Skip to content

NewsGPU

ComfyUI v0.46.0 Drops DirectML: What AMD and Intel GPU Users on Windows Should Do

ComfyUI v0.46.0, released on 18 November 2026, removes DirectML support and the --directml launch option because no current PyTorch version has a DirectML plugin. Anyone who used it should move to ROCm on AMD, PyTorch xpu on Intel, the portable builds or the CPU.

By , Editorial team

2 min read

ComfyUI v0.46.0 Drops DirectML: What AMD and Intel GPU Users on Windows Should Do

ComfyUI v0.46.0, published on GitHub on 18 November 2026, removes DirectML support. Older versions of ComfyUI's README offered DirectML (via the torch-directml package) as an option for AMD cards on Windows, started with a --directml launch option. That option is now gone. Forge PC hasn't tested the new release; this piece is based on ComfyUI's release notes, the pull request that removed DirectML and ComfyUI's current installation guide.

Why DirectML went

The change came in pull request #17711, "Remove directml support", by comfyanonymous, merged on 12 November 2026. His explanation is short: "The last pytorch version that had a directml plugin was pytorch 2.5 and we don't support that anymore."

The pull request deletes the --directml argument and the code behind it. The removed code had already printed a warning when DirectML was used: "torch-directml barely works, is very slow, has not been updated in over 1 year and might be removed soon, please don't use it, there are better options."

Who's affected

  • NVIDIA users: no change. NVIDIA cards use CUDA builds of PyTorch.

  • Anyone who launched ComfyUI with --directml: that launch option no longer exists in v0.46.0, so you'll need to switch to one of the supported routes below.

  • Anyone following the current install guide: its AMD and Intel instructions use ROCm and PyTorch xpu, not DirectML.

What to use instead

ComfyUI's installation guide now lists these routes for Windows users without an NVIDIA card:

  • AMD on Windows (ROCm 10.0): install AMD's multi-architecture PyTorch packages. The guide says to use Windows 11, a current AMD graphics driver and 64-bit Python 3.13, and that a separate HIP SDK install isn't needed. It lists RDNA 2, RDNA 3, RDNA 3.5 and RDNA 4 among supported architectures, with examples from the RX 7600 to the RX 9070 XT and Ryzen AI Max.

  • Intel Arc (Windows and Linux): install PyTorch with Intel's xpu support using pip, as described in the guide.

  • Portable builds: ComfyUI offers separate Windows portable downloads for NVIDIA, AMD and Intel.

  • No supported GPU: you can run on the CPU with the --cpu option.

Our ComfyUI install guide covers the basics, and we have separate guides to running local AI on an AMD Radeon GPU and on an Intel Arc GPU. If your card is short of video memory, see how to run ComfyUI on low VRAM.

What else is in v0.46.0

The release notes list plenty besides the DirectML change, including a new node to split image channels, support for saving single-channel PNGs, a Video Deflicker node with faster RAFT optical flow, support for Qwen3-1.7B in the Generate Text node, a warning when a GPU that supports CUDA 13 isn't using it, and a --force-output-rerun option. There are also many partner (API) node updates, including a MiniMax H3 Image node.

Our verdict

If you run ComfyUI on NVIDIA, nothing changes. If you've been using DirectML on an AMD or Intel card, v0.46.0 is the point to move to ROCm on AMD or PyTorch xpu on Intel Arc, as ComfyUI's own install guide recommends. If you can't switch yet, don't update a working DirectML install until you're ready to.

The bottom line

If you run ComfyUI on NVIDIA, nothing changes. If you've been using DirectML on an AMD or Intel card, v0.46.0 is the point to move to ROCm on AMD or PyTorch xpu on Intel Arc, as ComfyUI's own install guide recommends. If you can't switch yet, don't update a working DirectML install until you're ready to.

Sources

  1. 01ComfyUI on GitHub, Release v0.46.0 (18 Nov 2026)
  2. 02ComfyUI on GitHub, Pull request #17711: Remove directml support
  3. 03ComfyUI on GitHub, Installation guide (docs/installation.md)
  4. 04ComfyUI on GitHub, README at v0.3.0 (DirectML section)

About the author

Forge PC Staff

Editorial team

The shared byline for articles researched, written and edited by the Forge PC editorial team. How we source, check and correct our work is set out in our editorial policy.

More from Forge PC Staff

Read next

How to Work Out What Running Local AI Costs in Electricity (UK, 2026)

How-To

How to Work Out What Running Local AI Costs in Electricity (UK, 2026)

Multiply watts by hours, divide by 1,000 and multiply by your unit rate. At Ofgem's average cap rate of 26.32p per kWh (1 October to 31 December 2026), an RTX 5090 at its full 575W rating costs about 15p an hour. Measure your own PC with nvidia-smi or a plug-in energy monitor.

Forge PC Staff4 min

Intel Arc B580 vs RTX 3060 12GB for Local AI: Which Budget 12GB Card?

Versus

Intel Arc B580 vs RTX 3060 12GB for Local AI: Which Budget 12GB Card?

Intel's Arc B580 and NVIDIA's RTX 3060 12GB both have 12GB of GDDR6 for local AI. The RTX 3060 uses CUDA, which Ollama, llama.cpp and ComfyUI support directly; the newer B580, rated at 456GB/s, runs through Vulkan, SYCL or PyTorch xpu. Beware the 8GB RTX 3060.

Forge PC Staff3 min

How to Test Your Home Network Speed With iperf3

How-To

How to Test Your Home Network Speed With iperf3

iperf3 measures the speed of your own network rather than your internet connection. Install it on two devices, run iperf3 -s on one and iperf3 -c with its IP address on the other, then use -R for the reverse direction and -P for parallel streams to find Wi-Fi, cable or port bottlenecks.

Forge PC Staff3 min

How to Run MiniMax H3 Image Locally in ComfyUI

How-To

How to Run MiniMax H3 Image Locally in ComfyUI

ComfyUI natively supports MiniMax H3 Image, an open-weights model for text-to-image and reference-image editing in six fixed steps at up to 4096x4096. Update ComfyUI, download the diffusion model, VAE and Qwen3-VL text encoder, load the template, and connect images and the VAE to edit.

Forge PC Staff3 min