doc/html/platforms.shtml

<!--#include virtual="header.txt"-->

<h1>Platforms</h1>
<h2 id="os">Operating Systems<a class="slurm_link" href="#os"></a></h2>
<ul>
<li><b>FreeBSD</b> &mdash; Limited support, not actively tested.</li>
<li><b>Linux</b> &mdash; Slurm has been thoroughly tested on most popular Linux
distributions using arm64 (aarch64), ppc64, and x86_64 architectures. Some
features are limited to recent releases and newer Linux kernel versions.
Currently supported distributions include:
<ul>
<li>Debian 11 (Bullseye)</li>
<li>Debian 12 (Bookworm)</li>
<li>Debian 13 (Trixie)</li>
<li>RedHat Enterprise Linux 8 (RHEL8) and RHEL8 derivatives</li>
<li>RedHat Enterprise Linux 9 (RHEL9) and RHEL9 derivatives</li>
<li>RedHat Enterprise Linux 10 (RHEL10) and RHEL10 derivatives</li>
<li>SUSE Linux Enterprise Server (SLES) 15</li>
<li>Ubuntu 20.04</li>
<li>Ubuntu 22.04</li>
<li>Ubuntu 24.04</li>
<li>Ubuntu 26.04</li>
</ul>
</li>
<li><b>NetBSD</b> &mdash; Limited support, not actively tested.</li>
<li><b>macOS</b> &mdash; Slurm has run on macOS in the past, but does not
currently. It should be possible to fix this with some adjustments to linker
and compiler flags, and any patches would be appreciated.
</ul>

<h2 id="database">Databases<a class="slurm_link" href="#database"></a></h2>
<p>Slurm will be built with support for MySQL if it finds supported
development libraries at build time. MySQL 8.0.21 and MariaDB 10.2.7 are the
minimum required versions. The <b>slurmdbd</b> daemon checks the version of
the server it connects to at startup and exits with a fatal error if the
server is older than the minimum. Use a version of MySQL or MariaDB that is
still supported by its vendor.
</p>

<p>Further reading:
<ul>
<li><a href="quickstart_admin.html">Quick Start Administrator Guide</a></li>
<li><a href="accounting.html">Accounting and Resource Limits</a></li>
<li><a href="upgrades.html">Upgrade Guide</a></li>
</ul>
</p>

<h2 id="accel">Accelerators<a class="slurm_link" href="#accel"></a></h2>

<p>Slurm has optional support for managing a variety of accelerator cards.
Specific plugins have been developed for:</p>
<ul>
<li><b>gres/gpu</b> &mdash; several autodetection plugins are available for
different vendors:
<ul>
<li><b>AutoDetect=nvml</b> enables autodetection of NVIDIA GPUs through their
proprietary NVML library.</li>
<li><b>AutoDetect=nvidia</b> also enables autodetection of NVIDIA GPUs, but
through generic Linux interfaces rather than their proprietary library. This
plugin doesn't detect MIGs or NVlinks.</li>
<li><b>AutoDetect=rsmi</b> enables autodetection of AMD GPUs through their
proprietary RSMI library (tested on x86_64 and arm64).</li>
<li><b>AutoDetect=amd</b> also enables autodetection of AMD GPUs, but through
the amdgpu kernel driver in sysfs rather than their proprietary library.</li>
<li><b>AutoDetect=oneapi</b> enables autodetection of Intel GPUs through their
proprietary oneAPI library.</li>
<li><b>AutoDetect=nrt</b> enables autodetection of AWS Trainium/Inferentia
devices.</li>
</ul></li>
<li><b>gres/mps</b> &mdash; NVIDIA CUDA Multi-Process Service provides ways to
share GPUs between multiple compute processes</li>
<li><b>gres/shard</b> &mdash; Generic mechanism that provides a way to share
GPUs between multiple compute processes</li>
</ul>

<!--#include virtual="footer.txt"-->