ComfyUI ComfyUI 0.42.0 (commit cdf272f), PyTorch 2.14.1 CPU: our test log
ComfyUI 0.42.0 installed in 17.42 s and its CPU-only server was ready in 30.1 s with 1014 node types, but we did not generate an image.
Tested by us on 2026-10-26. Local open-weight model, not the vendor's hosted models. A smoke test on a CPU-only box, not a benchmark.
Result at a glance
| Field | Value |
|---|---|
| Result | Passed |
| Version tested | ComfyUI 0.42.0 (commit cdf272f), PyTorch 2.14.1 CPU |
| Date and time | 2026-10-26, 10:40 CDT (America/Chicago) |
| Task | Clone ComfyUI, install it into a fresh Python 3.12 environment with the CPU build of PyTorch, start the server with --cpu, and query /system_stats and /object_info. Pass means the server starts without a GPU and reports its nodes. No checkpoint was downloaded and no image was generated. |
| Model used | none (no image model loaded) |
| Hardware | Linux x86_64 virtual machine (Debian 13), 8 vCPU (Intel Xeon, 1 thread per core), 15 GiB RAM, no GPU |
| Install | 17.42 s, 2193 MB |
| Time to ready | 30.1 s |
| Node types | 1014 |
What we found
- The server reported ComfyUI 0.42.0 on a single CPU device and PyTorch 2.14.1+cpu.
- /object_info listed 1014 node types in 0.31 s, including KSampler, CLIPTextEncode, CheckpointLoaderSimple, VAEDecode and SaveImage.
- The checkpoint list was empty because we installed no model files.
What we ran, step by step
| Step | Wall time | Exit code | Notes |
|---|---|---|---|
| Clone repository | 2.33 s | 0 | Shallow clone, commit cdf272f. |
| Install with uv (cold, no cache) | 17.42 s | 0 | 2193 MB on disk including the CPU PyTorch build (2.14.1+cpu). Load 18.69. |
| Task command (start server, probe API) | 32.68 s | 0 | Ready after 30.1 s on a shared machine (load 18.42). |
# Clone repository
git clone --depth 1 https://github.com/comfyanonymous/ComfyUI.git
# Install with uv (cold, no cache)
uv pip install torch torchvision torchaudio (CPU wheels) && uv pip install -r requirements.txt
# Task command (start server, probe API)
python main.py --cpu --listen 127.0.0.1 --port 8189 (driven by tasks3/comfy_task.py)Raw output (excerpts)
{
"ready_s": 30.1,
"system_stats": {
"comfyui_version": "0.42.0",
"python_version": "3.12.15 (main, Oct 21 2026, 17:47:06) [Clang 23.1.3 ]",
"pytorch_version": "2.14.1+cpu",
"ram_total": 16779640832
},
"devices": [
[
"cpu",
"cpu"
]
],
"object_info_s": 0.31,
"node_types": 1014,
"sample_nodes": [
"KSampler",
"CheckpointLoaderSimple",
"CLIPTextEncode",
"SaveImage",
"VAEDecode"
],
"checkpoints_listed": []
}import subprocess, time, json, urllib.request, sys, os, signal
PY = "/workspace/tests2/work/py/comfyui/bin/python"
cwd = "/workspace/tests2/work/comfy/ComfyUI"
t0 = time.time()
p = subprocess.Popen([PY, "main.py", "--cpu", "--listen", "127.0.0.1", "--port", "8189", "--disable-auto-launch"], cwd=cwd, stdout=open("/workspace/tests2/logs/comfy-server.log", "w"), stderr=subprocess.STDOUT)
def get(path):
return json.load(urllib.request.urlopen("http://127.0.0.1:8189" + path, timeout=30))
res = {}
try:
for i in range(240):
try:
st = get("/system_stats"); res["ready_s"] = round(time.time() - t0, 1); break
except Exception:
time.sleep(1)
else:
raise SystemExit("server never became ready")
res["system_stats"] = {k: st["system"].get(k) for k in ("comfyui_version", "python_version", "pytorch_version", "ram_total")}
res["devices"] = [(d["name"], d["type"]) for d in st["devices"]]
t1 = time.time(); oi = get("/object_info"); res["object_info_s"] = round(time.time() - t1, 2)
res["node_types"] = len(oi)
res["sample_nodes"] = [n for n in ("KSampler", "CheckpointLoaderSimple", "CLIPTextEncode", "SaveImage", "VAEDecode") if n in oi]
res["checkpoints_listed"] = oi["CheckpointLoaderSimple"]["input"]["required"]["ckpt_name"][0]
finally:
p.send_signal(signal.SIGINT)
try: p.wait(20)
except Exception: p.kill()
print("RESULT_JSON " + json.dumps(res))Gotchas and things that broke
- Nothing useful can be generated until you add a checkpoint; the common Stable Diffusion files are multiple gigabytes.
- The full install, with PyTorch, is about 2.2 GB.
What we did not test
- Generating any image: no checkpoint was downloaded (the SD-Turbo file we considered is 5.2 GB, and the machine's free memory could not hold it).
- GPU use, custom nodes, the ComfyUI Manager, workflows from the web interface, and the browser UI.
- Speed of image generation on CPU or on a GPU.
Reproduce it
Versions change. These commands install whatever is latest on the day you run them unless a version is pinned, so expect different results later. The version we measured is in the table above.
git clone https://github.com/comfyanonymous/ComfyUI && cd ComfyUI
uv venv -p 3.12 .venv && VIRTUAL_ENV=.venv uv pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cpu && VIRTUAL_ENV=.venv uv pip install -r requirements.txt
.venv/bin/python main.py --cpu --port 8189
curl http://127.0.0.1:8189/system_statsHow to read this test
Every number on this page was measured on this one machine, which also ran other jobs, so timings are indicative and vary run to run. A pass means our small task worked once or as stated, a fail means it did not work as we ran it; neither says how the tool behaves with a vendor's frontier model, on a GPU, or at scale. Where a tool could not run without an account or a GPU, we say so instead of guessing.
See the testing rig for the hardware, the local model and the rules we follow, and the test hub for every other test.
Related tests: voice, speech and image
- faster-whisper (speech to text): faster-whisper 1.2.1 transcribed 41 s of speech at 2.4 percent word error with the base.en model in float32 (deterministic), but the int8 setting gave 2 to 60 percent on identical audio across runs.
- Piper (text to speech): Piper 1.8.0 synthesised 10 sentences (about 41 s of 22.05 kHz audio) on CPU in 1.4 to 1.5 s each including process start; 3.5 s uv install, 167 MB.
- whisper.cpp: We could not run whisper.cpp: no prebuilt Linux release was available for download and this machine has no make, g++ or cmake to build it.
FAQ
Does ComfyUI start without a GPU?
Yes: with --cpu the server was ready in 30.1 s on an 8-core CPU-only machine and listed 1014 nodes. We did not generate an image.
Sources
- ComfyUI repository (fetched 2026-10-26)
Related
Markdown version for agents: /tests/comfyui.md