my prior post in “new commers” shows my setup (ai & gaming $4 84gb vram) , but i am running Q8 only on these shards atm. 1x5060/8gb, 2x 3060/12gb, 1x1660s/6gb, 1x1070/8gb with 200sec response for hard questions (according to opus5 max). 550w usage. i have a renogy 2000w inverter, 2x 220ah 24v litime batts, 1400w solar panels (1100w actual). i also run 7 proxmox VM’s…6 websites, RAG/EMBED, grafana, other small programs and a trunas server (10 spinners and 4 ssd’s) spinners are idle most of the time atm due to not using jellyfin to much hence the 550w.
source & further reading
forum.level1techs.com — original article
Keychron Firmware 1.1.1 Nkey Rollover fix for KVM Hotkeys
Sanity check on my math: DGX Spark vs 2x Radeon PRO R9700 vs Mac Studio (M3 Ultra) for multi-user office LLM serving, does reality match the numbers?
Does anyone know how the vericoding benchmark ran GLM-4.5 and DeepSeek V3.1? Asking because one of them looks like the non-thinking variant