#benchmarks

2 articles

On-Prem AI 30 August 2026 19 min read

How I make AI videos with my own GPUs: 3090 vs 4090, first results

A fully self-hosted pipeline: script to voiceover to render, no cloud in the path. Two mascots, worst-to-best, the 3090/4090/DGX numbers, and why the right model depends on your character.

#video-generation
#wan
#comfyui
#rtx-4090
#rtx-3090
#benchmarks
#open-source
#mascot

On-Prem AI 23 August 2026 17 min read

GPU passthrough cost me nothing — until the GPUs had to talk to each other

A year after virtualizing my edge AI supercomputer, I finally measured what the hypervisor costs: single-GPU work is free — the VM even won — and four-way tensor-parallel serving pays a third.

#vfio
#gpu-passthrough
#proxmox
#vllm
#nccl
#tensor-parallel
#sdxl
#rtx-3090
#benchmarks
#virtualization