ComfyUI cuts VRAM use and doubles generation speeds
ComfyUI just introduced dynamic VRAM, a feature that streams model layers on demand and unloads them immediately after execution. Instead of crashing low-VRAM setups, mid-tier cards like the RTX 5060 can now run heavyweight models like Flux 2dev with render times cut in half. Running frontier generation locally just became a lot more accessible.
ComfyUI's new dynamic VRAM feature lets creators squeeze heavyweight models onto local hardware without running into out-of-memory errors.
How it works and what it delivers
Historically, running ComfyUI meant loading entire models directly into VRAM upfront. For creators with mid-tier setups, that usually meant crashes or settling for smaller, degraded configurations.
Dynamic VRAM changes that baseline: instead of holding everything in GPU memory, it loads model parts strictly when needed during execution and unloads them right after. This keeps memory footprints minimal, virtually eliminates OOM errors, and allows users to push for larger models or higher resolutions on the same card.
Surprisingly, this on-the-fly swapping doesn't introduce latency; it actually speeds up generation. Testing on an RTX 5060 with 1 2.2 showed clear runtime drops. With Flux 2dev, image generation times were cut in half across multiple cases. Users can test this immediately simply by updating ComfyUI to the latest build.
Why it matters
If you are running mid-range GPUs like an RTX 5060, this is an instant upgrade. It smartly replaces blunt brute-force VRAM allocation with granular layer swapping, effectively cutting generation times while opening top-tier image models to everyday hardware.
Sources (1)
- youtube.com(opens in a new tab)Originalyoutube.com