive, a design house that specialised in RISC-V. It will probably want to have a RISC design in its back pocket for lower powered applications. But it can also open up IA-64 and offer something more powerful that it still dominates due to being able to draw upon in-house heritage.
Fragmentation? Remember in the 1990s when there were several RISC workstations, an different proprietary UNIX to run on them. Never mind other weird stuff like LISP systems. Now everything is a dreary monoculture on Linux or Windows, both of which are rehashes of 1970s thinking.
You realise that the EPIC paradigm underlying IA-64 was actually HP’s invention? It was them who partnered with Intel to develop into a marketable product, might be considered part of this celebrated lineage of HP hardware! I am sorry if the memories are so raw, since it all happened a (meatspace) generation ago.
It is good to hear from actual experience from the time. Most of us would not, it passed consumers by. In fact til I read up about it recently I had no idea it persisted so long. When Itanium first came out I was moving from x86 to PowerPC mac, then a bout five years later back to x86 mac, and then owned ipod touch and mobile phones with ARM. In the meantime I would have played on consoles with MIPS and PPC. Never would I have had a chance to directly use anything with IA-64.
It is interesting to speculate about the technical merits, and the larger code size has been identified as a disadvantage. But since all software is so bloated these days, does it really make so much difference? Compiler design and everything else needed to optimise code has advanced since the early 2000s, so maybe the technical barriers have fallen away?
Now that a generation has passed since IA-64 was introduced, maybe it is time for us to revisit it as an ISA for the future? ARM is only gaining traction for serious applications now, and that was invented back in 1987, so I do not think it is too late for IA-64. In fact the lack of development might work to its advantage because it is less encrusted with myriad extensions that all the legacy ISA, and particularly x86, has.
We were all following IA-64 closely, and were very, very, VERY aware of the HP / Intel collaboration – an evolving mixture (as best I recall now) of thoughts that they might eventually migrate us to it and fear that focus on it would lead to abandonment of the platform we all depended on. In the end they discontinued us and stuck us with unpleasant migration choices. The alternatives HP proposed to us were rather offensively bad.
I put my thoughts into an essay here. Essentially I think an Itanium inspired single board computer would be a great publicity stunt for Intel’s new foundry service and a nice toy for hardware geeks.
What do you all think? Perhaps the similar Elbrus-2000 might be an option, if only international relations can be restored in the future.
Perhaps one of Framework staff would like suggest an IA-64 SOC when they next see their Intel sales rep?
