The crux of most supercomputers is the ability to operate on many pieces of data at once — something video cards are good at, too. Enter T1 (short for Torrent-1), a RISC-V vector inspired by the Cray X1 vector machine.
T1 has support for features, including lanes and chaining. The chip contains a version of the Rocket Core for scalar operations, but there’s no official support for using it. The project claims you could easily replace that core with any other RISC-V CPU IP.
By focusing on parallelism instead of out of order execution, the design gets to skip branch prediction, register renaming, and similar problems.
There is an emulator if you want to experiment. You can even grab a docker image for easy installation. This doesn’t look like something you could pick up in an hour, so prepare to spend some time. Everything is bare-metal, too, so leave your favorite development tools at home.
The project uses Chisel, which we’ve covered before. The build system seems very complex, but based on Nix Flakes, so it should be understandable.
If your high-performance RISC-V dreams are more conventional, there’s work going on in that area, too.
Title graphic from Freepik.
“Everything is bare-metal, too, so leave your favorite development tools at home.”
Espressif ESP chips were a P.I.T.A. until the core for Arduino was written. Sure, big shops could afford the expensive tool$, but the hobbyists really craved a simple development environment. Same can be said of Raspberry’s RP-2040.
I learned 6502 from hand-coded assembler, but these new chips are simply not worth the effort IMO.
Okay, so you didn’t like the ESP chips until there was an IDE for them, but you learned 6502 from hand-coded assembler. I’m not sure which way you are arguing: that IDEs are overrated, since you were okay with not having one for the 6502, or that somehow, having to invoke the assembler at the command line is now too much of a bother. I’ve personally never seen any benefit to using an IDE, and write my own Makefile for almost everything, so it’s not obvious from your comment.
I’ve always just used an editor like notepad++ or geany myself . And envoke compilers from the command line. Only ide that I really liked was Borland/Imprise Pascal/Delphi or the c++ builder back when. I too use a simple makefile most of the time for c/c++ projects. Starting to play around with cmake as that is what our energy management system source code build at work uses. It does some nice things too.
I think you missed the point. Neither of those was their argument.
The argument was about the complexity of the chip; they don’t like the ESP chip because it’s complex enough to require an IDE to work with effectively, while the 6502, even when programmed in assembly, was easier to work with.
Exactly! I learned Z80 assembly and was able to understand most of the ZX-Spectrum’s ROM. You pretty much wrote hardware or memory directly for I/O.
To be fair, back in the dark ages, if you had a 6502 or a Z80, (a) you were younger and more mentally agile then, (b) what you had was what you had, with, in most cases, not much chance of having any other hardware to play with any time soon, and (c) you didn’t have the web to doom-scroll on or other similar distractions.
Simpler times.
Also, to be scrupulously fair, the T1 has this in spades. Because, as of now, it’s not a real chip you can hold, and how many changes will there be before it is?
The Apple II was sold for, what, 6 years? And, probably for Ray, learning it was a gateway drug. He’s onto harder stuff now, and something that’s only a flash-in-the-pan for a few months just isn’t going to cut it.
The Commodore 64 was sold for 12 years, and yeah, 6502 didn’t take that long to learn.
“The Apple II was sold for, what, 6 years?” – 18 years for the entire 2 series
“if you had a 6502 or a Z80, (a) you were younger and more mentally agile then, (b) what you had was what you had”
Although that certainly comes into play, complexity still takes a big toll.
The 6502 and z80 were in a few computers I had as a kid and basic+ assembly was virtually a requirement.
Yet the 8086 at the same age felt far more complex and not worth learning asm for, despite seeing its popularity jump up and stay there (always in hindsight eh?)
I think these are two separate things that both narrow down over time.
Yep, I quit smoking while working for Jef Raskin (he didn’t like the smells carried on clothes) and my focus and concentration went to hell, at least in comparison to before. The geriatric effects can bbbbbbbbbbbbbbbbbbbbbb
Well, you learned a very different and difficult “IDE”, Make.
If you like this project then check out “Chipyard” on github. It’s a Chisel frame work that makes it easier to build cool and wonky experimental RISC-V SoCs.
As I side note it almost seems like VHDL and Verilog are becoming intermediate languages as projects like Chisel, MiGen, Amaranth, and SpinalHDL take shape.
Most serious HDL work is still done in VHDL or Verilog, I mainly only see open source or hobbyist projects using the other languages.
still waiting for 4096 epiphany cpu