mimblewimble team mailing list archive

[Ignotus Peverell <igno.peverell@xxxxxxxxxxxxxx>]

dear John,

Thanks for the update. Just curious, what's the time tradeoff for that 20x memory? This optimization becomes difficult, especially on GPU, over Cuckoo 30, correct?

- Igno

-------- Original Message --------
Subject: Re: [Mimblewimble] Status and how to help
Local Time: April 18, 2017 11:51 AM
UTC Time: April 18, 2017 6:51 PM
From: john.tromp@xxxxxxxxx
To: Ignotus Peverell <igno.peverell@xxxxxxxxxxxxxx>
mimblewimble@xxxxxxxxxxxxxxxxxxx <mimblewimble@xxxxxxxxxxxxxxxxxxx>

dear Ignotus/all,

> Some decent mining software that can handle multiple GPUs.

There has been some significatn development in Cuckoo Cycle.

Xenoncat, the anonymous Zcash miner contest winner, recently
claimed a $5000 CPU speedup bounty, by using 20x more memory
to bucketsort the edge->endpoint map and thereby avoid the high latency
random memory accesses.

Basically, with enough memory available, Cuckoo Cycle is transformed
from a latency bound PoW into a bandwidth bound one.

I'm in the process of writing my own version of his solver, after which
I will port that to CUDA. Then the next challenge will be to extract the
maximum performance out of the GPU solver. An OpenCL solver will
have to come later...

regards,
-John