I was thinking it would be pretty nifty to rig up something like the displays in Brazil using a big Fresnel lens and an old analog oscilloscope tube. The benchmark here is maybe the TX-2 display used to develop Sketchpad and GENESYS in the 1960s: 1024×1024 with a single brightness, a limited range of color (blue initially, yellow afterglow), typically less than 32768 points lit, and I think 100 000 points per second (?).
If we were to drive a 20MHz oscilloscope with the analog VGA signal from a modern video card, we could probably actually do better than that. The video card can produce about 72 million points per second, about 720 times greater than the TX-2, and has memory for a bit over a million of them, about 32 times greater. The tube becomes the limiting factor — if we figure that it can be persuaded to go to an arbitrary spot on the display 40 million times a second, perhaps with a bit of filtering to compensate for the 3dB attenuation and phase shift at that point, that’s one limitation. But that still means we could paint every point on the display 40 times a second — the slow phosphor becomes an even bigger limitation. And it probably doesn’t have 1024×1024 faceplate resolution, maybe more like 512×512 if we’re lucky.
A possibly more entertaining approach is to allocate, say, 100k points to each of many oscilloscopes, and refresh them each at, say, 10 Hz — with a fast enough switch you could multiplex one pixel to each display and thus drive 72 oscilloscopes off a single video card, but it’s probably easier to switch a 20MHz signal to each display for the requisite 2.5 milliseconds out of every 100, limiting you to 40 oscilloscopes but allowing you to use a 4kHz multiplexing switch.
If you wanted to drive them all from a single frame — so you didn’t have to synchronize your page flips with the switch — you could still drive about 10 oscilloscopes.