The Dual Celeron IdeaIntel recently announced the Celeron processor which uses PentuimII core but does not use any level 2 cache. This makes it a very low cost CPU able to run a low cost system.
This Celeron CPU is really excellent regarding overclocking tolerance. It has been confirmed by multiple sources that a 266MHz Celeron can run at 448MHz stability.
The PentuimII can run in a dual CPU configuration. Since the CPU core of Celeron is the same as PentuimII, I thought that I could use this good cost/performance CPU in a dual configuration.
The Problem of DualizationThe first problem when we try to use Celeron in a dual configuration is this: On the Celeron , BR1# is not connected from the CPU core to CPU board. That signal line is necessary for the CPU to operate in a dual configuration.
Through further invesitgation of a datasheet on the Intel(R) web page, the pin assignment of the Celeron is the same as the PentiumII except 66#/100#, which decide the base clock. By connecting the BR1# signal line to the land of B75 on the CPU card edge, I assume that the Celeron may operate in a dual configuration.
Getting the ComponentsNow that I have the goal, I get a dual motherbord and two Celerons (retail package SL2QG.)
412MHz overclockable Celeron (retail package SL2QG) 2
It is hard to buy the Celeron because in most shops
they are still sold out. The 1st one, I bought in the
middle of May, and second one I can bought at the
beginning of June. The two Celerons have different
production dates. The 1st one was produced 5/1/98, and
the 2nd one was produced 5/15/98.
Fortunately, both Celerons can operate easily at 412MHz.
Before AlterationBefore altering the processors, I installed the 2 Celerons in the motherboard and turned the power on. The result was negative: Even the BIOS did not boot. That's a matter of course, but there was a little possibility that it might work by mistake.
AlterationFirst, I prepare the wiring.
0.26mm JUNFLON narrow wire
I get 0.26mm JUNFLON narrow wire which looks quite narrow. Using this wire I connected BR1# in the CPU core to B75 on the board. Unfortunately, this 0.26mm wire is so thick that breaks or slides out when the Celeron is installed into the motherboard. So I find even narrower wire.
Narrow wire bought in Akihabara
I use the even narrower wire bought in a Akihabara shop named Sengoku. I forget the real name of the wire. Now with the narrower wire I reconnect the CPU core and board. There is no break or sliding out when the Celeron is installed into the slot. It should be OK.
BR1# in CPU core B75 on CPU conector
As you can see in the upper photograph, the wiring is very tight. If just a little slip happens, the wire could stick to next land. To prevent this I cover the area around B75 with masking tape. Then I start soldering.
Masking the board
The wire is coated with insulation, but to make sure that the wire is not shorted to another pin, I mask the board also.
2 Celerons wired
After checking carefully for any short, I replace the fan and heat sink, and the alteration is complete. The 2nd Celeron is then altered the same as the 1st.
Operation TestNext, operation tests.
Single operation test
Celeron operation check by 266MHz normaly
At first, just to be sure that the altered Celerons operate normally, install one by one and boot. It looks like it operates normally.
Now try Dual modeAfter checking that the altered Celerons do not have any problems, I try dual operation.
Install 2 CPUs
Result is negative. Even the BIOS do not operate once again. I recheck the wiring of the altered Celeron. There looks to be no problem. The wire does not short to other lines, and contact looks right.
I boot again as single. It is still no problem.
Just to take a chance, I remove the wire from only one Celeron, and boot dual again. It's looks OK. It boots with no problem. The joy is short-lived however, as the motherboard only recognizes a single processor.
I assume that only one of a pair is recongnized. Still there is reason for further investigation. For the time being this reserch is stopped. In the future, when the investigation progresses, I will update this page.
--To Be Continued--
RumorsNow there are some interesting rumors.
PentuimII processors have a function named APIC (Advanced Programmable Interrupt Controller) that is necessary for Dual CPUs operation. We can check if the CPU has this function by using WCPUID (Mr. H.Oda! build) which is very famous free software in Japan. But when Celeron was operating on single slot motherboard this software recognized the CPU as one which did not support APIC function.
Display of WCPUID
I do not have any other slot1 motherboard except ASUS P2B-DS, so I can not check what is displayed when using any other motherboard. I can not solve this difference in results. They come from either the motherboard automatic detection of the Celeron or a difference because of Celeron's S-spec. Without the APIC function, dual operation should actually be impossible. So it's very remarkable.
* Some systems whose environment is same as me, displayed APIC = Not Supported. What's that. It's not understandable.
--To Be Continued--
APIC Supported?As a test, I changed the state by removing the wire from BR1# to B75, and checkd the display about APIC. As I thought, It displays 'Supported'. With both Celerons, this result is the same. Same motherboard, same type of CPU, and same S-SPEC (SL2QG), but result is different. It's a really mystery.
* Later, I was found that depending on the OS the display about APIC was different. On Windows NT 4.0, it was displayed as supporting APIC.
--To Be Continued--
The voltage of BR0#/BR1#Next, I check a voltage of BR1# when Celeron is operating. Depending on the data sheet of PentiumII, BR0# is GTL+I/O and BR1# is GTL+Input, and both pins are pull-upped by 1.5V.
And now checking, BR0# is supplied 1.5V power, but
BR1# is always supplied 2.0V power. Though this voltage
looks little high, I cannot check the pattern from the
surface of the board to BR1#, I suppose this pin may be
pulled up in the middle of the board.
Without solving this problem, it will not be a success.
Now the investigation will continue.
--To Be Continued--
For dual operation I have to solve the problem about BR1# which is supplied 2.0V power. For some time I trace the pattern of the board of Celeron, and measure the voltage at various points, but I can not find a good idea. Finally I decide to use a really violent way. It is actually pretty easy. I just cut the 2.0 voltage to BR1# which may be connected to VCC-CORE.
So I prepare this.
Drill 0.5mm drill pin
Using a 0.5mm drill, I shave the BR1# of CPU from the reverse side of the board, until it does not touch to the board.
The image of shaving
Restart the Alteration
Now start to shave BR1#. Because there are some other lines on the reverse side of the board around the BR1#, shave slowly, carefully and vertically...
Continue to shave only a little beyond the thickness of the board. After shaving I remove the waste by an air duster or something.
The Excellent Hole Magnification
The section of BR1#
I holed another Celeron the same way. And 2.0 voltage supply which was a problem has gone away. I reinstalled the altered Celeron to the motherboard for confirmation. Then I measured the voltage of BR1#. If BR1# is 0V, the alteration is successful.
But only with the above alteration, Celeron will not
work as a dual CPU.
According to the data sheet of PentiumII, BR1# is GTL+Input. So it will decide 'H' or 'L' by 1.5V as a standard voltage. So now BR#1 has to be pulled up to 1.5V.
Pull-up BR1# Connect to RP6
Figure of connecting to RP6
Top 2 pins of RP6 on the Celeron board are free, so I connect from here to get 1.5V pull-up. It is easier to connect from the topmost pin.
Lastly, I carefully checked the jumper line for a short or lose solders. If there is no problem, the alteration is over.
Now Check OperationNow, I insert both Celerons and power on the system.
BIOS recognizes as 2 Processors.
If the system will not boot here, or the system recognizes as only 1 processor, you may need to recheck the wiring. Perhaps the wire is worn or pulled out of the connector in the slot. When even a single processor does not boot, Celeron may not be correctly installed in the slot.
Dual Operation Success!!
On July 25, I confirm Celeron dual operation on BeOS R3.1. This is a screen shot at that time.
BeOS R3.1 Dual runnning!
Exact dual operation. RC5-64 client program for BeOS works in dual mode. The key rate is double.
I confirm dual operation on Windows NT 4.0J Workstation. I do not find any strange behaviors at present.
Recognize as MPS 2 System processors
Two months have gone by since getting the idea to try dual Celerons. Now it is a success. Thank you all for the cooperation. Special thanks to all who gave me information in the message board at Celeron World. Much thanks!
A word of warning!!
This alternation is dangerous and it could destroy CPU physically. If you try to do this alteration, you bear the responsibility. I will bear no responsibility.
Added Benchmark result.
I added the English version of this page today.
Added 3D Rendering results to Benchmark results page.
The dual system which used one Pentium2 and one alteration Celeron was possible. (It worked!)
Added the English mirror page of dual Celeron article in cpu-central.com.
Celeron300A worked in dual by this alteration. New!
English transration :