Ardent Tool of Capitalism

P90 Complex Technical Considerations

(Photo from Peter Wendt/Dennis Smith's MCA card picture collection)

Content Copyright 2003, Jim Shorney, all rights reserved. Edited by Major Tom.

Why the "Y"?

The Pentium 90 IBM "Y" processor complex for PS/2 Model 90/95 computers is the most highly sought after PS/2 upgrade because of it's speed and upgradability. However, some people have experienced unexplained failures of an otherwise very reliable processor complex after upgrading. Why? I believe that some of the failures can be attributed to the power system for the CPU.

The Pentium 90 (actually, Pentium 75 non-MMX and above) CPU is a 3.3 volt CPU. The earlier P60 and P66 CPUs used a single 5 volt power supply. To provide 3.3 volts for the CPU and the I/O section of the 80497 cache controller, IBM added a second voltage regulator to the circuit board. While this regulator is adequate for the original 90 MHz CPU, it is my belief that is needs some rework to handle faster CPUs.

One important factor that many upgraders fail to take into consideration is the power requirement of the new CPU. This is OK for the typical socket 5 or 7 clone motherboard, since (for the most part) they are designed to handle a variety of CPU chips. The Y complex, however, was designed for only the P90 CPU, with no upgrade envisioned. Below is a table of the relevant power specifications for several CPU chips that have been used with the Y. Current specifications are worst-case values, and are taken from relevant Intel documentation; Regulator power dissipation was calculated assuming 1.7 volts drop across the CPU regulator.

Here we can see that even one of the simplest upgrades to implement, the 180 Overdrive, draws over 1 Amp more than the stock CPU, and increases regulator dissipation by over 2 watts! An interposer that provides a separate power source is one solution that is under investigation by others. I think my solution is equally viable, and has the advantage of being cheaper to implement.

The Regulator

The CPU voltage regulator is a three-terminal, low dropout voltage device, the LT1084CT. It has a typical current limit of 5.5 Amps, so it should be capable of handling the upgrade. In fact, this same device is commonly used on Socket 5 and 7 clone motherboards rated for 200/233 MHz CPUs. So where's the problem?

The Heatsink

IBM used two different heatsinks for the CPU regulator. One appears to be a Thermalloy PF430, similar to the Thermalloy PF432G, the other is similar to an Aavid/Thermalloy 576802. They are both very similar in size, and are rated for a thermal resistance of around 25 °C/Watt. I won't go into the thermal calculations here, but this unit seems entirely too small to dump over 7 watts of heat into the air. Also, the PF430 has a design flaw which would seem to make it inadequate for all but the smallest jobs. Look at this picture (all photos are clickable):

The area of concern is circled in red. The tab of the regulator IC rests against the horizontal flat. The raised area below creates an air gap between the regulator tab and the heatsink, which prevents efficient heat transfer to the heatsink. The next pictures show the air gap:

The Aavid design does not suffer from this flaw, as can be seen in this photo:

These sinks were removed from two different P90 complexes. Noticeably absent from both configurations is thermal heatsink compound between the regulator and the sink; further evidence that this was not meant to handle a lot of power.

Making it Better

One obvious solution to the problem is to install a better heatsink on the CPU regulator. I dug through a stack of dead clone motherboards, and found these heatsinks:

They are both from socket 7 75-200 MHz boards that use the LT1084CT regulator, so they should be adequate. I chose the gold one because it has two more fins, and, well, because it's prettier!

Important: The processor complex is a static-sensitive device! When doing any work on it, be sure to use standard precautions. Use a grounded soldering iron, and make sure that you are discharged before touching anything on the board. It is a good idea to get one of those large anti-stat bags that clone motherboards come in, and lay the complex on it while working or any time the board is out of the computer. Your local clone mill should have lots of these bags laying around. You don't need a big-money anti static workstation to do this, just a good dose of common sense.

The stock heatsink has one solder tab in the upper left corner that secures it to the board. Heat this up with a soldering iron and pull the sink up away from the board. It should slip right off the regulator once it is loose from the board (another sign of less than ideal thermal contact...). Pull the mounting pin out of the bottom of the new heatsink with pliers. Carefully stand the regulator straight up from the board by straightening out the legs, and bolt the new heatsink to the regulator tab. Be sure to use a thin coating of heatsink compound between the regulator and the sink. Make sure the sink does not come into physical contact with any other components. There should be enough clearance for everything unless you find a truly monster heatsink, and it should be mechanically stable enough unless you are planning to ship the machine somewhere or launch it on the space shuttle. In this set of photos, the stock P90 CPU is still installed.

The 'turbo-sink' Y complex with Intel Pentium Overdrive, operating at 60/180 MHz, fitted:

Tim Taylor would be proud!

On this complex, I also replaced the Thermalloy heatsink on the LT1085 regulator with the Aavid unit that was removed from the CPU regulator. So we're done, right? Well, not quite...

We need more power, Scotty!

What we have now is a CPU drawing more than 7 amps through one connection point: a plated-through PC board hole that is tied to an inner layer of the circuit board (the +5 Volt power plane). Call me paranoid, but I don't like that. I chose to add two jumper wires from nearby +5v points to the regulator input to spread the current flow out, as shown in the next set of photos. One wire connects to the anode of diode CR4 on the top side of the complex, the other to the positive terminal of capacitor C179 on the bottom. I also made the fan connection for the Overdrive CPU to C179 (not shown in this photo).

A word about the fan connection: you might think that there is enough airflow in the 95 case that you don't need the CPU fan on the Pentium Overdrive. Whether this is true or not, I don't know. But the fan is needed for another reason besides cooling: the POD has a tach sensor circuit built into it that reads a tach signal from the fan (the third electrical contact on the fan). No fan spin = no tach signal, and the CPU downclocks automatically to protect itself. It still runs reasonably fast, and can fool you with some benchmarks, but it's not running as fast as it should.

Was all this worth it? I made a simple temperature probe for my digital multimeter and did some tests. The stock regulator/heatsink combo hit 55 °C at 100% CPU load, while the new heatsink never got above 45 °C. I can't vouch for the accuracy of this crude test, but the results are reasonably close to what I would have expected. All the images on this page were prepared on the machine that is running this modified complex.

Why all this fuss about cooling? Click HERE for one explanation. Since PC Power and Cooling keeps changing the location of that link, here's the pertinent text:

Operating Temperature vs System Reliability

At elevated temperatures a silicon device can fail catastrophically, but even if it doesn't, its electrical characteristics frequently undergo intermittent or permanent changes.

Manufacturers of processors and other computer components specify a maximum operating temperature for their products. Most devices are not certified to function properly beyond 50°C-80°C (122°F-176°F). However, in a loaded PC with standard cooling, operating temperatures can easily exceed the limits. The result can be memory errors, hard disk read-write errors, faulty video, and other problems not commonly recognized as heat related.

The life of an electronic device is directly related to its operating temperature. Each 10°C (18°F) temperature rise reduces component life by 50%*. Conversely, each 10°C (18°F) temperature reduction increases component life by 100%. Therefore, it is recommended that computer components be kept as cool as possible (within an acceptable noise level) for maximum reliability, longevity, and return on investment.

* Based on the Arrhenius equation, which says that time to failure is a function of e^-Ea/kT where Ea = activation energy of the failure mechanism being accelerated, k = Boltzmann's constant, and T = absolute temperature.

Getting to 200 MHz

If you dare, the complex can be overclocked to 66 MHz on the bus, to achieve 200 MHz (or even 233 with an MMX interposer) with Intel CPUs. Here's one way to socket the oscillator module:

I took an 8-pin DIP socket with machined contact pins and bent pins 1-2, 3-4, 5-6, and 7-8 together in a 'V' shape. It's a perfect fit to the foil pads on the circuit board! Half-can oscillators that fit this socket can be purchased from Digi-Key.

9/4/2004 - Some comments on 233 MHz: it's not worth it, in my opinion. Based on empirical results posted to CSIPH newsgroup, 233 MHz is problematical and more often doesn't work well or at all. That extra 33 MHz of speed is insignificant, and more trouble than it is worth to achieve. Here are some comments from Peter Wendt on the subject:

I don't know the Kingston Upgrade too good, have seen it a couple of times, but don't have one for testing. So most of my reply is purely theoretical and based on assumption and experiences I made with different upgrades.

1. In a different thread in this group I'd shared my impression that a bus /core ratio above 3 is a bit tricky to handle for the complex card. Particularly for a 1:3.5 ratio, which *rarely* synchronizes with the 40 MHz DMA and MCA bus controller clock.
In addition the "time window" for synchronization is rather short anyway and gets even shorter if the bus clock and internal CPU speed differ more and more. I only have empirical data gained from weeks of experimenting with the Madex 486007 adapter and a known-good Pentium 233-MMX CPU in an otherwise stable and well-tested system. The number of crashes and bluescreens was significantly higher (about 10 times) than with the Pentium Overdrive 200 MMX or with the upgrade adapter running in 3x mode (200 MHz internally).

2. You should take in account that the power drawn by the processor *may* exceed the onboard 3.3V regulator capabilities - or that of its heatsink. Less likely however, but we should keep an eye on it. The upgrade is usually fed from +5 / +12V from the main power supply directly. Usually this is only the secondary supply for the CPU core - but the rest of the CPU still runs from off the standard 3.3V regulator. I don't know how the Kingston gets the core power. If it has no direct connection to the power supply there might be the reason for frequent crashes: According to the worst case specs from Intel the 233MMX draws 6.5 Amps on the 2.9V core and 0.75 A on the 3.3V supply. Altogether it sums up to over 21 Watts electric power - which is far outside the specs for the LT1084 regulator.

3. Cooling is important. Most likely the Kingston kit has a built-on fan. Does it work ? Is it fed from the main power supply as well ? Or is the Kingston originally intended for Socket-7 boards ("Overdrive Ready") only ? The P90 platform is a Socket-5. Not Overdrive Ready and therefore lacks the two +5V DC fan supply pins. If the CPU fan does not work the 233MMX reacts rather allergic, since it overheats in a few minutes already - and it has no thermal self-protection, other than the Intel Pentium Overdrive 200 for instance.

4. MMX technology and the 82497 cache controller usually work together - other than in P60/66 systems with the 82496. I myself run a machine with the Pentium Overdrive 200 and I did ran one with the Madex and a P-233MMX - but clocked down to safe 200MHz for an extended time with no problems.

5. The change from 60 to 66MHz does not matter (much) to cache controller and cache RAMs, even if they are spec'd as -60 parts. The 10% overclocking (if any) usually causes no problem. None of my machines ever complaint nor crashed more often than usual. At least not significantly. If you run Windows you are used to frequent crashes anyway.

Well - I guess that I can't tell you more at that point. As said: I don't have the Kingston and have no experiences with it.

9/27/2006 - More on the 233 MHz adventure: user reports indicate that BIOS level 08 seems to have the best chance of working with a 233 MHz CPU/interposer combination. BIOS levels higher than 08 are likely to hang with a 0129XXXX error.

Check Sandy's Pentium Interposers and Upgrades page for more info about interposers.

Thanks to Clinton M. (classic PS/2 driver) for digging this info out of the dark recesses of the internet.

To date, no AMD CPU attempts have been successful.