xiaohong@spark-b3e0:~/ComfyUI$ sudo dmidecode -t memory | grep -E “(Size:|Volatile)” [sudo] password for xiaohong: Size: 128 GB Memory Operating Mode Capability: Volatile memory Non-Volatile Size: None Volatile Size: 128 GB Cache Size: None Logical Size: None xiaohong@spark-b3e0:~/ComfyUI$ sudo dmes…

I got it working with fp8 following this workflow: Flux 2 | ComfyUI_examples But it wouldn’t work with full sized version (just got an error message ‘Killed’!)

Have you guys tried to runsudo sh -c 'sync; echo 3 > /proc/sys/vm/drop_caches'before executing ComfyUI?

Thank you for your reply, I have used

To run the full sized version of FLUX.2 you can either start ComfyUI with the --cache-none argument (which is slower because it does not keep the model in memory after execution) or use the full BF16 GGUF version of FLUX.2 from city96/FLUX.2-dev-gguf · Hugging Face There is also a thread on the of…

I wrote a custom node that loads .safetensors using the fastsafetensors library directly to VRAM from storage, bypassing any copies. This works exactly as expected on DGX Spark, loading the BF16 version of FLUX.2 in only a few seconds: [image] GitHub - phaserblast/ComfyUI-DGXSparkSafe…

hedin.e kralvis 1m Apologies in advance, i’m still new to Linux in general. but how do you actually install and use this. I did the git pull and the directory is there. so i did the pip install -r requirements.txt and that seemed to work. But I’m still getting the out memory errors so I’m obvious…

Ensure you’re cloning the repo into ComfyUI/custom_nodes. After restarting ComfyUI, add add the new DGXSparkSafetensorsLoader node, which will replace the default “Load Diffusion Model” node. Connect the output of the new DGX loader to the rest of your workflow. That should do it: [dgx-node]

I used your node, and it was still the same Killed

Are you using Flux.2? Post your workflow and I’ll have a look.

[Screenshot from 2025-12-04 17-33-09]

I ran ComfyUI on DGXSPARK, flux.2 reported insufficient memory error, please help me analyze it

Accelerated Computing DGX Spark / GB10 User Forum DGX Spark / GB10

vr8vr8 July 8, 2026, 7:01pm 23

If you’re on a 128 GB Spark, here’s a route that runs FLUX.2-dev in ~66 GB and ~3× faster — real NVFP4 compute (W4A4 torchao), not just weight-only storage (which saves memory but doesn’t speed up the matmul). Writeup with numbers + a PR to spark-vllm-docker here: FLUX.2 on the Spark: ~3× faster with real NVFP4 compute (not just weight-only) — PR to spark-vllm-docker

Topic		Replies	Views
Buyers beware: DGX Spark limited to 64GB in ComfyUI DGX Spark / GB10	26	3425	May 1, 2026
My Comfyui setup and patches DGX Spark / GB10	10	1953	August 9, 2026
Dgx spark comfyUI DGX Spark / GB10 cuda	1	576	April 29, 2026
ComfyUI: High-Performance Model Loading and DGX Spark Optimizations DGX Spark / GB10 Projects generative_ai	3	702	April 26, 2026
ComfyUI setup optimized for DGX Spark DGX Spark / GB10 Projects spark , generative_ai	10	3215	August 3, 2026
Spark-comfyui, a self-healing ComfyUI setup for DGX Spark DGX Spark / GB10 Projects	17	2016	August 19, 2026
Unlocking the Power of the Spark In ComfyUI - No Crashes! DGX Spark / GB10 Projects	13	3570	July 18, 2026
Spark: cudaErrorNotPermitted in comfyui - but only after Docker sits idle for hours DGX Spark / GB10 cuda , docker-machine-learning	1	456	December 24, 2025
ComfyUI container for DGX Spark DGX Spark / GB10 Projects stable-diffusion-xl	13	1862	March 28, 2026
Memory Creep on DGX Spark: Where Your 128 GB Actually Goes (And How to Stop It) DGX Spark / GB10 jetson , nemotron	2	1445	March 30, 2026

I ran ComfyUI on DGXSPARK, flux.2 reported insufficient memory error, please help me analyze it

Related topics