"In February 2003, the original founders saw that key problems with general purpose processor architectures simply were not being addressed adequately by any contemporary CPU, and so rethinking CPU architecture was something that 'just needed doing.' ...
In January 2004 Out-of-the-Box Computing came into existence as a “company in formation. ... In March 2014 after a decade of work Out-of-the-Box Computing was incorporated as Mill Computing, Inc. ..."
This is a rather leisurely pace.
Their paper on split-stream instruction decode is kind of interesting.[1] It's a statically scheduled very long instruction word machine, like the Intel Itanium.
That solves some problems while creating new ones. Most of the ones it solved relate to our of order execution, which, while complicated, works pretty well now. The ones it creates involve trying to make compilers smart enough to optimize the static scheduling. The team working on that at HP for the Itanium was beating their head against the wall in the 2000s, when they gave a talk at Stanford, but maybe it could be solved today.
Also, the niche they hoped to occupy, an alternative to Intel's architecture, is now filled by ARM in its 64-bit form, Apple's CPUs, and maybe RISC-V. They may have missed the window for this product.
I like this yearly exchange. I like that the original poster apparently keeps a calendar, and that the response is very prompt— and then they just let it go fora full year. No argument. Just one annual snipe with with one annual response. I eagerly await June 2023.
> There hasn’t been much news because we have no particular reason to put news out.
This seems kind of oddly dismissive for a project people are interested in. They don’t want to be seen as dead, but weren’t giving any information to counter the narrative other than, eventually, saying something like “we’re not”.
The next paragraph lists a ton of stuff they’ve been working on. Couldn’t they just mention some of that stuff over time? Not big detailed posts (how often to people bring up an environment for a totally new way of doing things? Blog it!), but just a simple little status update here and there.
I too am very disappointed in the lack of progress. IMHO the window for the Mill ever to achieve any kind of commercial success, even in a niche, has been completely co-opted by RISC-V. That's a shame, because while RISC-V is clean, it isn't innovative. (In particular, I find our continued reliance on page-based memory protection schemes prevents the adoption of new OS paradigms for security. But of course this is for general-purpose computing which, afaik, was never the focus of the Mill project.)
That's the triumph of vanilla hardware. There are lots of cool things you can do. Segmented architectures, all that call gate stuff Intel put in 32-bit mode, tagged machines, unified integer/floating point implementation, hardware key/value stores - they've all been built, and they're all history now.
There's a long history of exotic CPU architectures. The only ones that caught on were GPUs. We might see new ones for machine learning, which needs some very specific operations with huge parallelism. GPUs are gaining features for that, like tiny floating point optimized for the 0..1 range. The successful ideas in this space were driven by a specific problem. None were "build it and they will come".
At least tagging is making a comeback thanks to exausting all other mitigation strategies on how to fix C, Objective-C and C++ memory corruption issues.
The project is dead, they spent all the money and there’s none left. The things the responder says is exactly what a con artist would say to keep stringing people along.
The lack of specifics about this technology plus the lack of communication led me to believe early on that it was not real. The notion that progress can only move forward with a large team, vs a small one, seems more like an argument you use to get more money out of a VC than a real engineering statement. What type of work requires a large team? Heck even if you're building a sky-scraper you could get started with a small team by, for example, making blue-prints.
I catch a whiff of con-artistry, and I find it deeply distasteful.
I've always assumed it was earnest enough, but steeped deeply in the academic culture of "OK, we've done the hard part; implementation is left as an exercise for the reader"
Oh yes, I remember being inspired by the idea of reversible compute. That very much appeals to the physicist in me. But it was vague even back then, and resurfacing with this nonsense about needing a big team to make any progress tells me at least they aren't serious, and at most they are actively out to scam some money out of people.
Maybe you have this project confused with another one? This is an ordinary exposed-pipeline VLIW machine with a novel “temporal addressing” scheme rather than named registers. There’s no mention of reversible computation.
This is a rather leisurely pace.
Their paper on split-stream instruction decode is kind of interesting.[1] It's a statically scheduled very long instruction word machine, like the Intel Itanium. That solves some problems while creating new ones. Most of the ones it solved relate to our of order execution, which, while complicated, works pretty well now. The ones it creates involve trying to make compilers smart enough to optimize the static scheduling. The team working on that at HP for the Itanium was beating their head against the wall in the 2000s, when they gave a talk at Stanford, but maybe it could be solved today.
Also, the niche they hoped to occupy, an alternative to Intel's architecture, is now filled by ARM in its 64-bit form, Apple's CPUs, and maybe RISC-V. They may have missed the window for this product.