While professional series gpu’s generally don’t have advantages over the cheaper consumer models, maybe with rare exceptions like use of frame-sequential stereo goggles, that advice of VPixx is somewhat non-sensical. Graphics and display drivers between same gpu families of consumer and pro gpus usually share > 90% of code (and potential bugs), and the hardware is mostly identical, as you pay extra money for support, validation and extended warranties or availability of replacement parts, or maybe some features like ECC VRAM that may be useful for scientific gpu computing/number crunching.
And in general, whole families/generations of gpu models share the same features and driver/hardware bugs, not specific models. Consumer class gpu’s are bought more frequently, so in that sense are better beta-tested by the masses of consumers, except for the special niche pro use cases for which Pro gpu’s provide extra validation by the vendor for the extra money. Niches like computer aided design CAD/CAM or common pro 3D modelling (3DS Max, Maya, etc.) software, certainly not vision science / neuroscience.
That’s why I currently recommend against NVidia gpu’s, especially on Linux, because their proprietary drivers are undebuggable/unfixable by people like myself, so less help can be provided in case of trouble. And various special open-source features contributed by myself for vision science are missing from their proprietary drivers.
I can btw. confirm the brokeness of the NVidia GeForce GTX 1650 in Vpixx list, at least with Ubuntu 20.04.5-LTS current NVidia proprietary drivers. This is the only NVidia i have for testing atm. and it doesn’t work for identity pixel passthrough.
The advice against AMD Radeon WX 4100 is probably outdated. It is a Polaris gpu with DCE-11.2 display engine, which had broken pixel passthrough for a short period of time under Linux until i diagnosed and fixed it. The bugfix is in all currently shipping Linux distros since at least Ubuntu 20.04. In fact, Polaris class gpu’s are among the most well-tested gpu’s with PTB on Linux atm.
In the past I was better able to pro-actively test and fix such things at least on Linux, but the severe lack of funding on our side makes extensive testing a luxury now that we can no longer afford, not even on Linux which is generally and still by far the highest quality and easiest to handle OS for neuroscience applications, as long as most of our users don’t give much of a damn about the quality of their research software tools, or at least don’t express that in money contributions.
-mario