Old IBM 300 GL won't power up - somebdy save my 455 !!!1!!1

[anaqer]

OK, so I finally got around to get me an old comp for a DOS/Win98 box - except it won't power up. Actually, I have two different models of 300GL, a P100/16MB and a P133/32MB with two different power supplies too, but no matter how I juggle things around, I can't get either of them to work. Fans don't power up, chips are not getting warm, no sign of life whatsoever.

It could be that both of the PSUs are dead, but there's some oddity about them too - they seem to be of the AT standard (2x 6 pin main power connector for the mobo), but they have two additional +3.3V connectors, one for the mobo and one for the PCI/ISA riser card. Also, the power switch doesn't connect directly to the AC mains, but to the mobo just like ATX, so what I don't understand is, how the PSU is even supposed to know it should be on?

 

[anaqer]

OK, turns out the PSU was dead indeed. But at least I've found out that it's just ATX after all, with AT-style power connectors, so I'll probably be able to convert a spare PSU to work with the system.

New question (hope this one will be easier) - does the model 6282-380 support MMX processors? I could only recognize a single VRM on the motherboard, does this rule out split-voltage capability altogether, or is there hope yet...? If it's 3.3V only, what are the chances of an MMX processor operating at this, approx. 20% higher core voltage (assuming serious air cooling) ?

Here is a photo to identify the board (I can look up part numbers too if you tell me which stickers I should look for):
[ - Fig 1 - ]

 

[mczak]

New question (hope this one will be easier) - does the model 6282-380 support MMX processors? I could only recognize a single VRM on the motherboard, does this rule out split-voltage capability altogether, or is there hope yet...?

Looks like a single-voltage board to me. I believe all boards offering split-voltage have some voltage-adjustment jumpers. You could use a multimeter to figure out, get the pentium mmx datasheet, and look if i/o voltage pins and vcore voltage pins are directly connected.

If it's 3.3V only, what are the chances of an MMX processor operating at this, approx. 20% higher core voltage (assuming serious air cooling) ?

People have done this (on other boards), should work (don't know how long the processor survives this of course). However, it can put serious stress on the voltage regulator - if it's a switching regulator, chances are quite good, if not, the regulator might just burn. Judging by the small heatsink, I tend to believe it's a switching regulator, it doesn't look like that small piece of metal would be large enough for cooling even in case of an old pentium 133. Though the board could theoretically directly use the 3.3V supplied by the PSU (I've never seen that AT/ATX mix) for cpu voltage, in that case no vrm would be needed at all for vcore (you could also figure this out with the multimeter)...
If you'd want to use a 233MMX, you'd have to jumper it for 3/2 multiplier, which gets remapped to 7/2.
Also, the bios might not like pentium mmx cpus at all, no idea.

 

[anaqer]

Thanks for your input - I've whipped out my DVM but it didn't really clear things up, or at least I can't interpret the results. I measured between VCC2 and VCC3 pins (conveniently located on opposing sides of the socket) and it showed about 4 kOhms initially, which climbed to about 16.5 kOhms in a matter of seconds. :? Weird, but at least it didn't show ~0 Ohm so I still have my hopes.

I'm probably grasping at straws here, but what makes me wonder is the fact that the riser card uses one of the +3.3V rails from the PSU directly (ie. withouth any sort of regulation) - how much greater tolerance is there for the PCI bus on 3.3V than on the CPU? Because if the difference in tolerance is negligible, we could assume that the CPU I/O voltage comes also directly from the PSU (and is regulated right there based upon a remote sense lead coming from the second 3.3V connector near the L2 cache slot). Thus, it might be possible that the VRM with the tiny heatsink does only 3.3V to 2.8V conversion. (The lack of voltage configuration switches I'm willing to pin on the idea that IBM offered the model 6282 with just Pentium Classic and Pentium MMX (see this chart) and didn't bother with support for CPUs which used anything other than 3.3V/2.8V .)

I'll report back tomorrow when the new PSU for the main rig arrives and I can refit the current one with these AT connectors.

EDIT - I just checked the data sheet of the FET, and it's a MONSTER! It's an International Rectifier FZ34N hexfet that would probably burn a hole into the mobo before it fails, so adding some extra cooling to it should be enough. Theoretically, that is.

 

[mczak]

Thanks for your input - I've whipped out my DVM but it didn't really clear things up, or at least I can't interpret the results. I measured between VCC2 and VCC3 pins (conveniently located on opposing sides of the socket) and it show about 4 kOhms initially, which climbed to about 16.5 kOhms in a matter of seconds. :? Weird, but at least it didn't show ~0 Ohm so I still have my hopes.

If Vcc2 and Vcc3 are not directly connected, that's a pretty good indication that this board indeed has split voltage support - the "classic" old pentium boards really had Vcc2 and Vcc3 just connected (it's normal that you measure some resistance and that it changes somewhat when measuring).

(The lack of voltage configuration switches I'm willing to pin on the idea that IBM offered the model 6282 with just Pentium Classic and Pentium MMX (see this chart) and didn't bother with support for CPUs which used anything other than 3.3V/2.8V .)

Yes that's possible. Actually I think that not all boards with split-voltage regulator might have the adjustment jumpers, that might only apply to retail boards, but not those OEM boards in complete systems.

Anyway, if it turns out to be a single voltage design after all, I'll have to slap a bigger heatsink on the FET and pray.
I'll report back tomorrow when the new PSU for the main rig arrives and I can refit the current one with these AT connectors.

I think chances are pretty good this works just fine - since the same box was available with pentium mmx, I'd bet all use the same motherboard.

 

[anaqer]

Well then. The PSU surgery has worked just fine so far - power levels are nominal across the board, the system is running and seems to be stable. MMX compatibility testing has to wait until the CPU arrives, until then I'm stuck with a bloody multiplier locked 133MHz.

What's worse is that the machine won't recognize any of the two extra SDRAMs I got (32MB and 64MB), it'll only work with the factory default 16MB stick. (Which, you no doubt are all aware of, is a major pain in the back when it comes to running anything but Solitaire.) Did they change something between the SDRAM modules the 430VX chipset could handle and those that were used in 440 LX/BX/... systems?

EDIT : Piccie of the modules, original at the top. (Each of the three has 8 chips.)

 

[mczak]

Well then. The PSU surgery has worked just fine so far - power levels are nominal across the board, the system is running and seems to be stable. MMX compatibility testing has to wait until the CPU arrives, until then I'm stuck with a bloody multiplier locked 133MHz.

Multiplier locked? Sure? I'm fairly certain the first multiplier locked intel cpu was some PII.

What's worse is that the machine won't recognize any of the two extra SDRAMs I got (32MB and 64MB), it'll only work with the factory default 16MB stick. (Which, you no doubt are all aware of, is a major pain in the back when it comes to running anything but Solitaire.) Did they change something between the SDRAM modules the 430VX chipset could handle and those that were used in 440 LX/BX/... systems?

sdram did not really change, however all chipset have certain limitations (mostly maximum supported "chip density"). With the 430VX, these limitations seem to be _extremely_ bad, a quick googling returnted this page: http://www.plasma-online.de/english/help/solutions/vx_chipset_sdram.html (you could try getting the 430vx datasheet at developer.intel.com, but I don't know if it is still available, and furthermore the site seems to be down at the moment). But it looks like it will only accept modules built with 16mbit chips (maximum), and your 32MB and 64MB modules are built with 32mbit and 64mbit chips respectively.

 

[anaqer]

Yeah, meanwhile I found out about the RAM density limitation of the chipset (not at intel though, they've killed off all info on pre-TX stuff). Damn IBM, they didn't even bother with EDO sockets, so I'll have to track down some of these ancient moduls somehow. :? Time to go visit some office storerooms where they might have stuff like this, piled up in a corner under six layers of dust. (Although astonishingly, the 32MB stick kinda works if it's used together with the original one - it's recognized as 8MB though, amounting to a whopping 24 megs of system RAM. )

I'm pretty sure about the multiplier lock, at least it won't respond to the jumper settings - even 2.5*50MHz is broken (gives 100MHz) and the 1.5 multiplier seems to have no effect either. Google turned up this site and sure enough, it's a SY022 chip.

Oh well... it screams in DOS and later Win based games will just have to wait.
I'll bump once I've dug up those arcane upgrade parts and done some silly benchmarking.

 

[mczak]

I'm pretty sure about the multiplier lock, at least it won't respond to the jumper settings - even 2.5*50MHz is broken (gives 100MHz) and the 1.5 multiplier seems to have no effect either. Google turned up this site and sure enough, it's a SY022 chip.

Interesting. Looks like intel played around with locking earlier than I thought. Looks like some P-133, P-166MMX and P-200MMX are locked. Looks like intel just did not wire one of the BF pins on the cpu. Though the information I've found indicates 1.5 multiplier should work correctly, and 2.5 multiplier should give the same results as 1.5. Who knows though what intel did on different chips, since not all SY022 chips appear to be locked...
And since your board doesn't support higher bus speeds (probably the clock chip doesn't support them), you're only chance of overclocking would be exchanging the reference crystal .

mczak

 

[anaqer]

And since your board doesn't support higher bus speeds (probably the clock chip doesn't support them), you're only chance of overclocking would be exchanging the reference crystal .

Well, since I've got two motherboards but need only one, I'm inclined to go a bit radical and do some soldiering and stuff - except I'm having a hard time digging up data sheets on the clockgen. The only thing I've found is this one, but I'm not sure how much of a difference there is between revision -01H and -04H.

Do the numbers "3077 / 72" on the top of the crystal say anything useful about it (google pulled a blank), do I need an oscilloscope to find it out, or can I just assume it's the common 14.318 MHz variety? What formula do I need to use to find out what the new crystal should be like to get around 80MHz instead of 66MHz, and does a crystal swap effect, say, the USB clock frequency?

Sorry for the probably stupid questions, I've never done this before and it's hard to find info on this kind of hardware hacking...

 

[mczak]

Do the numbers "3077 / 72" on the top of the crystal say anything useful about it (google pulled a blank), do I need an oscilloscope to find it out, or can I just assume it's the common 14.318 MHz variety? What formula do I need to use to find out what the new crystal should be like to get around 80MHz instead of 66MHz, and does a crystal swap effect, say, the USB clock frequency?

I believe all common boards use the same 14.318 Mhz reference crystal (don't quote me on that...). And I think you could just assume linear scaling (so something like 17Mhz would do for 80Mhz FSB). Not sure which clocks are affected, but you can expect problems with some peripherals (pci/ide will certainly be affected, no idea about other i/o).

Sorry for the probably stupid questions, I've never done this before and it's hard to find info on this kind of hardware hacking...

Neither have I - I'm not really into hard-core overclocking

 

[anaqer]

Just wanted to report that I got my 233MMX and it's working just fine after all.
The tiny VRM heatsink does get real warm when the case is open (ie. in no airflow situations) but nothing too serious.
I also got another 16MB stick making it a total of 32MB RAM, but it was rather hard to find such antiquated PC66 memory.
Between these upgrades and the Diamond Monster 3D II, Quake 2 and even Half Life is quite playable, w00t!