First post, by raymangold
This is somewhat of a follow up to an older thread I did about the Kingston TC400. Finally purchased a PowerLeap PL-K6-III; which is vastly superior over the Kingston interposer: allowing for a greater range of voltages (as low as 1.30v !) And a few extra gimmicky features with isolated power and a clock throttle to turn down the multiplier if the CPU gets too hot.
Here are some pictures of it:
The performance increase is most definitely an extra 50% when all things are considered (games that were previously sluggishly unplayable are now playable). Not only that, but I have my full 384 MB of RAM now with no performance penalties--once the write allocation is enabled that is. The 300PL's BIOS does not automatically switch to it of its own accord. Turning off the cache in the 300PL's BIOS just turns off *all* cache, rendering the system absolutely insanely slow (Windows 98 was operating as if it was on a cacheless 286). It doesn't seem to matter when the old cache is left on, once write allocation is forced the memory is addressed through the K6-III.
Note: It could be different depending on your BIOS. According to Jan Steunebrink, many Socket 7s don't even natively work with the K6-2+ / K6-III.
The results have been added to Phil's Ultimate VGA Benchmark google excel sheet.
For a rough reference, the Kingston TC400 scored 42.9 FPS in Quake, whereas the PowerLeap PL-K6-III gets 47.7 FPS. When the K6-III Write Allocation is not enabled (where the system runs on its old Intel cache tag) it gets the same score as the K6-2 on the TC400; fairly impressive since going over by 320 MB makes the machine a *lot* slower.
Final installation (crap capacitors were replaced with rubycons and a "panny"... I must say the vias on these powerleap interposers are *very* irritating to work with since the diameter is unusually shallow, requiring the capacitor leads to be heated up as they're inserted...):
Here are the three 128 MB EDO DIMM modules installed (the memory ICs are visibly different than the ones on SDRAM modules):
I could force this old socket 7 system farther:
#1 Modify the FSB frequency to 75 Mhz instead of 66 Mhz | allowing 450 Mhz for the CPU clock, the faster bus would definitely increase performance
#2 Attempt to find ECC 256 EDO DIMMs (unbuffered) and see if the system can be forced to take 768 MB (assuming the system would be willing to accept memory beyond 512 MB before the write allocation is forced).
#3 Build a RAM interposer to reveal the 'fourth' hidden DIMM slot for 512 MB: which would be four 128 MB sticks
Going beyond this would require some hardware hacking and BIOS editing. These computers aren't exactly all that common anymore (the less featureful Pentium II counterparts are, but they suck). So due to lack of spare motherboards I'll forefeit for now. There are also different revisions of the board with components moved slightly around. I am not sure what the different revisions change...
One nice thing about the motherboards is that they're 'tray loaded', pull this latch and the mobo slides out from the rails (which keeps it from touching the chassis for proper isolation):
PS: I know vetz owns one of these interposers, not sure if he has experimented with it yet.