Metrixel on Linux: Full-Bandwidth 3D Pipelines on Your GPU

Introduction

For most of its life since its inception, Metrixel has run on macOS and Windows. However, people building 3D datasets for machine learning, including but not limited to training image-to-3D models, generating multi-view renders at scale, baking volumes for shape representation, mostly do that work on Linux and on machines with discrete NVIDIA GPUs. With 2026.2.1, Metrixel meets them there: Linux is now a generally-available platform, on equal footing with macOS and Windows.

In this article, I will explain what the Linux build includes, why it is considered a particularly good home for heavy 3D-ML work, and how it slots into headless, GPU-accelerated pipelines.

---

Why Linux, and Why Now

Metrixel is, at heart, a tool for turning messy 3D files into clean, reproducible, machine-ready datasets. Teams that need that most, such as AI/ML researchers, robotics and SLAM groups, simulation teams, run their pipelines on Linux. Their GPU training servers run Linux. Their CI runners run Linux. The SSH remote boxes run Linux.

For us, putting Metrixel on Linux is not simply about a checkbox for OS support. For us, we believe in being present where most of the work happens. On Linux, Metrixel uses your GPU's native Vulkan driver, so rendering and signed-distance-field generation run at full bandwidth on a discrete NVIDIA GPU. This is also noticeably faster than on Apple Silicon for large volumes. If you have ever wanted to point Metrixel at a rack of GPUs, this is the release that gets you started.

---

So What Ships?

We simply ship a single relocatable tarball that covers Ubuntu 24.04 and Red Hat Enterprise Linux 9 (x86_64). There is nothing to compile.

unzip Metrixel-2026.2.1-linux-x64.tar.gz.zip
tar xzf Metrixel-2026.2.1-linux-x64.tar.gz -C /opt
/opt/Metrixel-2026.2.1-linux-x64/install.sh   # wires up both Metrixel and EntVista Hub

A few things worth pointing out, on that note:

• Tarball bundles EntVista Hub: It is our sign-in companion that authenticates every EntertainmentVista Studio tool. One download will install both, so you do not need to chase down a second package.
• It is relocatable. Extract it anywhere you have space. This can be /opt, a home directory, a scratch volume on a training node. It does not simply assume a system-wide install location.
• install.sh is an optional polish. It adds the application-menu entry and command-line symlinks for desktop use. On a headless server you can skip it and run the binaries directly.

Uninstalling is symmetrical: ./install.sh --uninstall removes the desktop entries and symlinks, and deleting the extracted directory removes the rest.

---

Built to Run on Servers and in Containers

The Linux tarball supports a fully silent, unattended install. There're simply no display server, no clicks. This makes it practical on training nodes and in container builds. Here is the shape of a SageMaker / generic-Linux image:

# Pull in the minimal runtime deps for headless mode
RUN apt-get update && \
    apt-get install -y --no-install-recommends fontconfig libxkbcommon0 ca-certificates && \
    rm -rf /var/lib/apt/lists/*

# Unpack the relocatable tarball (bundles Metrixel + EntVista Hub)
COPY Metrixel-2026.2.1-linux-x64.tar.gz.zip /tmp/
RUN unzip -q /tmp/Metrixel-2026.2.1-linux-x64.tar.gz.zip -d /tmp && \
    tar xzf /tmp/Metrixel-2026.2.1-linux-x64.tar.gz -C /opt && \
    rm /tmp/Metrixel-2026.2.1-linux-x64.tar.gz*
ENV PATH=/opt/Metrixel-2026.2.1-linux-x64:$PATH

From there, Metrixel's command-line mode runs the full pipeline headless. The next post will cover this in detail. (install.sh is desktop polish; a container can skip it and put the extracted directory on PATH, as above.)

A note on size: the Linux tarball is large, as it carries the full rendering and tensor-export runtime. So please do give your build host and target enough scratch space.

---

How Linux Fits Alongside macOS and Windows

Linux is a first-class, production-recommended platform. The same engine and pipeline that we ship everywhere, is now validated on Linux training hosts. The relocatable tarball is how Metrixel is distributed on Linux, today and going forward: there is no separate native-package or installer step. Extract it, optionally run install.sh, and you are done.

---

So Who Is This For?

• AI/ML teams training image-to-3D or shape-representation models on Linux GPU servers,
• Researchers generating large multi-view datasets where GPU bandwidth matters,
• Robotics, SLAM, and simulation groups whose pipelines already run on Linux,
• Platform engineers packaging Metrixel into container images and CI.

---

Conclusion

Metrixel on Linux is about meeting 3D-ML work on discrete-GPU servers, in containers, and in CI. We are giving Metrixel the full bandwidth of a native Vulkan driver. The Linux tarball is ready today on Ubuntu 24.04 and RHEL 9 and bundles everything you need: one relocatable download is the Linux distribution format, today and going forward.

Up next:
Running the whole Metrixel pipeline from the command line: and exporting signed-distance fields for training.