libera #monero-pow

8 Oct 2023

Monero's Proof of Work algorithm (RandomX) and related topics. Use #p2pool-log and #p2pool-mini for P2Pool discussions. Anti-spam active: PM hyc for voice.

• 11:50
  m-relay
  <4​rkal:matrix.org> Are there any good guides on how to do that for a beginner?
• 15:22
  sech1
  Interesting twitter.com/WuBlockchain/status/1710880716383494207
• 17:10
  elucidator
  sech1: x5 killed bitmain?
• 17:12
  sech1
  the last straw
• 17:13
  elucidator
  iHope
• 17:18
  hyc
  brilliant
• 17:19
  hyc
  I wonder why they didn't just jack up the price of the X5 etc since they sold all they had
• 17:59
  m-relay
  <p​olar9669:matrix.org> Who would have bought it at high cost with a threat of pow change 😂
• 17:59
  m-relay
  <p​olar9669:matrix.org> Siphoned off all those secretly mined coins
• 18:28
  pauliouk
  well if they're not paying their staff, definitely not going to be any spare parts getting shipped
• 21:15
  tevador
  tevador/RandomX #275
• 21:15
  tevador
  took a lot more work than expected
• 21:17
  pauliouk
  hmm time to look into buying some RISC-V hardware :D
• 21:41
  sech1
  nice
• 21:41
  sech1
  I'll take a thorough look tomorrow
• 21:41
  sech1
  But tests (including risc-v) already pass, that's a good sign
• 21:44
  tevador
  I spent a day and half hunting for a bug that caused 1 in 50k hashes to fail. Those are fun.
• 21:45
  sech1
  what was it?
• 21:46
  tevador
  a bug in immediate materialization that manifested only for values 0x7ffff***
• 21:47
  sech1
  ah, probably something specific to instruction encodings
• 21:47
  sech1
  nasty
• 21:49
  tevador
  it's a different beast than x86 and ARM
• 21:52
  DataHoarder
  sech1: that makes more sense, I saw no reason why they would just sell these *now* except if they had a money crunch
• 21:54
  sech1
  tevador yeah, I remember writing jit for AMD GPUs, it was a different beast too. Spent like 3 weeks full time on it. The worst can of worms was this function: github.com/SChernykh/RandomX_OpenCL…ndomX_OpenCL/CL/randomx_vm.cl#L1406
• 21:56
  tevador
  This was another fun afternoon trying to figure out why GDB won't work: github.com/tevador/RandomX/blob/pr-…src/jit_compiler_rv64_static.S#L194
• 21:58
  tevador
  OpenCL not having directed rounding is a massive fail
• 21:58
  sech1
  I'd still use x4 for another imul_rcp value
• 21:59
  tevador
  not worth the trouble
• 21:59
  sech1
  make it compile time option to use x4, if you ever need the debugger
• 21:59
  sech1
  I'll have to do it for xmrig anyway :D
• 22:00
  sech1
  Right now you store only 10 rcp values in registers, 11 is better :D
• 22:00
  tevador
  yeah, I had a compile time flag for that but then I just remapped everything to avoid x4