Psychtoolbox 3.0.22.3 “The most fantastic Matala” was released on 14th August 2026.

You can download it as a zip file from the following link:

https://github.com/Psychtoolbox-3/Psychtoolbox-3/releases/download/3.0.22.3/3.0.22.3.zip

The complete development history can be found in our GitHub repository. The release tag is “3.0.22.3”, with the full tree and commit logs under the URL:

https://github.com/Psychtoolbox-3/Psychtoolbox-3/tree/3.0.22.3

Compatibility changes wrt. Psychtoolbox 3.0.22.2:

• Be reminded that macOS 26 Tahoe is currently not recommended for use with Psychtoolbox, due to severe Tahoe operating system performance bugs, caused by Apple and only fixable by Apple. These bugs will affect any software that tries to achieve robust, reliable, trustworthy and accurate visual stimulus presentation timing and timestamping. It causes framerates to be cut in half! Only software with broken timing support will not suffer this degradation, iow. to my knowledge all other toolkits in existence. This is the situation as of macOS 26.0 - 26.2, the status of current macOS 26 and later is unknown to myself. The same applies to upcoming macOS 27 Golden Gate.

Highlights:

• None. Mostly lots of bug fixes and smaller feature work and refinements. Most of the work done in the last months is “behind the scenes” work, preparing for the future, but not yet visible to users in the form of new features or obvious improvements, beyond what is listed here.

All:

• Improvements to OpenXR handtracking for VR applications. We can now make use of the XR_FB_HAND_TRACKING_AIM extension on VR devices that support it, to detect, and report properties of, “aim” gestures made by users.

• Procedural noise: Update (classic) Perlin noise to latest version. The functionality is demonstrated by ProceduralNoiseDemo(), now better!

  There were some changes in the upstream shaders circa 2021 that appear to visibly improve randomness/reduce banding artifacts. There may also be a minor performance improvement because some math was simplified, but may be system-dependent (I only saw 10s of microseconds changes). Contributed by Alex Forrence.

• GetGITInfo: Make simpler and more robust. Contributed by Alex Forrence.

• VBLSyncTest: Suppress beamposition plots on beamposition-less systems harder.

• Some improvements to the PsychColorimetric functions by David Brainard, as usually poorly documented in his non-conformant commit messages, so I won’t give details. Except: “Added CarrollNeitzNomogram and test.” and “Moved in routines from BrainardLabToolbox.” Other new functions seem to be TransmissionToOD, ODToTransmission (OD ~ Optical density?), “StockmanRiderNomogram” and “StockmanRiderNomogramDemo”, and “PhotopigmentNomogramDemo”.

• Various other smaller refinements, bug fixes and documentation updates. Various internal work in the software, not obviously user visible yet.

Linux:

• Psychtoolbox was built against Matlab R2026a and Octave 6.4 under Ubuntu 22.04.5-LTS. It only works on Ubuntu 22.04-LTS and later.

• RaspberryPi OS 13 is now usable with Psychtoolbox again, after my bug fixes for graphics driver bugs were accepted and integrated into it. RaspberryPi OS 12 suffered the same severe performance bugs, but the bug fixes were not applied to it, as the RaspberryPi foundation considers OS version 12 end-of-life. This leaves OS 11 and 13 as the supported versions for Psychtoolbox, but I will probably soon remove support for OS 11 as well. So upgrading to RaspberryPi OS 13 is recommended.

• Add a hack to allow basic HDR-10 mode on RaspberryPi 4 and later, with a very custom software stack: Mesa zink OpenGL driver from Mesa 25.1 or earlier, e.g., Mesa 25.0.7 from RPi OS 13, combined with a Mesa v3dv Vulkan driver from Mesa 25.3.0 or later. Also supports very slow HDR movie playback at ~5 fps for 4k HDR-10 content. Given the complexity of the setup needed, this is more of a “can pigs fly?” proof of concept at the moment, not something that just works out of the box plug & play.

• PsychOpenHMDVR: Fix it by fixing abuse of ‘clear Screen’ with a new method. Turns out this was broken quite a while. The fact I didn’t notice earlier, and nobody else complained, suggests there aren’t any users of this driver. Not surprising, given the auto-selected default driver is OpenXR, which is superior in almost any sense. We should probably just retire this driver.

• MultiWindowVulkanTest: Refine further for more stress testing.

Windows:

• Psychtoolbox was built against Octave 7.3 under Windows 10.

• Due to a lapse in our Matlab licenses, the mex files have only been updated for Octave, whereas the files for Matlab stay the same as in Psychtoolbox 3.0.22.2.

• PsychStartup: Fix GStreamer SDK root definition on MS-Windows better. Use fullfile() for building final path to GStreamer bin directory, based on sdkroot.

  Apparently, for unknown reasons, on some MS-Windows systems, the getenv() call returns a sdkroot without a trailing path separator backslash, whereas on most systems it does. Go figure! Lacking a trailing backslash caused the naive [sdkroot ‘bin’] concatenation to generate an invalid path. fullfile() deals with this quirk.

  Root of the problem found out, and fix proposed, by our GitHub user @HuwSwan in GitHub issue #669 , thanks!

macOS:

• Psychtoolbox was built against Octave 10.2.0 from HomeBrew on macOS 13.7.8 Ventura final update for Intel Macs, and on macOS 15.7.7 Sequoia with Octave 11.1.0 for Apple Silicon Macs. The minimum required version of macOS for use of the Vulkan display backend, ie. for HDR support on Intel Macs, is now macOS 11.

• Due to a lapse in our Matlab licenses, the mex files have only been updated for Octave, whereas the files for Matlab stay the same as in Psychtoolbox 3.0.22.2.

• Eyelink mex for 64-Bit Octave on macOS for IntelMacs fixed, and reverted to a over 18 months older version for Matlab. Turns out that Eyelink support was broken since over 1.5 years on macOS for IntelMacs without anybody noticing or complaining! The brokeness was due to IntelMac compatibility bugs in SR-Researchs older Eyelink SDK versions used for building the mex file. The bug is fixed in recent SDK’s.

  This gives an indication about how few users with demanding use cases are left on Intel based Macs. Our statistics suggest that less than 7% of all paying macOS users are still using IntelMacs. Both the drastically reduced number of users, and the increasing difficulty in maintaining macOS Intel versions, with all software vendors reducing or cancelling support, means that the days of Psychtoolbox on Apple IntelMacs will probably be limited. Apple’s next macOS 27 will no longer support IntelMacs, Matlab R2026a and later don’t support them, and Homebrew as provider of Octave also is announcing a phase out likely towards end of this year.

Enjoy!

Psychtoolbox 3.0.22.2 “The long long slog” was released on 2nd March 2026.

You can download it as a zip file from the following link:

https://github.com/Psychtoolbox-3/Psychtoolbox-3/releases/download/3.0.22.2/3.0.22.2.zip

The complete development history can be found in our GitHub repository. The release tag is “3.0.22.2”, with the full tree and commit logs under the URL:

https://github.com/Psychtoolbox-3/Psychtoolbox-3/tree/3.0.22.2

Compatibility changes wrt. Psychtoolbox 3.0.22.1:

• Now requires Octave 10 or later on macOS. Octave 9 and earlier will no longer work on macOS.

• HDR support on Intel Macs (via Vulkan display mode) now requires at least macOS 11, ideally at least macOS 13 is recommended for better results.

• Be reminded that macOS 26 Tahoe is currently not recommended for use with Psychtoolbox, due to severe Tahoe operating system performance bugs, caused by Apple and only fixable by Apple. These bugs will affect any software that tries to achieve robust, reliable, trustworthy and accurate visual stimulus presentation timing and timestamping. It causes framerates to be cut in half! Only software with broken timing support will not suffer this degradation, iow. to my knowledge all other toolkits in existence. This is the situation as of macOS 26.0 - 26.2, the status of macOS 26.3 and later is unknown to myself.

Highlights:

• None. Mostly lots of bug fixes and smaller feature work and refinements. Most of the work done in the last six months is “behind the scenes” work, preparing for the future, but not yet visible to users in the form of new features or obvious improvements, beyond what is listed here.

All:

• Eyelink: Fix for use with external display backends, most importantly the Vulkan display backed used for macOS on Apple Silicon Macs. This by rewriting PsychEyelinkDispatchCallback.m to abstain from using nested helper functions, and also some refactoring. Nested helper functions don’t play well with the Psychtoolbox imaging pipeline - which is always active on Apple Silicon macOS - due to some insane limitation that Matlab imposes onto the parent function of nested functions. Long story short: Avoid nested functions whenever possible for a happier life, and especially don’t use them in functions calling Screen('Flip')!

  Other parts of Eyelink toolbox and its demos have been refactored to remove use of nested functions in various demos.

• VideoCaptureDemo: Increase windowed size for modern cameras. For a better non-fullscreen viewing experience.

• VBLSyncTest: Add optional measurement of OpenGL gpu processing time. Calling VBLSyncTest([],[],[],[],[],[],[],[],[],1); will plot OpenGL per-frame processing time, excluding potential further processing by display backends like Vulkan, VR, etc.

• Vulkan backend: Improved debug output and other help messages. Various bug fixes and refinements. Some important bug fixes.

• HDR support: Skip initial HDR meta data setup during startup when not needed. On a conformant Vulkan driver, it should not ever be needed. Therefore only set it if native display HDR static metadata is available, as queried from Vulkan driver or OS, or if it is needed to workaround bugs in the Vulkan ICD. Currently the only driver known to require this to get HDR mode and HDR signalling going is AMD’s proprietary AMDVLK-Pro driver on Windows, and their now deprecated AMDVLK open-source driver for Linux. For these AMD provided ICD’s, we call the function always, but on these drivers we should get valid native display HDR metadata, so it should be fine.

• PredictVisualOnsetForTime(): Only query what you need. To avoid some warning message clutter on some system configurations with old graphics hardware.

• PsychLicenseHandling(): Refine some user messages and help text wrt. offline activation.

• Various other smaller refinements, bug fixes and documentation updates.

Linux:

• Psychtoolbox was built against Matlab R2025b and Octave 6.4 under Ubuntu 22.04.5-LTS. It only works on Ubuntu 22.04-LTS and later.

• Upgraded / Rebuilt the mex files for 32-Bit Octave on RaspberryPi OS 11 for 32-Bit ARM RaspberryPi machines. Those were left at v3.0.20, and are now also at v3.0.22. Other minor fixes for RaspberryPi. Should work on RaspberryPi OS 11 and 12. In principle, it also works on the latest RaspberryPi OS 13 based on Debian GNU Linux 13 “Trixie”, but in practice I strongly recommend against using it on RaspberryPi OS 13, as the graphics drivers in that OS version seem to have serious performance problems, leading to much lower framerates for stimulus presentation!

• Screen(): Improve troubleshooting help for Matlab library compatibility problems.

• PsychVulkanCore: Now allow use of Mesa version 25.3 and later Vulkan drivers for deep color modes on AMD gpu’s, 10 bpc, fp16 (~11 bpc) and 16 bpc (= At most 12 bpc on current AMD graphics cards in reality) modes. So far this required the now deprecated / unmaintainted AMDVLK driver to be manually downloaded and installed. This only works for single-display stimulation with Vulkan so far. For multi-display you’d still need AMD’s deprecated AMDVLK driver.

• PlayMoviesDemo: Use auto-selected pixelFormat on 64-bit ARM Linux.

• PsychLinuxConfiguration: Fix fallback symlinking to libglut.so.3.12 on Ubuntu 24.04-LTS. If users did not install sudo apt build-dep psychtoolbox-3, then this did not work. Reported by Github user @qx1147, thanks!

Windows:

• Psychtoolbox was built against Matlab R2025b and Octave 7.3 under Windows 10.

• Mark Windows 10 and Windows 11 as supported, all older versions as unsupported. This was quite outdated and slipped through the cracks. Note that development and testing still happens almost only on Windows 10, as my only Windows 11 machine is unsuitable for proper testing of visual stimulation, because it uses a known to be broken Intel integrated graphics chip with totally broken visual stimulation timing. Light testing of other features than visual stimulus presentation timing didn’t show apparent signs of trouble.

  My current Windows 10 machine can’t be upgraded to Windows 11, because Microsoft doesn’t allow it, and I don’t have the financial resources atm. to buy a suitable machine for Windows 11 testing. However, over 75% of all paying users of Psychtoolbox 3.0.20 and later already use Windows 11, less than 25% still use Windows 10, and one single machine still uses Windows 7! The lack of user feedback about problems with Windows 11 suggests that current Psychtoolbox should work fine, or at least as well as on Windows 10, on Windows 11 with suitable hardware and especially graphics hardware + drivers.

• Implement workarounds against the latest bugs in NVidia’s recent Vulkan drivers, which caused the Vulkan stimulus window to be hidden behind a static Psychtoolbox welcome screen, making it unusable, or at least neccessary for users to manually ALT+TAB the stimulus window into the foreground.

  Yet another NVidia problem: HDR doesn’t work by default anymore. One now needs to change some display driver setting in the NVidia control panels “Manage 3D settings” section. Specifically, the “Vulkan/OpenGL present method” must be set to “Prefer layered on DXGI Swapchain”, as the “Automatic” setting is now broken.

  This restores HDR on NVidia - ofc. only until the next time they break it with a new set of bugs… Also, as a new side-effect, it may break OpenGL based stimulus presentation timestamping, the other settings work better for OpenGL. This problem was observed on Windows 10 22H2 with a NVidia GeForce GTX 1650 Turing family gpu and NVidia driver version WHQL 591.74, Vulkan driver version 591.296.0.

macOS:

• Psychtoolbox was built and tested against Matlab R2025b and Octave 10.2.0 from HomeBrew on macOS 13.7.8 Ventura final update for Intel Macs, and on macOS 15.5 Sequoia with R2025b and Octave 10.3.0 for Apple Silicon Macs. The minimum required version of macOS for use of the Vulkan display backend, ie. for HDR support on Intel Macs, is now macOS 11.

  macOS 15 is now marked as somewhat supported and tested, at least on Apple Silicon. Current state of things:

  macOS 13 Ventura: Tested on Intel MacBookPro 2017 with Ventura 13.7.8 final update. No other macOS versions can be tested anymore on Intel Macs in the future.

  macOS 14 Sonoma: Tested on Silicon MacBookPro 2023 with M2 Pro once. No further testing will be done with macOS 14 in the future.

  macOS 15 Sequoia: Tested on Silicon MacBookAir 2020 with M1 once, and ongoing testing with MacBookPro 2023 with M2 Pro.

  The only actively tested macOS versions are now macOS 13 Ventura for Intel Macs, and macOS 15 Sequoia for Apple Silicon Macs. I strongly recommend not upgrading to macOS 26 Tahoe at this point, as there are known substantial performance regressions for visual stimulus presentation, due to severe macOS Tahoe operating system flaws which are not fixed by Apple at least as of macOS 26.2. The status on 26.3 and later is unknown at this point.

• Eyelink: Fix for use with macOS on Apple Silicon Macs.

• Switch Vulkan display backend from MoltenVK 1.2 to latest upstream MoltenVK 1.4.2 with some additional fixes by myself. While I hoped this would fix performance problems with macOS 26 Tahoe by increasing efficieny, those hopes were in vain for Apple Silicon Macs. It was worth a shot though. This raises the minimum required macOS version for use of the Vulkan backend to macOS 11, iow. no change for Apple Silicon, higher requirement for Intel Macs, but there it only affects HDR display support.

  Also add some tweaks to improve stability / reduce hangs. This is especially important on Apple Silicon Macs. The reliability of stimulus presentation still isn’t on par with Intel Macs yet. Substantially more work is needed. Right now, any Screen('Flip', win, twhen) with tWhen more than 1 second into the future, or in general pauses between flips of more than 1 second still cause stutter or timeout warnings.

• Screen: Fix “keypress makes annoying beep” suppression for fullscreen windows on the secondary (screen number > 0) screen on multi-display setups with Matlab R2025a and later on Intel Macs, and with all Matlab and Octave versions in terminal only mode. This beep suppression already worked on single display Intel Macs, or for use of screen 0 (the main screen) on multi-display setups, but failed on other screens of a multi-display Intel Mac.

• Fix SetMouse() and RemapMouse() on Retina displays in Retina compatibility mode.

• Fix panel fitter (e.g., PanelFitterDemo.m) on Apple Silicon Macs with Retina display. It is still somewhat broken on Intel Macs, except if opting to use the Vulkan display backend. Ie. the default high timing precision opaque fullscreen window display mode has somewhat broken panel fitting by default, e.g., in PanelFitterDemo.m. The problem is structural and unlikely to get fixed in the remaining lifetime of Psychtoolbox for Intel Macs. A way to use the panel fitter on such Retina Intel Macs is to manually specify the optional ‘fbOverrideRect’ parameter to PsychImaging(‘OpenWindow’, …) to be a rectangle the true size/resolution of the Retina display panel, to help Screen.

Enjoy!

• I am now on vacation throughout December and first half of January 2026, so user support is suspended until the end of my vacation.

• I strongly advise users of Apple Silicon Macs against upgrading to macOS 26 Tahoe until further notice. Follow this link to our forum discussion thread for background and updates

Merry christmas and a happy new year!

-mario