PyOverdrive vs dpnp in 2026

Q: Is PyOverdrive free and open source?

PyOverdrive is open source under the MIT License . The sources here do not state any paid tier, so the only price they give is free. You install it with pip straight from the GitHub repository .

PyOverdrive and dpnp compared on what each does for NumPy acceleration, what each costs, how each is installed and what to check before you pick one.

By SayDeploy · 2026-10-24

SayDeploy is published by LunarWerx, which makes some of the tools compared here.

dpnp is a library that can be installed with conda, mamba or pip and is published under the BSD-3-Clause license ^[1]^[1]. PyOverdrive patches supported NumPy functions so existing code uses faster paths, while calls it does not support run on stock NumPy ^[2]. dpnp's sources describe its install methods and license ^[1]^[1], and PyOverdrive's sources describe threaded paths for common NumPy operations, small-batch linear algebra, einsum optimization and isin hash-set membership ^[2]^[2]^[2]^[2]. PyOverdrive is free and open source under the MIT License ^[2], and dpnp is open source under the BSD-3-Clause license ^[1].

Option	License	Install	Python version	Price
dpnp	BSD-3-Clause license ^[1]	conda, mamba or pip ^[1]	Not published ^[1]	Not published ^[1]
PyOverdrive	MIT License ^[2]	pip from the GitHub repository ^[2]	Python 3.12 or newer ^[2]	Free, open source ^[2]

The short answer

dpnp is a library for NumPy acceleration that can be installed using conda, mamba or pip package managers ^[1]. PyOverdrive patches supported NumPy functions so existing code uses faster paths, and calls it does not support run on stock NumPy ^[2]. dpnp's page says it can be installed with conda, mamba or pip and is published under the BSD-3-Clause license ^[1]^[1]. PyOverdrive's sources cover threaded paths for common operations like np.sin, cos, tan, exp, log, log10 and tanh on large arrays ^[2], small-batch linear algebra for np.linalg.qr and np.linalg.eigvalsh ^[2], large np.einsum calls sent through NumPy's own optimize=True planner ^[2], and np.isin on 1-D StringDType or object arrays using hash-set membership ^[2].

What dpnp is built for

dpnp's page says it can be installed using conda, mamba or pip package managers ^[1]. The repository is published under the BSD-3-Clause license ^[1]. The facts provided here do not describe what dpnp does for a developer beyond its install methods and license. Check dpnp's own pages for what operations it supports, what hardware it targets, and how it accelerates NumPy code, since the facts here are limited to installation and licensing.

What PyOverdrive is built for

PyOverdrive is made by the team that publishes this site.

Calling pyoverdrive.enable() patches supported NumPy functions so existing code uses faster paths, and calls it does not support run on stock NumPy ^[2]. A fast path runs only when a check confirms the input is in its measured range; anything else, or a fast path that raises an error, goes to stock NumPy ^[2].

Threaded paths for np.sin, cos, tan, exp, log, log10 and tanh on C-contiguous float64 and float32 arrays run above a size floor calibrated per op and dtype between 3e5 and 3e6 elements ^[2]. np.linalg.qr on large stacks of 3x3 matrices uses a vectorized closed-form Householder approach, and np.linalg.eigvalsh has closed-form 2x2 and 3x3 paths for float batches above a size threshold ^[2]. Large np.einsum calls are sent through NumPy's own optimize=True planner once they pass a measured size threshold ^[2]. np.isin on 1-D StringDType or object arrays can use hash-set membership instead of the default method ^[2].

Pricing and fees

PyOverdrive is open source under the MIT License ^[2]. No paid tier price is stated in its sources. dpnp is published under the BSD-3-Clause license ^[1]. The facts here do not state a price for dpnp, so no price is published here ^[1]. Both are open source, and neither source states a paid plan or a fee.

Setting it up and daily use

dpnp can be installed using conda, mamba or pip package managers ^[1]. The facts here say nothing further about setup or daily use, so check dpnp's own pages for what comes after installation.

PyOverdrive needs Python 3.12 or newer ^[2]. You install it with pip straight from the GitHub repository ^[2]. The package is marked OS Independent, and the test suite has been run on Windows and on Linux x86-64 ^[2]. To turn it on, call pyoverdrive.enable(), which patches supported NumPy functions so existing code uses faster paths ^[2]. A fast path runs only when a check confirms the input is in its measured range; anything else goes to stock NumPy ^[2].

You can run python -m pyoverdrive --selfcheck to compare every fast path with stock NumPy on your own machine ^[2]. You can run python -m pyoverdrive --calibrate to re-time the threaded paths at their own thresholds on your CPU and switch off any that do not pay there ^[2]. pyoverdrive.explain() tells you which path a call would take and why, without running the call ^[2]. You can turn off one fast path with disable_path() or PYOVERDRIVE_DISABLE, or call pyoverdrive.disable() to restore the original NumPy functions ^[2]. python -m pyoverdrive --demo times the headline operations with stock NumPy and with PyOverdrive on your machine in about 20 seconds ^[2].

Things to check before you choose

Ask what operations you need to accelerate. PyOverdrive's claims name threaded paths for np.sin, cos, tan, exp, log, log10 and tanh ^[2], small-batch paths for np.linalg.qr and np.linalg.eigvalsh ^[2], large np.einsum optimization ^[2] and np.isin hash-set membership ^[2]. Check dpnp's own pages for the operations it supports, since the facts here do not list them ^[1].

Ask what hardware each tool targets. The facts here do not state what hardware dpnp targets ^[1]. The package for PyOverdrive is marked OS Independent, and the test suite has been run on Windows and on Linux x86-64 ^[2].

Ask what Python version you need. The facts here do not state a Python requirement for dpnp ^[1]. PyOverdrive needs Python 3.12 or newer ^[2].

Ask how each tool handles calls it does not accelerate. PyOverdrive sends unsupported calls to stock NumPy ^[2], and the facts here do not describe dpnp's fallback behavior ^[1].

How to choose

If you want to patch existing NumPy code with a single call so supported functions use faster paths, PyOverdrive's claims describe pyoverdrive.enable() ^[2].

If you need threaded paths for np.sin, cos, tan, exp, log, log10 or tanh on large C-contiguous float64 or float32 arrays, PyOverdrive's claims describe those paths ^[2].

If you work with small-batch linear algebra on stacks of 3x3 matrices or need closed-form 2x2 and 3x3 eigvalsh paths, PyOverdrive's claims describe those ^[2].

If you need large np.einsum calls sent through NumPy's own optimize=True planner, PyOverdrive's claims describe that ^[2].

If you need np.isin on 1-D StringDType or object arrays using hash-set membership, PyOverdrive's claims describe that ^[2].

If you want to install a library with conda, mamba or pip, dpnp's page says that is available ^[1].

If you want a tool you can calibrate on your own CPU and check against stock NumPy, PyOverdrive's claims describe --calibrate and --selfcheck ^[2]^[2].

Frequently asked questions

Is dpnp free to use?

dpnp is published under the BSD-3-Clause license, which is an open source license ^[1]. The facts here do not state a paid plan or a price for dpnp, so no price is published here ^[1]. Check dpnp's own pages if you need details on any paid offering or support terms.

Is PyOverdrive free and open source?

PyOverdrive is open source under the MIT License ^[2]. The sources here do not state any paid tier, so the only price they give is free. You install it with pip straight from the GitHub repository ^[2].

How do you install dpnp and PyOverdrive?

dpnp can be installed using conda, mamba or pip package managers ^[1]. PyOverdrive is installed with pip straight from the GitHub repository ^[2]. The facts here do not state a Python requirement for dpnp ^[1]. PyOverdrive needs Python 3.12 or newer ^[2].

Can PyOverdrive speed up existing NumPy code?

Calling pyoverdrive.enable() patches supported NumPy functions so existing code uses faster paths, and calls it does not support run on stock NumPy ^[2]. A fast path runs only when a check confirms the input is in its measured range; anything else goes to stock NumPy ^[2]. You can run python -m pyoverdrive --demo to time headline operations on your machine in about 20 seconds ^[2].

Can you turn off PyOverdrive paths or check them against stock NumPy?

python -m pyoverdrive --selfcheck compares every fast path with stock NumPy on your own machine ^[2]. python -m pyoverdrive --calibrate re-times the threaded paths at their own thresholds on your CPU and switches off any that do not pay there ^[2]. You can turn off one path with disable_path() or PYOVERDRIVE_DISABLE, or call pyoverdrive.disable() to restore the original NumPy functions ^[2].

It is open source under the MIT License. ^[2]

Try PyOverdrive