MacX VC supports Nvidia hardware acceleration. Intel Power Gadget shows that HD530 is also involved, but it can be OpenCL, the frequency of GT is the frequency of the 3D core. Intel Power Gadget does not have monitoring of another HD530 core, which is called RENDER. Documentation and utilities are available in the Intel Media SDK. But these utilities are only for windows and linux. Quicktime when capturing the screen uses Quick sync, in full screen mode. Install iStat monitor, it shows which video card and how it is loaded during the encoding-decoding operation. Even if this card is not displayed in the system report. Try to record the screen and see the CPU load. It should be very small, up to 10 percent. In this case, see the processor of which video card is loaded. Quicktime can also use Nvidia's hardware coding.
In addition, the FCPX uses 2 services. One is used to decode VDADecoderXPCServ, the second one is for encoding VDAEncoderXPCServ. They are used separately by the program. That is, the encoding and decoding are separated. This is logical. Skylake can either encode or decode via Quick sync, but not simultaneously. If it encodes via Quick sync, before it sends the stream to the hardware blocks, it decodes it from the original format, and it's not done through Quick sync. There is another way. Decode via Quick sync, and not encode through it. That is vice versa. But not both ways at the same time. And the second is slower in 2 times.
You can find out exactly if you have a test file 4K mov with a high bit rate for a couple of minutes of video. It is necessary to transcode it to h264 in different programs (only in one pass, 2 passes Quick sync does not support) and watch the metrics. The encoding time will be very different, depending on whether Quick sync, or OpenCL, or Metal is used.
To check the decoders you need a h264 4K file with a high bitrate. Quicktime with Quick sync playing such a file shows a very low CPU load, and without the Quick sync decoder, the CPU usage is several times higher.