Bug report is FB7799650. Also, discovered issue only occurs on discrete GPU, works as expected on integrated graphics.

Click on the user icon (looks like a person) in the upper right, then click on your username (which doesn't look like a link).

Since I can't attach a screen shot here nor provide a link to a screenshot on Flickr, I've opened a feedback with screen shot attached. FB7846981

Thanks for the response. The device is a AppleTV 4K (Model A1842). For anyone else coming across this, the performance limiters are supported on A11 and newer CPUs according to https://developer.apple.com/documentation/metal/optimizing_performance_with_the_gpu_counters_instrument

Thanks. Would you expect the buttons on a PS4 Dualshock controller to be usable as switches? I paired one to the AppleTV, but "Activate your external switch" doesn't detect any button presses.

I picked up a BT keyboard and it's working for Switch testing. I've also submitted feedback FB8481108 suggesting game controller buttons to be usable as switches. Thanks.

Submitted as Feedback FB8545183

I started getting this error in an app. Eventually tracked down to indexing beyond the end of a buffer passed into a fragment shader, due to an off-by-1 error.

Still seeing it with GM XC 12. Doesn't seem to cause a problem. No SwiftUI in my app.

I picked up a "FREEGENE HDMI Dummy Plug Virtual Screen Display Emulator for Headless PC 4096x2160@60Hz, 3 Pack Headless Ghost". And it works well. When I need to see what is on the screen, I'm able to use QuickTime Player "Movie Recording" as a realtime monitor without much negative impact on my app's performance.

One more note. When initially using the Headless Plug the Apple TV Video mode was 2160p60Hz SDR . When I changed it to 2160p60Hz HDR the Apple TV interface became unstable (focus movement erratic, list of available video modes being repainted every few seconds with focus moved to bottom entry). And it was very difficult (many attempts to quickly move through the settings before they repainted) to switch back to SDR. Once reset to 2160p60Hz SDR the interface was stable again.

To be a little more specific, the App's display fragment shader execution time reported by GPU frame capture increased for 2.5ms to 3.5ms for a 4K display.

I submitted a feature request for this in 2016. Was radar 28675140. And was closed as being a duplicate of another radar, so someone else had requested it before me. Maybe if we could get everyone who's posted a similar question in the forums to request it, it'll get some action.

Have you looked at the performance information from Capture GPU Frame? Unfortunately, I don't think you get the fine grained stats on Intel/Discrete that are available on A10X and newer SOCs. If you expect the code to live into the Apple Silicon era, exploring performance on a modern iPad might not be a waste of time. Also, would OpenCL be using both CPU and GPU while Metal is only using GPU?

I verified is still issue in Xcode 12.3 and submitted as FB8972110