- Short version
- Operating systems
- Additional software
- Basic hardware requirements
- Graphics hardware requirements
GNU Linux is strongly recommended. Also works on Microsoft Windows 10 (Intel), but will be much less well supported by the developers and may have restrictions in functionality, reliability, performance and precision for some features, compared to Psychtoolbox running on a modern Linux distribution. Especially multi-display setups and HiDPI Retina displays under Windows are highly problematic! Use on Apple macOS 10 is possible but strongly discouraged if you need any kind of reliable timing for visual stimulus presentation or precisely calibrated visual output, or use of special visual stimulators, e.g., from VPixx or CRS, or trustworthy visual stimulation at all, due to the large and growing number of bugs in the Apple operating system. macOS 11 is completely untested and not officially supported yet.
64-Bit Matlab version R2022a or later (older versions will likely work, but are no longer tested for compatibility or supported by us in case of trouble), and GNU Octave version 5.2 or later versions on Linux, and 64-Bit Octave version 6.4.0 on macOS and on MS-Windows.
Recommended are OpenGL 2.1 (or ideally better!) capable GPUs from AMD. Modern Intel graphics chips will also work well on Linux, not on Windows for more simple tasks with lower need for performance.
AMD GPUs are strongly recommended over NVidia GPUs on Linux and macOS, as they allow use of high quality open-source graphics drivers on Linux, and of additional useful PTB features for vision science on Linux (and also to some degree on macOS, for old AMD GPUs older than AMD Vega). On Linux, please do not install the AMD proprietary amdgpu-pro driver, but simply stick to the high qualiy amdgpu driver which is already installed on any new Linux installation (batteries included!), iow. if you don’t do anything, you’ll do the right thing.
The most well tested AMD models are currently from the AMD “Polaris” gpu family, with the most exhaustive set of low-level features supported by Psychtoolbox on Linux. AMD “Vega” family gpu’s should provide the same quality and features, and additionally have improved FreeSync support, if you need it. They are expected to work well with Psychtoolbox on Linux, according to testing by one lab so far. The latest generation “Navi” gpu’s seem to also work well, according to some user reports, but are not yet tested by the developers due to lack of hardware, and they do lack support for some special, but rarely needed, Psychtoolbox low-level debug features at the moment. So the sweet spot would be “Polaris”, or “Vega” if you intend to use FreeSync for interesting visual stimulation paradigms that require Psychtoolbox VRR functionality. AMD “Raven Ridge” gpu’s, integrated into AMD Ryzen processors, are also well tested and known to work under Linux and Windows-10. As “Raven Ridge” uses a “Vega” graphics core at its heart, this is another indicator that “Vega” should work fine.
Only choose NVidia if you absolutely must for some very good reason, e.g., if you must use CUDA during visual stimulation. Note that we won’t be able to help you much with problems caused by NVidia gpu’s even on Linux, as we can’t debug or fix their proprietary graphics drivers in any meaningful way!
Modern Intel HD-Series graphics cards should work reliably for medium complexity workloads on Linux. Current OpenGL-3/4 GPUs may expose additional useful features. The absolute minimum requirement is OpenGL-1.2 support, but functionality and performance will be very limited with such old GPUs, and using graphics cards with only functionality older than OpenGL-2.1 has not been tested by the developers in years.
Try to avoid PC laptops with hybrid-graphics, e.g., with NVidia Optimus or AMD Enduro technology! Under MS-Windows the visual onset timing of the high performance gpu will very likely be wrong on most such laptops, and there is no known way to fix this with software workarounds, so you might be restricted to the integrated low performance gpu. At least Intel integrated gpus are often very buggy under MS-Windows, so you would be out of luck completely! The behavior on Linux is generally better: Many Laptops will work reliably with the high performance gpu and the open-source graphics drivers when combined with an Intel integrated gpu, e.g., models with older NVidia graphics cards and most AMD graphics cards. The most recent gpu models from NVidia are more difficult to set up and somewhat rather inflexible to use, due to limitations of their proprietary graphics drivers, restricting such Optimus setups to single-display visual stimulation only if the proprietary graphics drivers are in use. This also only works with X-Server 1.19, which is outdated and only works on older distributions like Ubuntu 18.04.0 LTS, but not with the current X-Server 1.20 and, e.g., recent Ubuntu 18.04.5 LTS, or the recommended Ubuntu 20.04-LTS or later. The combination of AMD integrated gpu + AMD discrete gpu should work fine for modern AMD integrated gpu’s found in AMD Ryzen 3000 series processors or later, ie. more modern than AMD RavenRidge. Combining AMD integrated gpu’s with other discrete gpus, e.g., from NVidia or Intel currently doesn’t work well at all due to software limitations. At least the lower performance integrated AMD graphics chips should still work as a reliable fallback in such a scenarios on Linux though. Read “help HybridGraphics”, which provides up to date info.
On Linux and macOS, any card supported by the operating system should work well, as well as built-in sound chips. On MS-Windows 7 to 10, sound cards should work reasonably well with Psychtoolbox 3.0.15 and later, but Windows 10 should provide enhanced precision and lower latency over Windows 7. On MS-Windows with Psychtoolbox version 3.0.14 and earlier, no precise or low latency sound output is possible anymore. Click this link to find out why.
GNU/Linux is fully supported on Intel compatible PCs and suitable (== older) Apple Macintosh computers under
- Matlab 64-bit, version R2022a. Older versions likely work, but are no longer tested or testable by us due to lack of access.
- GNU Octave 64-bit, versions 5.2, 6.1 - 6.4.
- Additionally, Psychtoolbox from NeuroDebian supports 32-Bit and 64-Bit releases of whatever version of GNU Octave ships by default with your distribution, e.g., also Octave 3.2 to 4.4 on older distributions, or Octave 5.x, 6.x and 7.x on recent distributions.
Psychtoolbox testing and development occurs mainly on the most recent Ubuntu Linux LTS releases or flavors of them, currently Ubuntu 20.04.4-LTS. Distributions older than Ubuntu 20.04-LTS are no longer supported since Psychtoolbox 3.0.18.
NeuroDebian tests and supports PTB also on Debian GNU/Linux. According to user reports, Psychtoolbox seems to work reasonably well on Linux Mint, Arch Linux, Gentoo, and Fedora. We can’t provide much support on other distros than Ubuntu LTS flavors due to lack of time and resources.
It is recommended to stick to the latest long-term support (LTS) Ubuntu release, currently 20.04.4-LTS, if you want the most well tested and well supported setup.
The NeuroDebian project is an effort to provide convenient access to neuroscience-related software on the Debian and Ubuntu Linux distributions. The NeuroDebian APT archives include a Psychtoolbox snapshot release, which has been packaged to be ready-to-use with GNU Octave.
NeuroDebian curates some open-source neuroscience packages for Matlab as well, and provides two sets of packages, one for Octave (“octave-psychtoolbox-3”) and one for Matlab (“matlab-psychtoolbox-3”).
For details on these options, see the Linux installation instructions.
We generally strongly recommend Linux as the operating system of choice for demanding experimental setups, which require the highest timing precision, precision for color or luminance displays, general performance and flexibility. Our support for fixing bugs and other issues on other operating systems than Linux will be very limited. We strongly discourage use of Apple macOS for anything but entertainment and education purposes.
Apple macOS - Not recommended!
Psychtoolbox is also being developed and tested under Apple macOS 10. Psychtoolbox should “work” in principle on
- 64-bit macOS 10.15.7 “Catalina”, maybe macOS 10.14 (untested), maybe partially macOS 10.11-10.13 (untested).
- 64-bit Matlab R2022a, likely older recent versions upwards of R2014b, but those are untested.
- 64-bit Octave v6.4, probably also v7.1/v7.2 (expected to work, but untested).
- As of 2022, you will need to get Octave 6.4+ from a package manager like HomeBrew as standalone binary installers are not available yet.
Limited testing currently only happens on the latest version of macOS 10.15.7 “macOS Catalina” with 64-Bit Octave 6.4, and with 64-bit Matlab R2022a.
These are the only somewhat supported versions of macOS at this point in time. Psychtoolbox is compatible with Catalina in principle, and Catalina is the only currently tested system, but Catalina has fantastic new bugs and flaws, e.g., a slow-down of keyboard input by a factor of 5x, and various trouble wrt. keyboard input, sound input, and video capture, thanks to Catalina’s awful new security design – Prepare for lots of hassle if you choose Catalina, you have been warned! macOS is the most buggy and hazardous operating system you could use for visual stimulation, or DAQ digital/analog i/o, so running real data collection using macOS will likely bring you a world of pain (and possibly irreproducible research).
macOS Mojave was possibly a better working choice, although it is no longer tested by us and unknown if it is still fully compatible with Psychtoolbox 3.0.18. The fact that it is no longer developed or supported by Apple means that Apple will no longer break it with respect to our needs, but it is now an attractive target for viruses and hackers, a security hazard! Choose your poison…
macOS 11 Big Sur or macOS 12 Monterey are not officially supported or tested at all at the moment! macOS 11/12 on machines with Apple’s new ARM based SoC’s, e.g., the Apple M1 SoC, are not working with Psychtoolbox natively at all in any suitable way for data collection. There is no timeline for fixing this anytime soon. It is unclear if this is fixable at all, given the challenging and disruptive changes Apple made on Apple Silicon, especially switching from industry standard AMD, Intel and NVidia graphics to Apples own proprietary graphics chip. But also due to the disappointing lack of financial support from most of our users, we con’t have the funds to investigate this any further at the moment. It is possible to run Ubuntu Linux for ARM in a Virtual machine on Apple M1 for training and education purpose, according to a user report, ie. not for data collection, only for learning on Linux. It is also possible to somewhat run Psychtoolbox on macOS 11/12, but with completely broken visual stimulation timing. May be good enough for basic training purposes though. See this link for reference about the current state.
Toolbox version 3.0.14 and later releases do not work under macOS 10.10
or earlier anymore. The last working version on macOS 10.10 was v3.0.13.
The last version that worked on macOS 10.9 and 10.8 was v3.0.12. The last
version that worked on macOS 10.7 and 10.6 was v3.0.11. The last version
that worked on macOS 10.5 and 10.4 and also with 32-Bit versions of Matlab
was v3.0.10. Those old and unsupported versions of Psychtoolbox can be
downloaded by specifying the special
DownloadPsychtoolbox downloader script.
The PowerPC platform is no longer supported by the version 3.0.10 Psychtoolbox. If you need to use a PowerPC machine, stick to version 3.0.9.
In general, only operating system versions 10.4 “Tiger” and 10.6 “Snow
Leopard” seemed to be relatively unproblematic in use, i.e., most of the
many operating system bugs we found now have workarounds implemented in
Psychtoolbox. 10.5 “Leopard” was a rather buggy operating system,
especially for multi-display stimulus presentation and stereoscopic
stimulus presentation. 10.7 “Lion” and later mostly has restrictions in
the precision for visual stimulus timestamping due to various bugs in the
operating system. Please install the Psychtoolbox kernel driver to help
resolve at least some of these issues on AMD and NVidia graphics cards
(see PsychtoolboxKernelDriver or
>> help PsychtoolboxKernelDriver).
Psychtoolbox should work on
Matlab 64-bit. Currently tested and supported with release R2022a. External mandatory requirements: Microsoft C MSVC 2015-2019 runtime and GStreamer 1.18.5 MSVC. Installation of GStreamer before installation of Psychtoolbox is mandatory on Matlab or the Screen mex file will not work.
GNU Octave 6.4.0, 64-Bit. Installation of GStreamer before installation of Psychtoolbox is mandatory on GNU Octave or the Screen mex file will not work. The current download location for official Octave-6.4.0 64-Bit is reached by clicking this link.
If you choose to use Matlab, you may need to install Microsoft Visual C runtime
libraries to make it work, specifically
vcredist_x64_2015-2019.exe. The installer should
give you instructions on how to do that if necessary, ie. on install failure. The
Psychtoolbox/PsychContributed/ subfolder contains the neccessary
executable for your convenience in such a case.
Psychtoolbox-3.0.17 is no longer officially supported for Windows-7 / 8 / 8.1. The current expectation is that it still mostly works on these systems, but just as with v3.0.16, only Windows-10 releases from 2021 (21H2 edition specifically) are tested and supported in case of trouble, and some functionality like sound output should be better on Windows-10, other functions like HDR High Dynamic Range display support will only work on recent Windows-10.
Psychtoolbox-3.0.16 no longer works on Microsoft Windows XP, should continue to work on Windows Vista and later, but just as with v3.0.15, only Windows-10 is tested and supported in case of trouble.
Psychtoolbox-3.0.15 still worked under Microsoft Windows XP, Windows Vista, Windows 7 and Windows 8/8.1, but we don’t actively test for compatibility with any system but Windows 10 and won’t provide any bug fixes or troubleshooting help for any issues that can’t be shown to be also present on Windows 10. Specifically, moving away from Windows XP, Vista and Windows 8/8.1 is strongly recommended. Windows 7 should continue to work without major problems at this point in time, but we don’t test for this anymore. For best audio support, Windows 10 is strongly recommended.
We do aim to keep the toolbox working under these and future versions of Windows, but full support for all features is a lower priority for us than Linux. Without the kind of access to the source code and use of an open-source development process like on Linux, it is also simply not possible to make improvements to the operating system itself if needed, or to fix various bugs or even diagnose bugs for development of workarounds. It is hit and miss…
We cannot recommend Window at for dual-display stereo stimulus presentation, HiDPI setups, or for tasks with a need for high visual timing precision.
Generally we recommend switching to a modern version of Linux, e.g., a flavor of Ubuntu 20.04-LTS.
Multimedia engine: GStreamer 1.18.5 required on Windows and macOS
Installation of GStreamer version 1.18.5 is mandatory for movie playback, movie recording, video capture and video recording. Multimedia functions won’t work on macOS without GStreamer 1.18.5 being installed. Neither will high quality text rendering work, unless you use Octave instead of Matlab. If you want to use Psychtoolbox for visual stimulation on Windows you will have to install GStreamer first, even if you do not need any multimedia functions, or Psychtoolbox won’t work. GStreamer 1.18.5 MSVC variant is needed, earlier versions or MinGW variants will not work on Windows.
On Linux you also need GStreamer, but GStreamer is a de-facto standard component that ships with all modern Linux distributions. GStreamer 1.8 should work, GStreamer 1.16 will work better, and GStreamer 1.18 is required for full convenient support for HDR movie playback, whereas GStreamer 1.16 will need some “hand-holding” by user scripts for more limited HDR playback support.
See GStreamer (or
>> help GStreamer) for installation
instructions for the different systems.
Microsoft Kinect support on Linux and macOS: libfreenect-0.5 required
For use of your PsychKinect / PsychKinectCore driver on Linux or macOS,
at least version 0.5 of libfreenect is needed. See
>> help InstallKinect
for instructions. Kinect support on macOS is expected to still work, but deprecated.
Basic hardware requirements
Intel PCs: Any Intel-compatible PC that is capable of running the 64-bit versions of Microsoft Windows-10, or the 64-bit versions of GNU/Linux Ubuntu 20.04 LTS or later.
Intel Macs: Not recommended! Any Intel-based Macintosh computer that is capable of running 64-Bit macOS 10.11 “El Capitan” or later macOS 10 versions, or a 64 Bit GNU/Linux distribution. However: At this point in time, most NVidia graphics cards have broken visual stimulation timing under macOS. Most AMD graphics cards under macOS 10.12 and later have broken visual stimulation timing under macOS, and Intel graphics chips also seem to have trouble under at least macOS 10.13 “High Sierra”. If you install Psychtoolbox 3.0.16 or later, these timing problems will be worked around at least for standard precision 8 bit per color framebuffers on many gpu + display combinations. Higher precision framebuffers are still unfixably broken, and some machines may still have problems, e.g., the new MacBookPro 2019 16 inch. So for visual stimulation there essentially doesn’t exist any supported Apple hardware that would work acceptably under macOS in all configurations.
Old hardware may perform fine under Linux. Apple MacBook’s or MacBookPro’s from the year 2016 or later are known to be mostly unusable with Linux for practical purposes, as basic things like wifi, suspend/resume or audio won’t work. Any current Apple hardware with Apple’s T2 security processor will not even allow accessing the internal disc drive - and thereby won’t allow installation of Linux on the internal drive. For these reasons we don’t recommend use of any modern Apple hardware.
Apple ARM Macs, e.g., Apple M1: Broken! Do not use!
RaspberryPi models 2B, 3, 4, 400 under the most recent Raspbian operating system. The RaspberryPi 2B and 400 are actively tested for compatibility and works well for not too demanding visual and auditory stimulation tasks, USB i/o and digitial i/o via the programmable GPIO pins. The Pi model 3 is not tested, but expected to work just as well as the model 2B. Pi models 4 and 400 are substantial improvements over the older models in hardware capabilities.
Graphics hardware requirements
Basic Psychtoolbox functions should work in theory on any OpenGL 1.2 capable graphics card with at least 16 MB of video ram (VRAM), with the mentioned operating system specific restrictions in mind, ie., somewhat broken on Windows multi-display and broken on macOS for non-trivial use cases. Fast stimulus drawing and use of the more advanced features requires recent graphics hardware. However, we do strongly recommend at least OpenGL 2.1 capable graphics hardware for full functionality and good performance. Older graphics cards are no longer tested.
For optimal performance and functionality on Linux we recommend AMD or Intel graphics over NVidia graphics, due to the high quality open-source drivers for AMD and Intel graphics. For advanced functionality like fine-grained timing via FreeSync or Displayport adaptive sync, choose AMD. Ditto for HDR display support.
In general, you should not try to skimp on the GPU, as performance of your stimulus script and the types of visual stimuli you can create with ease will depend much more on the horse power and features of your GPU than on the performance of your CPU.
The recent generation of integrated Intel HD graphics cards, e.g., Intel HD 2000, HD 3000 etc., as found in many modern “Intel Core” processors, provide decent functionality, accuracy and performance for not too demanding tasks on Linux. Use on other operating systems than Linux will usually go much less well. These cards are OpenGL-3 / Direct3D-10 compliant. Numeric precision is on par with recent NVidia or AMD cards for most (but not all) accuracy tests that have been executed on a Intel HD card under Linux. Absolute graphics performance is of course significantly lower than that of current discrete NVidia or AMD cards. But for not too demanding visual stimulation paradigms, these cards are now somewhat suitable.
Older Intel graphics cards are problematic for all but the most trivial visual stimulation tasks: While the Intel GMA X3100 series cards and similar are also Direct3D 10 compliant in theory, in practice they suffer from limitations. Users of Intel-based Macs should be aware that some Macs (e.g., old Intel Mac Book) use a built-in Intel GMA graphics adaptor. The GPUs of the GMA-950 series are known to have very low graphics performance and a very restricted feature set. They are cheap and sub-standard by any definition. See this Wikipedia article for further information.
Products from Matrox, Via and S3 or from other niche vendors are not recommended. As Matrox and S3 seem to have mostly retreated from the 3D graphics market, most of their products are not a good choice for OpenGL based applications like Psychtoolbox. Even the products that nominally claim to support hardware accelerated OpenGL, have a pretty limited feature set and performance.
For dual-display work (e.g., binocular stereo stimulation), we strongly recommend using dual-head or multi-head graphics adapters (i.e., one card with two or more output connectors) instead of multiple separate adapters. We expect dual/multi-head single-card performance to be higher and the likelihood of graphics driver bugs to be lower. While separate cards may work, we do not guarantee this and do not provide any support for troubleshooting. Note that stereo work may benefit from the display synchrony provided by some of the dual-head cards. Synchrony is usually hard to achieve with separate cards. We also recommend to avoid Mac macOS and MS-Windows for dual display real-time stimulus presentation. Apple seems to be mostly incapable of or uninterested in implementing decent support for high performance, tear-free dual display support. For static stimuli or use as a control monitor, macOS may be good enough. MS-Windows has equally severe trouble with multi-display visual stimulation, and often even for single-display stimulation on a multi-display setup.