Your reasons why also remind me of this thread by Agner Fog about the possibility of creating a new ISA.
Ah yeah, I saw that! I haven't read through it all yet but it is definitely an exciting idea, and Agner Fog would definitely be the person for the job.
He's since started a new thread
because the old one was becoming a bit unwieldy. It has been a fruitful effort however in that the spec is solidifying enough that he's put out some ISA documentation on a new site dedicated to this effort
, as well as a GitHub repo
The main concern about the ISA being brainstormed there is that (I believe?) it isn't being designed for compatibility with what we have today. That is a fine long term strategy so hopefully it takes off, but it won't be something people will have in their machines any time soon. RISC-V is exciting in that it offers an open architecture that can run Linux, and we may have SoCs being produced as early as 2017.
Sure, but the same can be said for RISC-V as well. It's not compatible with x86, ARM. MIPS or POWER either, to name but a few. Reading through the initial thread a while ago, I believe he was aware of RISC-V, but felt it didn't address all of the niggles he had with the x86 architecture.
IBM's POWER architecture is also interesting; Raptor Engineering is looking to produce a high end workstation, but that is $2-3K for what is essentially server grade hardware, which is a tougher sell. I'm pretty excited about RISC-V in that it is likely we'll be able to buy something roughly equivalent to a raspberry-pi within the next few years.
Leaving aside for a moment the question of whether or not everyone has the same concerns, I think both propositions offer value over the status quo. They're also facing an uphill struggle getting reasonable adoption, but as you point out here, RISC-V at least is close to having working hardware.
There's still another CPU design with a new ISA that I'm sort of kind of interested in, taking an altogether different approach to other extant and nascent designs, that being the Mill architecture
designed by Ivan Goddard. The videos they've put out (linked from their site under docs), certainly show an interesting new way to think about CPU architectures.
Whether or not we'll actually see working hardware any time soon is anyone's guess. If they can live up to their claims about better ILP and the hardware offering a much better performance/Watt than existing solutions, I expect some of that fame to rub off on other clean rethinks ISA efforts. Although I've been following their developments with some interest since their announcement a few years ago, I'm going to treat it as if they announced an improvement in battery tech: it's always 5 years away. Until actual hardware starts being sampled, it's too soon for me to tell whether it's really a sea change waiting to happen or not.
Getting back to your proposed project however, one of the reasons I like it is that it's not entirely unlike an educational project for Handmade Network that I've had in mind doing for the past year. In that multi-stage effort I'd walk people through designing an ISA from scratch, implementing an emulator for said ISA, porting a toolkit and bringing up Linux, and then finally teaching low-level optimisations work. Well, that and/or doing much the same but emulating even wimpier hardware (roughly 386 equivalent ISA, but running at 16MHz with 16Mb RAM) and write something like a 'scene demo' as a vehicle for exploring low-level optimisation.
That would however be a multi-year effort, and it's unclear what level of interest there would be in it. I may end up having to pare down the stages to just the ones with the most interest, like just showing how to bring up Linux on new hardware, or how to port GCC to a new target. That or make it a cooperative effort with others, so that one project shows how to design an ISA from scratch. Another shows how to write an emulator from scratch and so on. We'll see. It's not as if it's the sort of knowledge that'll be less relevant going forward, so there's time yet to weigh the pros and cons of these ideas. (And I do have around 5 years worth of projects lined up already, not counting this one, so paring things down may be an inevitability).
In any event, I for one am looking forward to your project and would like to subscribe to your newsletter, as the saying goes.