Javelin Internals

Justin Shumaker

If the Ubicom SX48AC is operating at 25MHz on the javeline stamp and has an average instruction length of 3 cycles, that makes the javeline roughly 1/1000 th the speed of native code on the Ubicom SX48AC. I recognize that there is an IC for doing bytecode translation to the native Ubicom SX48AC instructions and that there may also be some buffering involved assuming there is a pipeline in the architecture. I would think that this whole process would consume somewhere on the order of 20 to 50 cycles. Could somone explain why it takes roughly 1,000 cycles for a single java bytecode operation to process? I'm wondering what sort of optimizations could be done in a future version to make this more efficient. Also, why was the Ubicom SX48AC selected to run at 25MHz and not 50MHz?

Post Edited (Justin Shumaker) : 1/21/2005 5:11:16 AM GMT

Jon Williams

Engineering -- as with product selection -- involves choices. Design choices cause the Javelin to be what it is. I can speak specifically to speed; the original Javelin module (not for sale) did run at 50 MHz, and if memory serves me consumed about 120 mA while running. We felt like this would not fly. The only choice we had to reduce current consumption was to reduce the clock speed of the SX to bring it back to a manageable level.

▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Jon Williams
Applications Engineer, Parallax
Dallas, TX· USA

Justin Shumaker

Is the oscillator on the PCB next to the SX ? I'm wondering if I can take it up to 50Mhz myself... ideas?

Jon Williams

If you change the oscillator you will do two things:

1) Void your warranty
2) Upset the operation of the Javelin (everything is designed and timed for 25 MHz), causing it not to work

▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Jon Williams
Applications Engineer, Parallax
Dallas, TX· USA