Hey Packard Bell enjoyers. Got a mystery for you guys.
I recently performed the full gamut of upgrades on my Legend 204CD with the PB450 motherboard. My system is now voltage modded (image attached) using a method similar to Hoping's method posted above: Re: Packard bell PB450 repair / mods / upgrade. I have also added 512K of cache, 32MB RAM, the Micro Firmware BIOS, etc.
I put in an AMD 5x86-133 ADW, and have had mostly positive results. Except for one strange issue. A small number of programs will cause the keyboard to reset constantly, about twice per second (LEDs flashing). So far this is most noticeable in Descent: If you get into the game and then pause with Esc, while on the "Quit Game" prompt the LEDs will flash. This makes it very difficult to get back out of the menu since the keyboard responds poorly. In addition, sometimes in-game the keys will get "stuck" in the down position or the game will lock up. The only two other places I have observed this so far is the Blood setup program (and only when doing the sound effects test), and the installer for XCOM: UFO Defense (keyboard LEDs flashing during disk copy). It's mostly been a nuisance up to this point but now I've also run into some minor file system corruption across multiple disks now, which is suspicious.
I performed the following tests/swaps to rule things out:
- removed all cards
- memory test
- cache test + tested with cache removed
- power supply swap to a modern ATX w/ adapter
- slightly increased voltage to the 5x86 to 3.5v
- switched FSB from 33mhz to 25mhz
That basically left the voltage mod as suspect, so I bypassed the mod and tried the original DX2-66 @ 5v with Descent since it shows the most obvious issues. No problems. Then to test the voltage mod, I tried running it at 3.3V (which it works fine at)... also no problems. Hmmm. So I pulled out as many 486's as I could find in storage and tested them all with interesting results.
- Intel 486DX2-66 (original CPU for this system, not sure of S-Spec due to glued on heatsink) - no problems
- AMD Am486DX4-100 (SV8B, 25544 package code) - problems
- AMD Am486DX4-120 (SV8B, 25398 package code, running at 100mhz since this board doesn't do 40mhz FSB) - no problems
- AMD Am5x86-133 ADW (25544 package code) - problems
- IBM/Cyrix 5x86C-100HF - no problems
So the issue is seemingly limited to AMD CPUs based on the 5x86, which use the 25544 package code. The big difference between those and the DX4-120 that does work is that the 25544s are a later 350nm package with 16KB L1 according to the info here. The thing is, though, that plenty of people have run 5x86s on this board, both through voltage mods and standard upgrade kits with regulators, and I've seen no mention of problems. I even checked Usenet archives to see if anyone reported anything back in the day. What exactly is going on here?
I'm using the IBM 5x86C for now since it seems to be the fastest CPU that isn't presenting issues, but I would love to know what in particular is different about the AMD 5x86 that is causing it to act up. If anyone has more CPUs to test maybe we can see if my findings are a coincidence, or localized to my specific machine.
Also a side mystery, can anyone help me figure out what VRAM I need to do the VRAM upgrade? Per the manual, it wants "two 256x16 (70ns) DRAM ZIP chips with (2) CAS lines (symmetrical)". I bought some NEC chips I thought would be compatible (NEC D424170V-70), but when installed the system then only detects 512K of VRAM instead of the default 1MB or the upgraded 2MB. Very strange. I'm guessing they don't fit the "symmetrical CAS lines" requirement since it seems oddly specific, but I couldn't find enough info to tell. I am fairly sure the chips are not defective, as I have 10 of them and I find it hard to believe that all 10 would be bad.
The OPL Archive - Preserving MS-DOS music in a unified format!