Sporadic Linux Test Failure

This page documents differences in platform architecture between two pre-release runs, in which one passed and one failed.

As for the difference, only the archspec was found different.

Passed Run

libmamba version : 2.3.3
micromamba version : 2.3.3
curl version : libcurl/8.14.1 OpenSSL/3.5.4 zlib/1.3.1 zstd/1.5.7 libssh2/1.11.1 nghttp2/1.67.0
libarchive version : libarchive 3.8.2 zlib/1.3.1 bz2lib/1.0.8 libzstd/1.5.7 libb2/bundled
envs directories : /home/runner/micromamba/envs
package cache : /home/runner/micromamba/pkgs
                /home/runner/.mamba/pkgs
environment : tardis
env location : /home/runner/micromamba/envs/tardis
user config files : /home/runner/.mambarc
populated config files : /home/runner/work/_temp/setup-micromamba/.condarc
virtual packages : __unix=0=0
                   __linux=6.11.0=0
                   __glibc=2.39=0
                   __archspec=1=x86_64_v3
channels : https://conda.anaconda.org/conda-forge/linux-64
           https://conda.anaconda.org/conda-forge/noarch
base environment : /home/runner/micromamba
platform : linux-64

Failed Run

libmamba version : 2.3.3
micromamba version : 2.3.3
curl version : libcurl/8.14.1 OpenSSL/3.5.4 zlib/1.3.1 zstd/1.5.7 libssh2/1.11.1 nghttp2/1.67.0
libarchive version : libarchive 3.8.2 zlib/1.3.1 bz2lib/1.0.8 libzstd/1.5.7 libb2/bundled
envs directories : /home/runner/micromamba/envs
package cache : /home/runner/micromamba/pkgs
                /home/runner/.mamba/pkgs
environment : tardis
env location : /home/runner/micromamba/envs/tardis
user config files : /home/runner/.mambarc
populated config files : /home/runner/work/_temp/setup-micromamba/.condarc
virtual packages : __unix=0=0
                   __linux=6.11.0=0
                   __glibc=2.39=0
                   __archspec=1=x86_64_v4
channels : https://conda.anaconda.org/conda-forge/linux-64
           https://conda.anaconda.org/conda-forge/noarch
base environment : /home/runner/micromamba
platform : linux-64

If we look closely:

2.log (FAILED) - Line 2073:
  virtual packages : __unix=0=0
                     __linux=6.11.0=0
                     __glibc=2.39=0
                     __archspec=1=x86_64_v4    ← Intel Xeon with AVX-512

  3.log (PASSED) - Line 2060:
  virtual packages : __unix=0=0
                     __linux=6.11.0=0
                     __glibc=2.39=0
                     __archspec=1=x86_64_v3    ← AMD EPYC with AVX2

What This Means

x86_64_v4 vs x86_64_v3:

• v4 supports AVX-512 instructions (Intel Skylake-X and newer)

• v3 supports AVX2 instructions (Intel Haswell and newer)

When GitHub assigns a v4 runner, NumPy/SciPy detect AVX-512 support at runtime and use different SIMD code paths. This causes tiny numerical differences (~1e-13 relative error) that exceed our tolerance levels.