Shop OBEX P1 Docs P2 Docs Learn Events
Chinese supercomputer named as world's fastest - Page 2 — Parallax Forums

Chinese supercomputer named as world's fastest

2»

Comments

  • HumanoidoHumanoido Posts: 5,770
    edited 2013-07-01 22:36
    The GA144 chip confounds me. You have a grid of 144 processors, but rather than 12 x12, the device is long and narrow in a 8x16 grid. The processors on the edge are consumed in part by special i/o functions.... somewhat limit i/o for all the power. And there is no clock. As I haven't yet gotten one of the two chips I bought to actually interface. But in theory, an RS232 boot is possible. It really needs the outboard RAM and storage provided in the Demonstration package to be off any use to anyone. And in all honest, I strongly suspect the interior processors are gridlocked.. unless there is some brilliant concept that I have missed. The best thing I have learned is that the Propeller i/o scheme with every processor having the ability to reach the outside world or to share data around the hub is a lot more pleasant to work with than bumping data from one processor to the next just to get a startup configuration. In sum... the GA144 is weird and driving me nuts.... very little i/o for 144 processors. I guess there is a number crunching or a video handling aspect that I just don't know. It seems a lot of CPU time is spent in passing left, right, up, or down. Thank God the Propeller doesn't do that. You can just set aside i/o pins to use alone, or to share.

    I appreciate your sharing these findings and admire your bold willingness to explore a strange new world with GA. I wanted to get 100 of these chips to add 14,400 processors to the Propeller Powered Big Brain. I changed my mind when I discovered some GA144 processors in the chip were not directly accessible due to its unique architecture and that I could not add processors by restructuring the way I could with the Propeller Cogs. The idea looked less attractive for my application and GA was put on hold. There certainly could be other methods useful to this chip but with the relatively limited support, that could take some time to develop. I think the Propeller is more understandable with its layout, SPIN and other language choices, and each Cog is directly addressable and it's relatively easy to set up a parallel machine with multiple props, with many pins, and the Propeller chip has many prospects for enhancements. Sorry, I wasn't trying to make this into Prop spam.
Sign In or Register to comment.