I wonder if we should look into making a unikernel build of this so people can boot up their own dedicated mining ASIC ... :P

<g​ingeropolous:monero.social> u think we could squeeze more h/s/w from a rig?

maybe a tiny bit. mostly because there would be no other code running besides the kernel and the miner. no other processes on the box.

we could make things totally turnkey. add a broadcast listener to p2pool, have the miner broadcast to find its p2pool server. just plug it in and it works.

like good old game servers :D

distribute it on bootable USB flash or microSD cards.

it can even be p2pool agnostic, the broadcast listener can just be something you run on the network to send config

sends stratum url and args to queriers

target can be anything

it could ... but there's no reason to support anything besides p2pool ;)

sweep fees might grow considerably past FCMP++ btw, so I'm trying to get some aggregation ongoing via multisig groups on the fly (with fallback txs ahead of time)

<g​ingeropolous:monero.social> id want to support local daemon mining. i don't use p2pool because of the dust. would rather just accept the variance and have larger outputs

fair enough. monerod itself could answer that broadcast query

tbh, mining on main gets you way less "dust" as payouts unless you have high hashrate

compared to mini/nano etc.

we should write an IETF RFC spec for this Dynamic Miner Configurartion Protocol :P

just do UDP broadcast on port 3333

the spec part is defining the message format

I am suggesting the port

let's maybe first find if this "turnkey" solution is that much efficient :P

sure. the first attempt can use a hardcoded destination.

<g​ingeropolous:monero.social> nonsense. cart -> horse

LOL

at least we have the road defined

<g​ingeropolous:monero.social> well it could have value for ppl launching massive mining infra. i guess they could just clone images onto drives as is, or pixy boot or whatever

sure. but people are doing MSR tweaks and other such stuff for tiney fraction of percent improvements

maybe there's a fraction of a percent to be found here too

hyc so basically ragerx for p2pool :D

the one big problem with that is the lack of drivers

no driver for GPU = GPU will be burning energy instead of going to the lowest energy state

So need to package at least the latest AMD/NVIDIA drivers to put the GPU to idle

hyc I'm still waiting for the review tevador/RandomX #316

RandomX v2 is almost done, only more documentation and PPC64 (big endian) intrinsics are left

👏

<g​ingeropolous:monero.social> the ipc from zen2 to zen4 is no joke. my monerosim run goes from 9 minutes to 3 minutes.

Did a writeup on RandomX v2 changes: github.com/SChernykh/RandomX/blob/v2/doc/design_v2.md

might indeed be good to get more datapoints there to show the impact on quoted Ryzen 9 3950X

Need 3950X test, but I don't have it

I took the numbers from XMRig's benchmark page

<g​ingeropolous:monero.social> i got one.. where are these tests...

<g​ingeropolous:monero.social> i got em. tryna compile

<g​ingeropolous:monero.social> oh jesbut. this box is currently a space heater in a far away room thats on a bullshit wifi extender so apt upgrade is gonna be slow

You will need to compile the v2 branch from my repo, then run randomx-benchmark with the same command line I used for 9950X. But first run randomx_boost.sh from XMRig scripts, and enable huge pages

oh, and the command lines are here gist.github.com/SChernykh/6058ecf01c929883b9d19c7eeadc8809

<g​ingeropolous:monero.social> of course v2 builds but not v1: paste.centos.org/view/1ad72230

import via "#include <cstdint.h>" afaik

what do you mean not v1? The latest code has everything, you can switch between v1 and v2 with command line

old code might have some unfixed issues of course

sech1: there is ppc64be JIT now?

No

But the interpreter mode might not work if the new intrinsics are wrong

hyc back in cnv0 days I used to have a stripped down kernel with bare minimums needed for storage+ethernet, initrd just brought up network, allocated hugepages, started my custom static xmr-stak build (built w/ icc, hot funcs had runtime dispatch for various uarch from westmere - skylake)

nice easy boot and mine setup

would be pretty fun to do the same for randomx

sech1 I have access to ppc64 host, let me see if I can build and run tests