molbals-stats

Single ComfyUI output node for prompt diagnostics.

Node

• molbals-stats

Connect the trigger input to the final value whose generation you want to wait for, such as the final latent or decoded image. The node displays and returns JSON plus separate numeric outputs:

• peak_vram_mb
• peak_ram_mb
• total_seconds

peak_ram_mb is the ComfyUI process RSS peak sampled during the prompt. peak_vram_mb is sampled from the active ComfyUI torch device; on CUDA it uses torch.cuda.mem_get_info, so it reflects device memory in use during the prompt, not only tensor allocations.

The node is an output node and forces fresh execution with IS_CHANGED, so it should not replay cached stats from an earlier prompt.

Example JSON output

{
  "complete": false,
  "device": "NVIDIA GeForce RTX 3080 Laptop GPU",
  "node_id": "284",
  "peak_ram_delta_mb": 1009.0,
  "peak_ram_mb": 15070.91,
  "peak_vram_delta_mb": 6067.86,
  "peak_vram_mb": 7425.36,
  "prompt_id": "8a2c25f1-722f-40b1-943f-c2a54c134e70",
  "sample_count": 13644,
  "sample_interval_seconds": 0.05,
  "total_seconds": 847.442
}