We are still unfortunately in a space where commercial displays are all marred by one problem or another. I bought an 2019 Aorus KD25F 240Hz 0.5ms GtG response time Twisted-Nematic (TN) display and a 2020 Aorus FI27Q-P 165Hz 1ms GtG response time IPS 10bit HDR to test. I was particularly hopeful of the 165Hz display, as a native 10-bit display with fast refresh is exactly what I am looking for. I also use the vision-research specialist Display++ from CRS (10-14bit 120Hz with scanning backlight). This serves as my gold-standard[1] for the time being, it is linearised in hardware.
I have performed photodiode and spectrophotometer tests of both monitors, alongside a standard EIZO IPS 60Hz monitor as a baseline. I use an TSL-251 photodiode (has a built-in amplifier so it converts light directly to voltage) recorded at 2KHz using a LabJack T4, and a Seeeduino Xiao sending a 5ms TTL out after PTB flips[2]. I use a SpectroCal2 spectrophotometer for measuring luminance and color reproduction etc. If you want to drive a 240Hz monitor, you need a reliable and fast PTB machine[3], I’m using Ubuntu 20.10 with AMD WX5100 GPU and 8/16 core i7, and it handles 240Hz just fine.
Main problems with both monitors are all the gaming settings. Even if you try to deactivate everything, the monitor is still doing something to make motion look “better”, or contrast range higher (dynamic contrast adjusts the backlight to make min and max luminance appear greater). You need to really carefully test and choose the best compromise. An excellent site for this is BlurBusters, they provide details about how to build a pursuit camera and test stimulus motion on these higher refresh LCD, where for their usecase they want to get back to the strobed “smoothness” of classic CRTs (Here is their LCD Motion artifact FAQ). They have recently started “certifying” monitors to guarantee good motion response, and the only monitor so far is a Viewsonic 240Hz IPS display (I haven’t tested it). They have user forums, amazing to see so many people so passionate about motion reproduction on modern display technology; I normally assume it is only vision research geeks who are interested in this stuff!
OK, some feedback on these 165Hz and 240Hz displays:
Aorus 165Hz FI27Q-P
First off the 10-bit resolution works great, also kind of works at 11-bit; this is native OS controlled and PTB uses it just fine (tested using the spectrocal2 stepping 1/1024 increments). Therefore these should be excellent for stimuli close to the luminance threshold or that vary in small increments like low-contrast gratings. It supports the DCI-P3 color space, so better than sRGB, though I didn’t try to measure this myself.
At 165Hz, perceptually I would say it “looks” better subjectively for motion smear that the Display++ at 120Hz. But there are still some motion artifacts. Using the photodiode it has about a 20ms onset lag and can take longer than the inter-frame interval to fall back to the background luminance (leading to some potential ghosting or motion smear).
Linearising using a SpectroCal2 works fine.
One major flaw (even with the latest firmware), is that even with dynamic contrast disabled, going from black as a stimulus increases luminance the backlight still switches a bit. On a grey background this is not an issue, but it is important to keep in mind.
Aorus 240Hz KD25F
The 240Hz display has the best latency and lag of any monitor I’ve tested, basically lag is around 5ms and rise-fall time is around 5ms, within the inter-frame interval. I assume this is not real 8-bit, like most TN displays, but the step test looks fine. One thing I notice at 240Hz is a small ripple at 1Khz, probably some controlling backlight or voltage control to keep the fast timing. Subjectively motion looks smooth, with less motion smear.
Linearising using a spectrocal2 is fine.
Summary
I would use these displays for general subject training and standard tasks that don’t require either very precise motion or luminance (i.e. a working memory tsk showing pictures etc). For electrophysiology, you really need to make sure you measure lag carefully and account for it in your analysis (lag can vary depending on the start->end luminance values). This is the great advantage of the Display++ or ViewPixx, they provide TTL output locked precisely to actual display onset. Having said that, the 240Hz display lag is as close as I’ve seen any LCD and may be fine for some experimental designs.
If luminance/contrast or color matters without too many motion artifacts, I think the FI27Q-P is worth considering. If determinstic lag and fast response and minimal motion artifacts are more important, then the faster KD25F must be used.
It is still worth considering writing grants and buying a dedicated visual display from CRS or VPixx, as commercial LCD displays still come with a lot of issues you have to carefully calibrate yourself to deal with.
I can upload some plots and data files if anyone wants to look at them.
[1] Specwise I imagine the overall vision-research gold standard display is VPixx’s ProPixx — but it costs 10x more than the Display++
[2] There could be a small delay from my TTL as the Xiao is USB-based, but when I’ve measured the fastest rate of communication from MATLAB is is ~0.5ms, so I think it is not much of an issue and is pretty deterministically sub-millisecond timewise.
[3] I can dual-boot windows, but on the same hardware the timing reliability is really much better under Linux.