Is this possible?

[Qythyx]

Do you think that Transmeta could program their Crusoe chip to crack RC5 in hardware. For example, could they program a special instruction (or set of instructions) that would greatly increase the speed of cracking RC5 under their chip.

If so, could they also somehow put this instruction into the Intel compatible version such that it just works seemlessly or would the client need to be compiled for it?

Just wondering. It's not like it's going to happen anyway.

 

[Train]

the RC5 clients are written in low level languages (most likely C++ or even assembly) which are already great number crunchers. And youll notice they have a different client for each type of chip, so they have done a good job of optimizing it for each instruction set.

A custom instruction set for RC5? interesting Idea, im sure it could be done, but i dont see more than a 1 or 2% increase in cracking speed. But the cost for a custom chip would be enough to buy at least 3 or 4 retail chips, so its really not worth it.

 

[SocrPlyr]

hey just think about the alti vec whatever instructions in the G4 same thing you are talking about

 

[Train]

actaully, now that i think about it, you could probably make a custom chip go pretty fast on RC5.

Cracking keys is really just looping from a start number to an end number, testing each key, hardly any memory is needed. But most chips put a lot of effort into addressing memory (some chips can address a VERY LARGE amount of RAM) so if you made the actual memory the chip had to address very small, and focused on speeding up the registers and the Floating Point/Integer units, you could probably get a nice jump in performance for a task like RC5, but trying to run windows or something like that on that chip would be a disaster because of its limited memory. If made, youd probably want a special, very light weight OS to run the RC5 client on, or just have the RC5 client be the OS itself.

 

[SocrPlyr]

any instruction that increases integer performance will help out rc5 (just as long as the client is setup to use it)
rc5's performance is based solely on integer performance that is why for each mhz the speed goes up linear
(the client etc fits within the L1 cache so it is all integer performance and there are no FPU calculations either)

 

[vss1980]

This is going outside of what D.net's RC5 is meant for. Custom hardware has been used for cracking RC5 before. But seriously, I wouldn't use a Crusoe or G4 processor for a serious dedicated hardware cracker due to the fact that they are general purpose chips. Dedicated hardware would mainly have just the needed instructions and probably use another computer for control and to retrieve the blocks from.