Bottleneck vs PyOpenCL in 2026

Q: What license does Bottleneck use?

Bottleneck is distributed under a Simplified BSD license . The facts here do not state a price for Bottleneck, so none is published here. Check the license text and the project's own pages before you rely on its terms.

Bottleneck and PyOpenCL compared on what each does, what each costs, how each is installed, and what to check before you choose one.

By SayDeploy · 2026-10-26

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

Bottleneck ships ready-made install packages for the most common platforms ^[1]. PyOpenCL is open-source under the MIT license and free for commercial, academic and private use ^[2]. PyOverdrive runs faster versions of supported NumPy functions after one call ^[3].

Option	What it does	Price and fees	Install and setup	Licence and status
Bottleneck	A collection of fast NumPy array functions written in C ^[1]. It includes a built-in benchmark suite ^[1]. Only int32, int64, float32 and float64 arrays are accelerated, and other data types use slower unaccelerated functions ^[1].	Price not published	Binary wheels on PyPI for the most common platforms ^[1]. A source install needs Python greater than 3.9 and NumPy 1.16.0 or newer ^[1].	Simplified BSD license ^[1]
PyOpenCL	Exposes the full OpenCL API, including every get_info() query and all CL calls ^[2]. Its base layer is written in C++ ^[2]. It turns all OpenCL errors into Python exceptions ^[2]. Its object cleanup is tied to object lifetime, an idiom often called RAII in C++ ^[2].	Free for commercial, academic and private use ^[2]	Not covered by the facts here; check its own pages.	Open-source under the MIT license ^[2]
PyOverdrive	Runs faster versions of supported NumPy functions after one call ^[3]. Sends any input outside the measured range to the standard NumPy ^[3].	Price not published	Installed with pip straight from the GitHub repository ^[3]. Needs Python 3.12 or newer ^[3].	Open source under the MIT License ^[3]. Version 1.0.0, status Beta ^[3].

The short answer

The table below gives each tool's job in plain words, and the facts outside it are limited to the lines cited here. Bottleneck provides ready-made install packages, called binary wheels, for the most common platforms ^[1]. Building it from source needs Python greater than 3.9 and NumPy 1.16.0 or newer ^[1].

On needs, Bottleneck's sources point to ready-made installs and a Simplified BSD license ^[1]. PyOpenCL's sources point to terms that allow commercial use ^[2]. PyOverdrive, covered in its own section, runs faster versions of supported NumPy functions after one call ^[3].

What Bottleneck, PyOpenCL is built for

Bottleneck's own page says it provides binary wheels on PyPI for the most common platforms ^[1]. Those are ready-made install packages. The page also says a source install needs Python greater than 3.9 and NumPy 1.16.0 or newer ^[1]. It is distributed under a Simplified BSD license ^[1]. The table gives the facts its page states about its functions.

PyOpenCL's own page says it is open-source under the MIT license and free for commercial, academic and private use ^[2]. The table gives the facts its page states about its interface.

The facts here do not compare the two tools with each other, so the choice depends on what you need.

What PyOverdrive is built for

PyOverdrive is made by the team that publishes this site.

Calling pyoverdrive.enable() patches supported NumPy functions so your existing code uses the faster versions, and calls it does not support run on the standard NumPy ^[3]. A faster version runs only when a check confirms the input is in its measured range, and anything else, or a faster version that raises an error, goes to the standard NumPy ^[3].

The command python -m pyoverdrive --selfcheck compares every faster version with the standard NumPy on your own machine ^[3]. The command python -m pyoverdrive --calibrate re-times the threaded functions at their own thresholds on your processor and switches off any that do not pay there ^[3]. The call pyoverdrive.explain() tells you which version a call would use and why, without running the call ^[3].

Faster versions of sin, cos, tan, exp, log, log10 and tanh run on large arrays stored in one continuous block of 64-bit or 32-bit numbers, above a size set for each function and data type between 3e5 and 3e6 elements ^[3]. Faster versions of np.linalg.qr exist for large stacks of 3x3 matrices ^[3]. The function np.linalg.eigvalsh has faster versions for batches of 2x2 and 3x3 floating-point matrices above a size threshold ^[3].

Large calls to np.einsum go to NumPy's own optimizer, called with optimize=True, once they pass a measured size threshold ^[3]. np.isin on one-dimensional string or object arrays can use a faster lookup method instead of the default ^[3]. The command python -m pyoverdrive --demo times the main operations with the standard NumPy and with PyOverdrive on your machine, in about 20 seconds ^[3].

Pricing and fees

No price for Bottleneck is published in the facts here. Its distribution terms are a Simplified BSD license ^[1].

PyOpenCL is free for commercial, academic and private use, and it is open-source under the MIT license ^[2].

PyOverdrive is open source under the MIT License ^[3], and its price is not published.

The facts here state no fees beyond the lines above, so check each project's own pages before you rely on any cost detail.

Setting it up and daily use

To install Bottleneck from source, you need Python greater than 3.9 and NumPy 1.16.0 or newer ^[1]. The facts here do not give install steps for PyOpenCL, so check its own pages for setup.

PyOverdrive is installed with pip straight from its GitHub repository ^[3] and needs Python 3.12 or newer ^[3]. The current version is 1.0.0 and its status is Beta ^[3].

For daily use, pyoverdrive.enable() turns on the faster versions ^[3], and pyoverdrive.explain() shows which version a call would use before you run it ^[3]. To turn one faster version off, use disable_path() or the PYOVERDRIVE_DISABLE setting, and pyoverdrive.disable() restores the original NumPy functions ^[3].

Things to check before you choose

Ask which NumPy functions your code calls, and check each tool's own pages for the list it covers, since the facts here name only some of them.

Ask what the tool must run on or connect to, and check each project's own pages for what it supports.

Check your Python version. PyOverdrive needs Python 3.12 or newer ^[3], and a Bottleneck source install needs Python greater than 3.9 ^[1].

Check your platform. PyOverdrive is marked OS independent, and its tests have been run on Windows and on Linux x86-64 ^[3]. Bottleneck's binary wheels cover the most common platforms ^[1], and the facts here do not cover PyOpenCL's platform support.

Check the status. PyOverdrive is labeled Beta ^[3], so weigh that before you depend on it.

Check the license text yourself, since the facts here name each license only briefly.

How to choose

If you want ready-made install packages for the most common platforms, Bottleneck's sources describe those ^[1].

If you want faster versions of supported NumPy functions after a single call, PyOverdrive's claims describe that ^[3].

If you want a check that the faster versions match the standard NumPy on your own machine, PyOverdrive's claims describe a self-check ^[3].

If you want terms that allow commercial use at no charge, PyOpenCL's sources say it is free for commercial use ^[2].

Frequently asked questions

Is PyOpenCL free to use for commercial work?

According to its page, PyOpenCL is open-source under the MIT license and free for commercial, academic and private use ^[2]. The page states this directly. The facts here add no paid tier or fee, so the license text and the project's own pages are the places to confirm the full terms before you rely on them for a commercial product.

Does PyOverdrive require rewriting existing NumPy code?

How do you install Bottleneck and PyOpenCL?

Bottleneck provides binary wheels on PyPI for the most common platforms ^[1]. A source install needs Python greater than 3.9 and NumPy 1.16.0 or newer ^[1]. The facts here give no install steps for PyOpenCL, so its own pages are the place to check. PyOverdrive installs with pip straight from its GitHub repository ^[3].

Can PyOverdrive compare its faster versions with NumPy on the same computer?

Yes. Running python -m pyoverdrive --selfcheck compares every faster version with the standard NumPy on your own machine ^[3]. Running python -m pyoverdrive --demo times the main operations with each version in about 20 seconds ^[3].

What license does Bottleneck use?

Bottleneck is distributed under a Simplified BSD license ^[1]. The facts here do not state a price for Bottleneck, so none is published here. Check the license text and the project's own pages before you rely on its terms.

It is open source under the MIT License. ^[3]

Try PyOverdrive