To add a little more commentary on why this is particularly impressive, they're more or less using the stock hardware, with the only "modifications" being some

To add a little more commentary on why this is particularly impressive, they're more or less using the stock hardware, with the only "modifications" being some custom peripherals. But those peripherals still have to move data through the I/O ports of the original machine. The pouet comment by one of the authors says that it isn't even able to read data off of the IDE hard drive at 16kB/frame @ 30 fps (480kB/sec) because it's so slow. On top of it, COVOX audio device requires the CPU to generate the sound and pulse it out of the parallel port where it acts as a DAC. Here it's generating a 22.725Khz stereo signal along with the hard drive I/O along with the computation required to decode the video stream and display it. This machine is ~.5MIPS and is a clone of a PDP-11. In demoscene conventions, this actually is a controversial production because it's technically a "video" and demoscene productions need to generate their display and run in real time. But there's lots of argument that the ability to harness this machine in the way they did is computationally impressive enough that there's some debate over it. However there are some categories of demoscene works that also allow for these kinds of productions and the debate is really over how it was competed more than what it is. For comparison, here's what games on the BK normally look like https://www.youtube.com/watch?v=-IOre6AAcGc https://www.youtube.com/watch?v=-IOre6AAcGc https://www.youtube.com/watch?v=X7zVVeirSdI&t=161s https://www.youtube.com/watch?v=X7zVVeirSdI&t=161s