That was a failure.
See, a few days ago, I bought myself something that seemed real nifty – a Matrox Mystique 220 with memory expansion add-on. Just imagine, an 8MB MGA! Think of the possibilities! (well, mostly, playing games in 1024×768 really slowly)
So it arrived. At first it looked real nice, but then I noticed something amiss: the expansion module is too big, for starters. Also, it had what looked like a second MGA chip on it. And finally, the PC still only saw it as a 4MB card. A quick googling for the MYST/RRSTI code revealed the truth: it’s actually a Rainbow Runner Video add-on card, essentially an AVI and MPEG1 encoder/decoder. Useful, I guess, but not for my purposes.
Oh well, at least I got a Mystique 220 out of it, right? Well, that was something of a disappointment too. The chip is an MGA-1164 (compare to my old Mystique, which was MGA-1064), and compatibility overall is actually a lot lower. Most tests and games will freeze at random, if they even start at all. The only solution I could find was disabling Bus Mastering, but that will reduce performance severely.
Apparently, with bus mastering, the card can start working on polygons while the CPU is still busy. Or maybe the opposite? Anyway, Turok has absurd slowdowns, 3DMark is a lot slower, sound in Forsaken seems slowed down, and overall stuff doesn’t run as well. At least it didn’t crash anymore, but at that point I might have as well used the older Mystique instead. Surprisingly, Quake even shows that the card doesn’t support the same 256 colors resolutions as its older brethren: nothing higher than 640×480 is shown.
So yeah, a big disappointment overall. I need to check my sources better. Can’t really fault the seller, because he probably didn’t know what he was actually selling. A lesson for me, I guess. The next card coming, and possibly the last one (I know I always say it…) is a SiS 6326 AGP. Not sure how much memory, but the memory chips make it look like a 4MB model. Strange, since I believed 8MB ones would be more common when it comes to the AGP model. Either way, it should be better than this mistake here.