4 ms·

Chuck Moore (invented Forth) announced new CPU: 144 cores, 100 billion ops/sec

60 points by csmeder 17y ago

csmeder 17y ago
Sorry about the title, I couldn't fit in who Moore is and have it make sense. I wanted to say: "Chuck Moore, the inventor of Forth, announced GreenArrays new CPU: 144 cores, capable of 100 billion ops/sec"
- pvg 17y ago
  Forth inventor Chuck Moore announces 144 core CPU capable of 100 billion ops/sec 80 chars on the nose.
joshu 17y ago
Anyone ever worked with one of his cpus?
- csmeder 17y ago
  The only commercial company that I know of, but I don't know much, who has explored using one of his CPUs is BMW. . If you want to try it out you can use their emulator: http://greenarraychips.com/home/documents/greg/cf-intro.htm http://greenarraychips.com/home/documents/greg/cf-intro.htm softsim is an instruction level simulator for GreenArrays chips, using code you compile for ROM and/or RAM, and with provision for adding testbeds to simulate I/O.extend softsim to allow you to define and simulate external peripherals to interact with using I/O pins.
patrickgzill 17y ago
I really, really want to like these guys - I have been following Forth, Charles Moore, and colorforth since first reading up on Forth in 1996 or so. However, the site is a dog's breakfast, and despite clicking around I cannot seem to locate an actual binary to download and run colorforth on Windows. Where / when can I buy an eval board? How do I connect it to ethernet, or USB? Further, it seems to be a company where you have a lot of smart guys who are retired, and they are just putzing around without a burning desire to really take over the market they are targeting. BTW it seems each core only has a very limited amount of RAM on it, 64 "words" which means 32 bits on PCs and 18 bits (???) on the g18 core CPUs.
- joe_the_user 17y ago
  I think that colorForth is the simulator. I just downloaded it from: http://greenarraychips.com/home/documents/greg/code/af-34k2-ga144-1-10-PD.zip http://greenarraychips.com/home/documents/greg/code/af-34k2-... It's only two clicks from the main page. I don't know if I'll do anything with this or not.
blue1 17y ago
Very naïf question from a software-only guy: given that such a chip exists, how is one supposed to USE this kind of thing? Like, maybe plugging some kind of card on a PC? Or is it just interesting for hardware manifacturers?
- ableal 17y ago
  Given the chip, you can't get to the "software only" part until you have: - The chip mounted on a board with at least memory and communication. - Software on your machine to talk to the board and hear back from it. In general, the most rudimentary development kit possible is to have the chip on a board with a ROM socket, and some development setup that allows you to burn code to a ROM. Then you plug the ROM in the board, turn the power on, and see if it works ...
rbanffy 17y ago
Wow. As an undergrad (equivalent - I live in Brazil), the best project I did (the one I am most proud of) was a stack-based CPU that ran something very close to Forth. This is really, really cool.
ajkirwin 17y ago
100 billion instructions per second is what.. fast than the current Intel i7s? Am I reading this right?
- Retric 17y ago
  I think it's about 10% as fast as a high end GPU, and around 2x faster than a CPU. Edit: That's flops not instructions, it could be faster than that if it's doing multiple flops per instruction. Also GPU's have significant limitations so it's not really possible to get those speeds on a generic workload.
  - blue1 17y ago
    If those are the numbers, it seems to me that -while certainly an interesting concept- it has no future in practice, unless it is extremely cheap.
  - jules 17y ago
    Where did you get that information? This hardware cannot do 1 flop in 1 instruction. Far from it, actually.
    - Retric 17y ago
      Never mind. I was thinking of a diffrent chip then.
bbot 17y ago
Some basic spec comparisons: A Intel i7 900 series processor has a die size of 296 mm^2, with a process size of 45 nm. The GA144 has a die size of 21 mm^2, at a process size of 130 nm, which is a couple generations old. If the i7 900 series was manufactured at 130 nm, it'd be 2488 mm^2. So, it's smaller. A lot smaller. I'd compare MIPS too, but that would be worse than useless, since it implies that it actually means something, which is far from the truth. Comparing raw instructions per second values between different architectures means nothing. (Plus, Intel have passworded their datasheets!) But it really doesn't look good. The Greenarray guys might manage some deep magic that makes their processor twice as fast per transistor, but it's still 118 times smaller. There's just not enough silicon!

60 points by csmeder 17y ago

Sorry about the title, I couldn't fit in who Moore is and have it make sense. I wanted to say: "Chuck Moore, the inventor of Forth, announced GreenArrays new CPU: 144 cores, capable of 100 billion ops/sec"

pvg 17y ago
Forth inventor Chuck Moore announces 144 core CPU capable of 100 billion ops/sec 80 chars on the nose.

joshu 17y ago

Anyone ever worked with one of his cpus?

csmeder 17y ago
The only commercial company that I know of, but I don't know much, who has explored using one of his CPUs is BMW. . If you want to try it out you can use their emulator: http://greenarraychips.com/home/documents/greg/cf-intro.htm http://greenarraychips.com/home/documents/greg/cf-intro.htm softsim is an instruction level simulator for GreenArrays chips, using code you compile for ROM and/or RAM, and with provision for adding testbeds to simulate I/O.extend softsim to allow you to define and simulate external peripherals to interact with using I/O pins.

patrickgzill 17y ago

I really, really want to like these guys - I have been following Forth, Charles Moore, and colorforth since first reading up on Forth in 1996 or so. However, the site is a dog's breakfast, and despite clicking around I cannot seem to locate an actual binary to download and run colorforth on Windows. Where / when can I buy an eval board? How do I connect it to ethernet, or USB? Further, it seems to be a company where you have a lot of smart guys who are retired, and they are just putzing around without a burning desire to really take over the market they are targeting. BTW it seems each core only has a very limited amount of RAM on it, 64 "words" which means 32 bits on PCs and 18 bits (???) on the g18 core CPUs.

joe_the_user 17y ago
I think that colorForth is the simulator. I just downloaded it from: http://greenarraychips.com/home/documents/greg/code/af-34k2-ga144-1-10-PD.zip http://greenarraychips.com/home/documents/greg/code/af-34k2-... It's only two clicks from the main page. I don't know if I'll do anything with this or not.

blue1 17y ago

Very naïf question from a software-only guy: given that such a chip exists, how is one supposed to USE this kind of thing? Like, maybe plugging some kind of card on a PC? Or is it just interesting for hardware manifacturers?

ableal 17y ago
Given the chip, you can't get to the "software only" part until you have: - The chip mounted on a board with at least memory and communication. - Software on your machine to talk to the board and hear back from it. In general, the most rudimentary development kit possible is to have the chip on a board with a ROM socket, and some development setup that allows you to burn code to a ROM. Then you plug the ROM in the board, turn the power on, and see if it works ...

rbanffy 17y ago

Wow. As an undergrad (equivalent - I live in Brazil), the best project I did (the one I am most proud of) was a stack-based CPU that ran something very close to Forth. This is really, really cool.

ajkirwin 17y ago

100 billion instructions per second is what.. fast than the current Intel i7s? Am I reading this right?

Retric 17y ago
I think it's about 10% as fast as a high end GPU, and around 2x faster than a CPU. Edit: That's flops not instructions, it could be faster than that if it's doing multiple flops per instruction. Also GPU's have significant limitations so it's not really possible to get those speeds on a generic workload.
- blue1 17y ago
  If those are the numbers, it seems to me that -while certainly an interesting concept- it has no future in practice, unless it is extremely cheap.
- jules 17y ago
  Where did you get that information? This hardware cannot do 1 flop in 1 instruction. Far from it, actually.
  - Retric 17y ago
    Never mind. I was thinking of a diffrent chip then.

bbot 17y ago

Some basic spec comparisons: A Intel i7 900 series processor has a die size of 296 mm^2, with a process size of 45 nm. The GA144 has a die size of 21 mm^2, at a process size of 130 nm, which is a couple generations old. If the i7 900 series was manufactured at 130 nm, it'd be 2488 mm^2. So, it's smaller. A lot smaller. I'd compare MIPS too, but that would be worse than useless, since it implies that it actually means something, which is far from the truth. Comparing raw instructions per second values between different architectures means nothing. (Plus, Intel have passworded their datasheets!) But it really doesn't look good. The Greenarray guys might manage some deep magic that makes their processor twice as fast per transistor, but it's still 118 times smaller. There's just not enough silicon!