Upgrading the P90 Complex

Madex 586007

Back by demand, by those with the "If it ain't broke, you're not trying hard enough" attitude.

WARNING! Before you do this, read Dr. Jim's P90 Upgrade Information! In summary, the Y complex was designed for the power draw of a P90 cpu. An Overdrive Pentium draws more! (Table stolen from Dr. Jim)
CPU Speed
Current (A)
Dissipation (W)
180 OD
200 OD

  Also, look at Peter Wendt's P90 Platform Tuning interesting stuff on clock multiplier hacking!

Pentium® Overdrive 180 Mhz upgrade from 90 Mhz Pentium® processor complex in IBM PC Server 500 S/390

   I would like to thank Aron Eisenpress for sending me this bit of trivia. Aron is not responsible (nor am I) if you screw up your P90 complex doing this. Please read the disclaimers before proceeding. This modification voids any warantee or service contract that you have left for your PS/2 8590 / 8595 /9590 / 9595, or Server 500. 

Modifying your processor complex with these instructions will void your warranty and may cause irreparable damage resulting in a non-functioning processor complex if performed incorrectly. There are no warranties expressed or implied. Modify at your own risk. (Manufacturers warranty on newly purchased PC Server 500's was usually three years depending on country of purchase. Check your warranty information if concerned or unsure. Optional service contracts may have altered or extended your period.) 

The following instructions can assist you in modifying the 90Mhz Pentium processor complex that was originally manufactured in the IBM PC Server 500 System390 so that you can use the Pentium® Overdrive 180Mhz with MMX chip. If you unplug the Pentium® 90Mhz chip from the complex and plug in the 180Mhz Pentium® Overdrive without this modification, you will experience two problems: 

I. The electric fan that cools the Pentium® Overdrive chip will not have power to turn the blades. You will 'cook" (burn out) the Pentium® Overdrive due to inadequate cooling. (The 90Mhz Pentium® chip uses a heat sink instead of an electric fan.) 
2. Without providing 5 volts to the overdrive chip, the chip will run at about 25Mhz which is slower than the installed 90Mhz chip. 

* Locate and have a working reference diskette for the PC Server 500 available. You will need it if you use a different processor complex board than the one currently installed. You should not need it if you remove, modify, and replace the same processor complex board that is currently installed. Better to locate and have it ready than to not have it. 
* The processor complex does not use a ZIF (zero insertion force) socket for the Pentium® processor. The old 90Mhz processor must be carefully removed. (A good quality jewler's screwdriver was useful to the author in starting to pry the chip from the socket and then working around the perimeter of the chip increasing in small increments. Insert the tip of the screwdriver between the socket and the chip. Use only the socket surface and not other chips or edges on the board as fulcrums for prying.) 
*Solder an insulated wire from the +5 volt regulator to the two unused pins on the Pentium® 'internally no-connect'. 
* The two pins are technically labeled AN01 and AN03. There are no labels visible that say AN01 or AN03. 
After the wire has been soldered to all three points and allowed to cool, carefully align all pins of the new processor to their respective holes taking note of the key (missing pin on one corner). Carefully push the new processor into the holes. The new processor pins may be longer than those of the old processor and therefore there may be a gap between the bottom of the new processor and the socket surface. 
* Reinsert the processor complex into the PC Server and reconnect the signal wire. Power on the server and assure that the fan on the Pentium Overdrive processor is turning (a flashlight aimed at the fan when power is applied will probably be necessary). 
* You will note that the LED display on the front of the PC Server 500 will say "60Mhz" instead of "90Mhz." This is normal. 

Bottom side of 90 Mhz complex IBM FRU P/N 06H3739 "CUBRUN" 

Modify at your own risk! Modification will void warranty. 

Photo of back of complex showing modification. 'Ihere is no significance to the loop in the wire toward the left. The wire could have been shortened to eliminate the loop. 

Closeup of the connection to the 5 volt regulator 

Closeup of connection to pins on the Pentium® connectors 

There is no significance to the loop in the wire toward the left. The wire could have been shortened to eliminate the loop. 

Trademarks Pentium® is a registered trademark of Intel Corporation. 

From Tam Thi Pham (24 May 99) 
   Hi Louis. I tried out Aron's P90 to P180 conversion over the weekend on two complexes, and discovered that it doesn't seem to be a reliable / stable mod. For the most part the system *seems* to work fine, but when I ran extensive diagnostics on the machine with Aron's mod the system would occasionally lock-up. At first, I thought that maybe it was just a sampling error of some sort, so I modded a different complex to confirm the results. Sure enough, exact same symptoms. Testing was done on a machine that had been running my P200 mod reliably. 

(Ed. I've seen stock Pentium class complexes fail diags. But I have a P complex with an FDIV cpu that passes. Aron and Tam are beating this to death) 

From Aron Eisenpress
I dunno... it's actually not my mod originally (I did it here for one of the guys who has 2 Server 500's) but I've not heard of any problems with it.  The mod only adds the electrically connection to run the fan on the Overdrive cpu, and I don't know why the Overdrive shouldn't be a compatible replacement for the original cpu.  (Only problems I know of are when the cpu isn't in firmly - the socket is a stupid so-called LIF socket which almost requires pounding the cpu in - or when the card isn't seated properly.  But I'd not expect Tam to have those problems...). 

From Tam
Hey Louis, some tinkering over the weekend (shortened by another wedding) yielded interesting results. I had three P90 complexes (1 with the always-stable "terminauter" mod at 200MHz, 1 with Joe's Overdrive mod at 180MMX, and 1 with the Evergreen interposer at 233MMX) testing on a "bare" 95a system with just the SVGA adapter and a fast/wide adapter for hours on NT. 

Guess what? All complexes ran perfectly fine and never saw a glitch in testing! So, this leads me to believe that the problems I've seen in the past with the MMX processors (including the 233MHz setup) under NT were a result of incompatibility with either the Cheetah RAID controller or the ATI GUP. All my prior testing was done on my maxed-out 3QT, which I'm learning isn't the best "controlled" test environment due to the complexity of the system. Anyways, when I get a chance I'm going to try and isolate the problematic component. It's a little early for any definitive conclusions, but I think that there's good possibility that MMX *can* run well on the 
type-4 platform. 

So this is somewhat a bit of potentially good news since a stable 233MMX system would kick some serious butt. Just might have to sacrifice RAID or GUP capabilities, though. 

CM-174 Universal Adjustable Clock Multiplier Adapter Kit
From Carroll Bloyd
   Using a clock multiplier kit sold by Concept Manufacturing, I have successfully (so far) upgraded my P90 complex to 150 MHz.  (Why a 150? Because that's the only extra Pentium I had on hand.) 
   The kit consist of an interface socket that plugs into the CPU socket, a wafer thin printed circuit board (PCB) that sits on top of it, and a very high quality heatsink/fan combo.  The pins on the CPU pass through matching holes in the PCB.  You end up with a sandwich composed of CPU, PCB, and interposer, which is then plugged into the Pentium socket. 

   The interface socket does not carry all the CPU pins through to the socket -- I think three are missing--I suppose these are the ones that determine clock speed. 
   The PCB has conductors where these three pins pass through and some circuitry that leads to a four position dip switch.  Using these switches you can set the clock multiplier anywhere from 2.0 to 5.5 in .5 step increments. (Ed. 5.5? Mathochistic!!) 
   The heatsink/fan unit clips wouldn't fit the tabs on the CPU socket on the P90 complex, so I used another heatsink/fan.  The Concept unit also looked liked it would be too tall to fit over the power supply in the 9595. 
   I'm running Warp 4 on this Model 9595, so I used Sysbench (version 0.9.4d) to do before and after testing.  The results show a 60% CPU speed increase, as expected. 
   The entire process was easy as could be and I was even able to adjust the dip switches with the complex in place in the system. 
   Oh yes, you can reach Concept Manufacturing at .  Look at the CM-174 Universal Adjustable Clock Multiplier Adapter Kit (Sockets 5 and 7).  It costs $30.95 plus shipping, and you can order online.  Concept has a variety of other interesting products also worth looking at. 

Hold The Phone! (As of 7 Oct 99)
  The P90 complex in my 9595 OPT went dead during the night two days ago. The P90 complex appears perfectly normal--I can see no obvious problems.  I have removed the Concept Manufacturing clock multiplier and P200, and replaced the original P90 and still no go.  Doesn't work in my 9595-OQT either. 

  Ed.  This is a reason for caution. We haven't figured out the reason for this yet... 

P90 to P180 w/o Soldering
   I have installed an interposer socket from terminaut. This socket has three pins connected with a simple wire. No soldering is required. 

   My P90 complex accepted the Terminauter without comment. I am using a P200 with the stock 60MHz oscillator which gives me 180MHz (3x clock). The LED displays 90MHz, but the processor speed under NT 4.0 and Netfinity 5.00 is 180MHz. After a year, I started getting odd behavior.
   Ed. This will eventually overwork your regulator!

From Tony Ingenoso 
   One possible non-solder approach is to ignore the fan voltage (the Evergreen fans fail and sound like a lawnmower within 6-9 months continuous anyway), rip their bogus fan/sink off (diagonal cutters) and go with a full heatsink or sink/fan combo powered by an external plug to one of the std power connectors 
   On my MXPro, the sink was attached to the CPU with some double stick tape. I Clipped the leads off, went external with a mondo fan/sink and now all is well again (and quiet<g>).  It will take some *careful* prying with a thin blade to get the old sink/fan off, but it will come off eventually.  Tape remains can be scraped clean with a single edge razor blade. 

Dr. Jim fires back 
   While this should work with the Evergreen, my understanding is that the Intel ODP will downclock or outright shut down if it doesn't see a tach signal from it's fan. 

   Is your MXPro in a PS/2 - to be more specific on a Complex 4? I have been loaned one - and it does not work - same error : / 

   Mine is currently in a PC 750.  I believe others have gotten the Centaur C6's to work in T4's though. (assuming yours is actually a C6 and not something else...) 

   The one I have just been lent is an MXPro 200  from Evergreen - so have absolutly no idea. What I do know is the interposer underneath is essentially the same piece of kit as on the Spectra 233 upgrade - and SUPRISE SUPRISE - I get the same error as stated erlier on in the news thread. 

Intel overdrive processor - socket 4, POPD5V133

Al Savage
> So, CPU gurus, will this work in my 8590-OLF w/Type4 P60? 

Jim Shorney
You need to stop skipping the meetings, Al...  I just tried this last week, both with and without an Intel EUCD35 interposer.  Didn't work. Sometimes would get as far as CP:8, sometimes wouldn't even post.  I friend of mine here in Lincoln who doesn't read the NG has also tried it, and he had even less luck then me; his machine wouldn't post at all. 

I'm hanging onto my chip, though, in case Tony or someone else comes up with a solution.  I suspect timing loops in the BIOS, but I'm neither a programmer nor an engineer... 

Cache Controller/CPU Incompatibility
   From the recesses of my mind- Someone tracked down the fact that the POD133 will not work with the cache controller on the P complex, BUT the individual had swapped that cache controller out with another type. I can't find the details... 

From Jim Shorney
   I can answer that: none.  Tam Pham discovered in his arcane book of shadow knowledge that the Pentium Overdrive 120/133 is not compatible with the 496 cache controller on that complex.  This seems to be the most likely reason that all interposer upgrades to date have failed as well.  Many have tried, all have failed.  There is a possibility of using a 497(?) cache controller on this complex, but it appears that it may require a voltage interposer of it's own.  This prospect is still awaiting investigation (anyone got a spare 497 they can send me?). 

From Dennis Smith
   Jim,  I'll let you *borrow* the 82497-60 from my dead P90 complex.Be advised, Tam found out that the 497 has nine pins set for 3.3V.  I ran a 497 in my P60 complex for about a half an hour with no ill effects though. 

Peltier Junctions
  This site has some details you should consider before slapping down your money for a Peltier. 
The Heatsink Guide Peltier Coolers

From Ian Brown
>Anyone, with experience? 

   OK, let's have a go at this (based on experience). First of all, a Peltier, when operated correctly, could cool a CPU more than the standard heatsink/fan combination. i.e. the core temperature could be lower. 
   But that is only half the story. The other half is what happens to the heat that it is removing from the CPU *plus* the heat that the device generates as part of it's normal operation. PLUS the heat generated by the extra loading on the PSU. 
   What I'm really saying is that a Peltier is great for removing heat from localised 'hot spots', but only if the total thermal dissipation of the equipment AS A WHOLE is sufficient. And this last point is the one that is so often overlooked, sometimes with disastrous consequences. You have 
to make sure that the hot side of the Peltier has adequate cooling (and you put it on the right way of course). And you are not cooking the PSU. 
   Having tried various cooling methods for CPU's and similar components, I must admit that a good heat sink with sufficient air flow takes a lot of beating in practical terms. The Peltier looks to be better, and indeed is in theory, but by the time you have put a decent heat sink and fan on top of it (if you have the room), and then remembered that your long suffering PSU is probably running a lot hotter, the use of a Peltier begins to lose it's charm. 

A worst case scenario would be something like: 
Attach Peltier to CPU. 
Attach heat sink and fan to Peltier. 
Fit extra fan to case to remove heat from the above. 
Fit extra fan to PSU to compensate for the increased loading that has ensued from the above steps. 
Wait for the whole bloody lot to go bang. 

And don't laugh, I have actually been up that road. And it wasn't funny. 

It's up to you, but my general advice is that if you can't cool your CPU enough using a standard heatsink/fan combination, then you've got a design problem. 

From Richard Ross
We tested two early units a couple of years ago, when the Cyrix (uhhggg!!!) P200+ single-rail CPU was released. This CPU was capable of consuming between 32-39 watts, much of which was dissipated via the CPU's top thermal surface. After cooking (seriously!) a perfectly good 
AAVID fan/heatsink unit under 24/7 high load testing (and 200 lockups!), we decided to give two Destech Solutions, Inc. Peltier units a try. The heaviest unit worked flawlessly for about six months, at which time the CPU was replaced. The lighter duty unit cooked its own fan, due to the intense heat and undersized heatsink. Peltiers can also *undercool* a processor, if their thermostat isn't operational, or if it's a cheap unit which runs constant. Not really undercool, but they can cause condensation to form upon power up, and low power states, which, of course, is bad, bad news for the components that get wet. The trick to using a Peltier is to remove the heat they produce, along with the heat the CPU is producing. The case must have lots and lots of properly designed airflow. Cool air in and hot air out. Touching a heatsink of a Peltier under full load will demonstrate the point! In my opinion, Peltiers are nice for hobby/experimentation, but not for anything even remotely mission critical. If you need a Peltier, you need to redesign or step that clock back down to spec. :-)) 

Madex 586007 for S 5/7 CPUs, switching type
Product outline original HERE
Jumper Settings (auf Deutsch!) 

Max. I out: 13 A
V out: 1.8 - 3.5 V
Input Capacitance : 3200 µF
Output Capacitance: 6900 µF
Inductance: 2 µF 

Suitable heat sinks: No 1380, 1327, w/o clip for S7. No 1389 or 1387, for over-length heat sink clips for direct attachment to S7. For adapting S5 and S7 main boards to the new generation of processors.

Charles Lasitter sez
   If I understand Peter correctly, this version of the adapter is the only one that has the correct clearances for our favorite processor board.

9595 Main Page