Conversation

Notices

1. Embed this notice
   on-lain ✔ᵛᵉʳᶦᶠᶦᵉᵈ (lain@lain.com)'s status on Wednesday, 01-May-2024 04:54:22 JST on-lain ✔ᵛᵉʳᶦᶠᶦᵉᵈ

   • :suya:
   • Phantasm
   • Johnny Peligro
   • e(Ag)le 🦅
   @newt @SilverEagle @phnt @mischievoustomato there’s no need to dedicate, it’s dynamic. I run largel models like llama3 70b on a Mac. Macs are currently the cheapest way to run huge llms at acceptable speeds.
   • Embed this notice
     on-lain ✔ᵛᵉʳᶦᶠᶦᵉᵈ (lain@lain.com)'s status on Wednesday, 01-May-2024 19:00:35 JST on-lain ✔ᵛᵉʳᶦᶠᶦᵉᵈ

     in reply to

     • :suya:
     • :marseyloadingneon: m0xEE :marseyloading:
     • Phantasm
     • Johnny Peligro
     • e(Ag)le 🦅
     @m0xEE @SilverEagle @phnt @mischievoustomato @newt it gets close, m2 ultra has 800gb/s, rtx4090 has 1008gb/s. the dedicated gpus are still a lot faster, but again, you can't 190gb of vram at any reasonable price with nvidia hardware.
   • Embed this notice
     :marseyloadingneon: m0xEE :marseyloading: (m0xee@breloma.m0xee.net)'s status on Wednesday, 01-May-2024 19:00:59 JST :marseyloadingneon: m0xEE :marseyloading:

     in reply to

     • :suya:
     • Phantasm
     • Johnny Peligro
     • e(Ag)le 🦅
     @lain @SilverEagle @phnt @mischievoustomato @newt
     Isn't this RAM considerably slower than the one discrete GPUs have? 🤔