nvidia-mathdx

library25.6.0pypypi✓ verified 79d ago

MathDx is a NVIDIA device library for accelerated math operations on GPUs, providing high-performance implementations for common mathematical functions. Version 25.6.0 is the latest. This package is part of the NVIDIA CUDA ecosystem and is updated frequently with new GPU architectures.

ai-ml

pip install nvidia-mathdx

package ↗homepage ↗docs ↗

INSTALL

IMPORT

SIG · NVIDIA-MATHDX

nvidia-mathdx

ai-mlpythonv25.6.0

Install

2.4s avg

Import

—

Disk

137MB

Pass rate

10/ 10

Env Coverage10 / 10

glibc

3.9–3.13

musl

3.9–3.13

Install & Compatibility

Where this runs

tested against v25.6.0 · pip install

no network on importno background threads

Install × environment matrix

Each cell = how many times install + import succeeded across repeated harness runs. Partial = flaky.
glibc = Debian/Ubuntu slim · musl = Alpine Linux

musl

py 3.10–3.95 runs

installs and imports cleanly · install 0.0s · import 0.000s · 138.5MB

glibc

py 3.10–3.95 runs

installs and imports cleanly · install 2.4s · import 0.000s · 139MB

137MB installed

● package 137MB

Code

Verified usage

Verified import paths — ran on the pinned version, not inferred.

nvidia

✓ import nvidia

✗ import nvidia_mathdx as mathdx

Basic usage showing import pattern and calling an elementwise math function on a CuPy array.

import nvidia_mathdx as mathdx import cupy as cp # Create a CuPy array a = cp.array([1.0, 2.0, 3.0, 4.0]) # Use mathdx elementwise math (example: sin) b = mathdx.sin(a) print(b)

Debug

Known issues

breakingThe package was originally distributed as a single module; in recent versions it has been split into subpackages (e.g., nvidia_mathdx.fft, nvidia_mathdx.linalg). Old code importing `nvidia_mathdx` directly may break if relying on functions that moved.

fix

Update imports to use subpackages: e.g., `from nvidia_mathdx.fft import fft`.

affects: >=24.12.0

gotchaThe function `mathdx.sin` may return a GPU array that is not automatically synchronized with the host. Accessing `.get()` on the result without explicit synchronization can cause segmentation faults.

fix

Use `cp.cuda.Stream.null.synchronize()` before reading results back to host, or use CuPy's default synchronization.

affects: all

deprecatedThe top-level function `mathdx.matrix_inverse` is deprecated in favor of `mathdx.linalg.inv`.

fix

Replace calls to `mathdx.matrix_inverse` with `mathdx.linalg.inv`.

affects: >=25.03.0

Upgrade

Version history

25.6.0latest on PyPI · released Nov 13, 2025

Audit

Dependencies

cupy-cuda12xoptionalMathDx often depends on CuPy for array operations and GPU context.

nvidia-cuda-runtime-cu12optionalRequired for CUDA runtime if using MathDx kernels directly.

Agent activity

2 hits · last 30 days

node

Resources

packagehttps://pypi.org/project/nvidia-mathdx/ ↗

homepagehttps://developer.nvidia.com/cufftdx-downloads ↗

docshttps://docs.nvidia.com/cuda/cufftdx/index.html ↗