AT&T's Hobbit Microprocessor (2023) | Hacker News Reader

AT&T's Hobbit Microprocessor (2023)

thechipletter.substack.com

105 pointsklelatti36 comments

However the Hobbit wasn't a stack machine, it was a memory-to-memory machine with a stack cache.

The LLVM uses a register machine with an unlimited number of once use registers. The Hobbit could easily run LLVM programs with no change to the number of registers used. Compiling as to reuse registers would only decrease the amount of stack memory traffic.

The stack cache is effectively a form of register window. The ISA could have or might have (I don't recall) short form instructions that take shorter offsets from the top of the stack which would be similar in performance to register-to-register operations.

Most compilers either use an "infinite registers" abstraction or a "no registers" abstraction in the layer that LLVM occupies. It is easy enough to translate from there to "16/32/64 registers."

The Hobbit was the original CPU for the BeBox, one of the most fabled 1990s “what could have been” hacker machines:

https://en.wikipedia.org/wiki/BeBox

I dream of owning one of those one day! I was a huge BeOS fan as a teenager, and even though the BeBox was from before my time I’d love to own one! They very very rarely seem to come up for sale, though. And the ones that do are pretty crazily priced..

I have one. What would you consider a reasonable price?

Of course it depends on the model and the condition, but somewhere around or under 5-600 euros?

IIRC there were only ~1000 66MHz and ~800 133Mhz BeBoxes produced.

In comparison, I'd be curious as to how many, say, SGI O2s were ever produced in total.

My husband worked for Go Corporation and there is a small pile of EO Personal Communicator bits and pieces sitting in our garage.

I was always on Team Psion, and I still miss the foldable-with-keyboard form factor today. It seems perfect, but despite one failed attempt from Planet Computers, no longer exists in any useful product.

>> With apologies to J.R.R. Tolkien

Why? The word "hobbit" predates Tolkien and was used in a variety of ways. It might seem pedantic, but while the use of old words is one thing, there is a trend as of late whereby authors who popularize old words then claim ownership. Tolkien is beyond this but others such as the Potter franchise are not (Padfoot).

https://en.wikipedia.org/wiki/Hobbit_(unit) https://en.wikipedia.org/wiki/Hobbit_(word)

The apology is for the parody of Tolkien not for the use of the word Hobbit.

Because the "Hobbitses" intro is a rephrasing of Gollum/Smeagol.

Oddly, the Wikipedia page doesn't include the most likely inspiration, the Hobyahs: https://en.wikipedia.org/wiki/The_Hobyahs

The parallels are pretty unmistakable. I believe the good Professor when he says he could recall no prior inspiration for the word, but: Bag-End. That's the Hobyah's whole thing, the bag. That and turnips.

I ended up writing about the EO Personal Communicator a few years ago:

https://www.inverse.com/input/features/fax-on-the-beach-the-...

Was never able to get the device working (I have two) but I still have them in my possession. So I have some of the few Hobbit microprocessors in the wild.

I had one of these and I think I bought it from Staples, Office Depot, or whatever the predecessor to this was. I can't remember the exact timing but I picked up an IBM Simon around the same time.

Both were half baked products that ended up in a drawer after a couple of weeks. The EO had lousy battery life and not as good as pen and paper for note taking which is why I wanted it.

I had various devices over that time frame which included the Palm Pilot, Sony Magic Link, all of the Newtons, a Sharp Zaurus, and so on. The one that really stuck was the Psion MX5 as I could actually touch type on it.

The person who sent along the EOs also sent a Simon. I couldn't get that working either. Sigh.

If it's any consolation you're really not missing much!

I’m always surprised that when people talk about stack computers they don’t mention the Burroughs B55/65/67/7700 and A Series computers. They ran Algol (pre C) super fast.

fastish - our B6700 had roughly microsecond core and as a result was ~1MIP

> So Larry Tesler, now in charge of the Newton project, orchestrated a switch to the ARM architecture in 1990.

ARM seems to have worked out pretty well.

> The CRISP architecture was described as a "2½ address memory-to-memory machine", where instructions can employ zero, one, or two memory addresses and can employ a stack entry called the accumulator for computation results.

Ewwww

This sounds like something from the 70s and not like a new architecture for a mid-1990s microprocessor - and indeed the 70s seems to be where the design actually came from.

The instruction encoding has strong S/360 feeling to it. Which is not necessarily a bad thing in itself.

I have an EO 440 in my house. The disk is slightly corrupt and some of the apps (including the configuration app) don't work, unfortunately.

Reminds me of Transmeta Crusoe. Innovative but insufficiently performant.

OTOH, LISP stack-based machines were fairly successful in their day.

> Reminds me of Transmeta Crusoe. Innovative but insufficiently performant.

According to this article (circa 2003?), Crusoe was slower but more power efficient than Pentium-M. It suffered from a recession in Japan (affecting Crusoe-based laptops), poor execution and faulty chips leading to product delays, and being "outfoxed" by intel:

https://www.cnet.com/tech/tech-industry/transmeta-are-the-ch...

An interesting bit is how intel was apparently ahead of TSMC at that time.

It's a shame that AMD is the only x86-compatible CPU company that survived competing with intel, but we're lucky that they did. Some Transmeta alumni also went on to Nvidia.

The following progression is a clear downward trajectory: transitor, C/Unix, this thing.

> In addition, the Hobbit was buggy and, as the Hobbit was exclusive to Apple, AT&T wanted millions more dollars from Apple to continue development for the Newton. So Apple looked at the ARM architecture instead. Benchmarks showed the ARM design outperforming the Hobbit

So maybe TSMC shall learn something from history and not rely only on Apple for some processes.

I'm pretty sure that Apple's internal silicon group rolled their own version, as I recall it came back after they decided to switch to ARM and worked perfectly.

Like a lot of Apple projects of the era someone put their heart and soul and a year or more of their life into a project and it got arbitrarily canned - we hired him ....

Apple did so many of these reinvent-the-world projects in the systems area alone that somewhere on Earth-87 they’re running their Hobbit-based phones on NuKernel and the apps are all written in Dylan.

I don't believe there was an Apple "silicone" group at that time. Thats a recent invention.

Apple definitely dabbled in silicon around that time, the headliner was the Aquarius RISC CPU project. They famously bought a Cray for it.

Before they decided on PowerPC, they had a failed attempt at building their own RISC CPU:

https://thechipletter.substack.com/p/the-first-apple-silicon...

https://archive.org/details/scorpius_architecture

not "silicone" that's something else.

They were rolling their own north-bridge-ish chips at the same time for 2FX and later machines (all pre the power PC switch)

The first true ARM MMU design was Bob Welland's, who had joined Apple after his stint at Commodore (C900, Amiga 500). It doesn't look like traditional MMUs, because it needed to support the Newton OS and shared address spaces.

I stand corrected :)

They were really really bad at actually shipping products for a long time. I love all that ATG stuff: SK8, Dylan, all that blue-eyed object oriented optimism. But hundreds of man-years were put into all of that and Copland and Taligent, and they had nothing to show for it. Apple was so dysfunctional under Sculley and Spindler. People say Apple killed Hypercard but really what happened is people worked on it for years but it got bogged down in feature creep and focus switches, and they never get a functional version 3 out the door.

It is impossible to know what you mean by this sentence.

Surely it's Apple who rely upon TSMC for 'some processes'?

Reply on news.ycombinator.com