Mining software concept: surf mode. Imagine you generate a ton of pending work units in memory. Each of them needs specific data from specific snow field segments.
You read through the snowfield from block 0 till to end. Just one big sequential read. You use that data to advance the work units as you read the data they need.
If you are lucky, then the next data they need is still coming and they are riding the wave, advancing along ahead of or with your read wave.
If they need something behind the wave, then they are off the wave and either get abandoned or wait for the next wave.
If your SSD can read at 3GB/s (fairly reasonable) then you can do a complete wave in 43 seconds.
You can do 6 waves (the most you'll ever need for all work units to complete) in 256 seconds.
This would be pushing it, but is below the block time.
and you can also try things like you only both with work units that make it to completion in 2 or 3 waves, discarding a huge number but still keeping a lot
is that just betting very wide on untangling the linked list onto a hilbert curve?
I don't know enough about hilbert curves to answer that. Those are the square space filling things, right?
yeah, ip address space wiggly snakes
i’ve been thinking of something similar, your random is too good, so just shuffling should always make seeks easier?
if your queue is deep enough, you can kinda simulate memory mining
don't even need the wave to be in order
just load the chunk with the longest current queue of work
i’ve been thinking of something like that for distributed mining across the internet
the bandwidth probably gets brutal
does it matter if some participants are slow?
glacial mining, melting the fields
not really
there are always more work units that could be started
if a particular queue is slow or dead, that is fine
exactly
so i guess people can participate if they have even one chunk
sure
https://pcpartpicker.com/list/2Wb8mq - current plan for a cheap 50kh/s miner (up to field 8)
missing a psu and a case
cases are for suckers
but yeah
mostly a combo deal could be cheap
With those things, probably at least $300, so $6/kh
which seems like a lot
prolly a bit less, but yeah
but those intel 760p seem pretty damn great for the price
how about you add a divider to your index for ’works up to n fields into the future’
assuming they can do what the specs say
they seem pretty legit and not thermally constrained either
intel is a boring enough company
right
My old intel 1.2TB PCIE nvme still kicks ass
450k/s random reads
that’s some 900 series workstation card?
I bought it in 2015
i remember people getting those for adobe scratch volumes
I got it for bitcoin/electrum servers
i ran mine on spinny disks, xfs and enough ram
@Clueless what was that m.2 pcie riser you liked?
that’s the asus one
i’m morbidly curious as to how 760p would do on VROC on that
@Fireduck I'm totally ready to jump in mining, budget of 1-2 BTC
ctrl-f ’jump first’ http://telefinn.blogspot.com/2011/11/matti-nykanen-quotes.html In in recent weeks, I have made references on a couple of occasions to the legendary Finnish ski-jumper and bon-viveur Matti Nykänen . In pa...
"If things are going well, then call me. I can fuck it up In seven seconds."
one can now get epyc on amazon
Me too thinking of mining but maybe I'm too late at 128gb ram i don't have any machines/facility have to start from zero though if i mine it'll help the network get more decentralized as i think maybe nobody from where I'm at mine this yet
128GB gets you quite far at this point still
@Fireduck there's your upcoming unicorn tears again, 4TB per socket EPYCs sometime in 2019 https://www.anandtech.com/print/13561/amd-previews-epyc-rome-processor-up-to-64-zen-2-cores AMD Previews EPYC ‘Rome’ Processor: Up to 64 Zen 2 Cores