3090. Models like flux or wanvideo with fp16 text encoders will push your VRAM usage over 20GB. LLMs with functional context (or workflows with multi-model, not

3090. Models like flux or wanvideo with fp16 text encoders will push your VRAM usage over 20GB. LLMs with functional context (or workflows with multi-model, not modal, tie in) will all break 16GB. There's no point to performance without context. At its pricepoint, the 5080 is purely a gaming card.