Bitcoin Institute

delete

1 messages BitcoinTalk ArtForz September 29, 2011 — September 29, 2011
ArtForz September 29, 2011 04:54 UTC Link to this message
Quote from: Lupus_Yonderboy on September 29, 2011, 3:49:10 AM UTC

The algorithm was written by an fpga expert and it runs very well on fpgas (shockingly) and not so well on gpus.

Hey, that’s news to me, so far my math and synthesis attempts say it’ll suck on FPGAs, too. (wow, a algorithm designed to be expensive on specialized hardware ends up expensive on specialized hardware). so… [citation needed] or stop spreading FUD.

Hmm and here I thought you had the necessary skills to efficiently program fpgas to solve blocks using an algorithm you wrote. My apologies. There are a handful of people on this board with the knowledge and equipment to try and implement this on FPGAs. I am not one of them and therefore must concede the point. The rest of what I wrote is true.

Wait, what? an algorithm I designed?

Rule #1 of cryptography: do not design your own algorithms.

What I did was modify multicoin to make replacing the block PoW hashing function easier, then plugged in scrypt (http://www.tarsnap.com/scrypt.html) with parameters of N=1024, p=1, r=1, feeding in the block header as password and salt, output size of 32 bytes.

While those parameters would be way too low for a good password hashing/key derivation function (you want lots of margin for the future there), my initial educated guess and further experiments suggest they’re still enough to “pessimize” current GPUs and FPGAs to a point where CPUs will easily be competitive… GPUs growing several MB of fast random access on-chip memory in the future might change that.

And yes, choosing such “unusual” parameters is skirting the rule, but in this case imo acceptable risk. Worst case… someone manages to make a “efficient enough” GPU/FPGA/… implementation or a new gen of GPUs comes out, scheduled chain fork switching to higher N and p. Up to N=4096,p=8,r=1 or so time to verify the PoW hash on a CPU shouldn’t be an issue, beyond that you’d have to add some measures to prevent “junk block spam” DoS.