on my ga-586hx 2.01/k6-233/matrox millennium setup, i am normally using 2 32mb 60ns edo hyundai sticks. these are already enough to run almost everything at fastest timings, that is: ras# precharge time 3, r/w leadoff 6/5, ras# to cas# delay 2, read burst 333/444, write burst 222, turbo read leadoff enabled, speculative leadoff enabled, cache timing fastest. with this, i get 106.97 mb/s in speedsys (yes, k6 scores for this are really low) and 64.6 fps in pcpbench vgamode. so the only thing it won't do is read burst at 222/333, it simply hangs at post.

i decided to test some other memory and found 2 lgs 16mb edos. these would run with that setting at 222/333 but to my surprise this actually resulted in degraded performance instead of outright freezing, the numbers are 78.25 mb/s in speedsys and 59.5 fps in pcpbench. so clearly it's not stable and would probably just crash quickly when trying to run windows. some other sticks with the same hyundai parts as above but only 16mb capacity each would just hang again, but curiously they would give about 1mb/s more in speedsys upon reverting to 333/444, when compared to the 32mb sticks.

there is an entry here that runs the fastest timings (#424), but it doesn't specify what ram was used. is this timing typically inaccessible for 60ns rated edo, or is there maybe some ic/datecode that's more likely to run it? i'm sure pcb design and capacitors must play some role here, for instance my 32mb hyundai sticks only have capacitors under the ics on one side for some reason. finally, i know about the 50ns parts and would definitely expect those to do the trick, but they tend to be a bit expensive to just get 64mb of them.