Celeron Buyers guide, PIII`s too!

[Toaster]

Well, I had fun. A friend who owns a computer store allowed me to do a test with various Cel-mines and PIII`s.
All CPU`s tested were of the "coppermine" variety and all FC-PGA. He had no "slot variants available".
It became quite clear after 4 CPU`s that the following gave the best results:
1. If the Celeron2 CPU had an advertized voltage rating of 1.50 volts, it was almost guaranteed to fly at 100FSB and 1.85 volts. Whenever we tried any 1.50 volt CPU, at 100 FSB and 1.85 volts, it flew, every time.
Tested units were:
Cel-2: 533 ->800mhz
Cel-2: 566 ->850mhz
Cel-2: 600 ->900mhz
Cel-2: 633 ->950mhz

The 633 unit came in BOTH 1.50 volts *and* 1.65 volts.
The 1.65 volt unit flew, but it required add`l work and cooling. This was the case where ANY CPU that had a stock voltage of 1.65 volts.
Bottom line, its easier to "crank" the units with a 1.50 volt rating but the others fly too but require add`l effort.


For the PIII`s, the same applies. However, they start at 1.65 volts and go up from there.
Tested units were:
PIII-600/E -> 800mhz (133fsb)
PIII-700/E -> 933mhz (133fsb)
PIII-800/E -> 1064mhz (133fsb) *note*

These units flew at 1.90 volts or less with little problem except the 800/E variant.
The 800/E was a different stepping, both internal and external steppings. One we tried wouldn`t gdo anything above 120FSB where the other flew at 133fsb.
It seems the "external" stepping is almost meaningless and the internal stepping is the key to the faster PIII`s.
Intel DOES NOT advertize the "internal" stepping and the only real way to tell is to call up an internal register which some programs do. The "external" stepping is simply for matching pairs in dual configs.

The bottom line is thus:
Choose a CPU with the lowest "advertized" core voltage and chances are quite good it will fly.

[mosquito]

that makes sense.

Give me your opinion on this one, the 1.65 cpu's failed on the 1.5v test by intel, and worked at 1.65v (resulting in a higher yield). Which indicates that they will score bad (not as good) when you try to overclock them (as they already failed the tests at normal voltage).

I don't know if this is true, I just want your opinion about this.

[Toaster]

That is what i`m surmizing and evidence points to it as well. I would guess that Intel sets a base voltage of 1.50 volts for the Cel-Mine and 1.65 volts for the PIII.
As the chip fails each test, the core voltage is raised and its multiplier and or FSB are lowered. This is why I truely believe that the "copper-mine" CPU is configured *after* production where it is then decided wheather its a PIII or a Cel-Mine and its speed.
In *every* case where the Cel-Mine had an advertized voltage of 1.50 volts, it flew at least at 100FSB and often faster. When CPUs of 1.65 volts were cranked, most flew but with some difficulty. The same boar out in the PIII.
So, yes, I buy that scheame, it makes sence to me.
Iv`e also desyphered the internal steppings as well to some degree. The internal steppings seem to be an internal "Intel only" code to differentiate "core testing phases" during manufacture. Its odd, that both the PIII and the Cel-Mine CAN and DO on occaision the same internal stepping.
This isn`t a guarantee by any means but i`d feel real comfortable saying: If it`s 1.50 volt, IT WILL FLY!

[Edited by Toaster on 11-21-2000 at 01:46 AM]

[mosquito]

I will pay attention to it when I purchase a new system.
thx for the info

[Floppyman]

Agree with you here Toaster. Both my slot 1 PIII 600Es do 800 at default voltage. My motherboard does not allow me to raise the voltage any so I'm very lucky. Sometimes my system becomes a little finicky...not sure whether it's the cpu or the M$ OS though. Have been running 2 clients of SETI@home and the system does not lock up while running both at the same time while the cpu usage for both cpus is 100% when running both clients. Thx for the info!

[Toaster]

I have yet not found a PIII-E/600 that wouldn`t fly at 800mhz (133FSB) This is the choice for entry level into the PIII IMHO. Aside from this, they are cheap at right around the $100.00 mark.
THIS DOES NOT APPLY to the "EB" CPU which requires 133fsb to attain 600mhz!
The more I look, the same applies to any PIII-E/Celmine, if it has the nominal voltage of 1.65/1.50 volts, it will fly.
I recently converted a Cel-Mine 600 to a PIII-900 to which all the cache (256K) is enabled and even the Intel software calls it a PIII. When I come up with the "easy way" i`ll do a HOW-TO for those brave enough and with EXCELLENT soldering technique. This has a "pukker-value" of about 8.5 on a scale of 1 to 10 where 10 is EXTREME Pukker value for the few brave souls.

[HAL9000]

I'll be looking forward to that one Toaster!

[mosquito]

Me too.

Is this converted Cel SMP enabled? It would be great if I could build a cheap 900Mhz dual system.

[Toaster]

As to wheather the converted Cel-Mines are SMP enabled, I can`t say at this time. I however "assume" they are but I havn`t tried it first hand. As yet, I havn`t found a dual capable system board I would want to own. The ABIT board is about the only thing i`d try as its a BX board and the industry considers it a decent board. As this info comes to bare, i`ll adize. The Cel-mine I converted is of the CA0 stepping which is also used in the PIII. The intel software that quizzes the CPU directly says its a PIII with the full 256KB of cache enabled and the L2 latency lowered to "1" from a Cel-Mine standard of "3". I`m starting to think that the "internal" stepping suggests "which" assembly line it came from, either a high yield line of a more exact low yield line. This for the time being is pure speculation and not fact. My friend who has the computer store suggests that intel will be releasing a 1GHZ Cel-Mine by 2nd quarter 2001. This too is rumor and speculation as well as Intel moving to the .14u core over the entire PIII production.

[mosquito]

If you mean the Abit BP6, you can't run PIII's dual in that board without heavy modifications to the sockets. There is a converter from powerleap, but it's not confirmed that it's working. (check [url]http://www.bp6.com[url]). The guy who runs the site tried to set up a dual system with CA0 stepping, and failed. But I found reports on the site from people that got the thing working.

If I'm going to implement the mod I'll do it on a Dual PIII capable board (my BP6 with celerons is working fine, and I want to keep it that way.)

[Toaster]

Iv`e been seeing folks with problems getting the CA0 stepping PIII`s to work in a dual config. I cant suggest whats going on as yet because I havn`t tried it personally.
The CB0 stepping is often found on PIII`s of 850mhz and faster. Still, i`ll give it a try when time permits and a decent board becomes available. I would like to avoid the use of slockets when possible as they tend to load the CPU differently and this creates extra heat and other problems.

[mosquito]

Can you keep me updated when you're able to test it, I'm planning on a new dual CPU setup and I'm looking around to see what the options are. If this celmine setup works out, I can build a cheap powerfull rig for photoshopping.

[Toaster]

I`ll repost when info becomes available.
A friend is working with 2 Cel-Mines as we speak but nothing on this as yet. He seem confident and would try 2 converted Cel-Mines at my leisure. I now only have one converted chip and within a week or 2 I`ll be able to sit down at the rework bench to convert another. More as it comes in.
(Dual Cel-Mine 900`s would be a screamer)

[mosquito]

Yeah, I'm going out for the mobo this weekend.

[raydioactive]

Is there a way to modify the cpu that breaks the multiplier lock on a celeron?

[Toaster]

With the mod we`ve done, yes and no to the unlocking of the multiplier. The chip we converted was the Cel-2 600. It has the original multiplier of 9.0 x 66mhz. Now, it runs 100% stable and even passes the Intel CPU test and verification (see note) at 900mhz @ 100mhz x 9.0. The multiplier is what I tend to call "flexible" for a reason I`m still investigating. When the system board is set to 10x, the system posts and runs at 1GHZ and is quite stable. After about 1.5 hours running SETI @ home the system locks and requires a restart. Then, it will run for another 90mins and lock up once again. The chip seems to tolerate and recognize a multiplier change of 1x plus or minus. The system still hung even when the multiplier was forces to 8x or 800mhz.
This is still being investigated. So far, we surmize that during the manufacturing, the multiplier was chosen and bus speed during final assembly. Still, we can only surmize and not speak fact. I contacted a good friend down Intels way and he simply said "your getting closer,this is a bad time to stop" but would say no more.
We can do the following:
1. Change the operating voltage of the chip.
2. Enable the intact 128KB of L2 cache.
3. Have the CPU "request" a 100mhz+ CPU bus.
4. Enable the CPUID to report itself as a PIII.


We obtained a "true" PIII-900 and ran benchmarks.
The benchmarks stayed within 1% at all times. In effect, the Cel-Mine IS a PIII and performs as one at the same clock rate. That is, there was no real difference between the two and the "1%" we concluded as normal variance resolution of the benchmarks.

The benchmarks are:
3D max99
3d Max 2000
SETI
HD TACH
Final reality
Linux 6.2 RH (this OS quizzes the CPU(s) directly and gives a rather exhaustive information screen claiming it to be a PIII. When a Cel-Mine at 100mhz was used, it was reported to be a "celeronII" @ 900mhz.


Note to the above:
The Intel verification software reports the CPU as:
PIII-900 (e) w/256k L2 @ a latency of 2. The CPU "serial" number DOES NOT MATCH any known record and gives us a telephone number to call for more information. (no thanks)


While what we have done is legal, it would be illeagal to claim the CPU as a "genuine PIII-900".

[Edited by Toaster on 11-29-2000 at 02:15 AM]

[raydioactive]

Thanks Toaster, good luck with what you are trying to achieve, if anyone can do it you can:) I'll be very interested to see your findings. I have a celeron 633 with a 1.65 default voltage, can get it to work at 950, but only for about 5minutes before she locks up, guess this explains it.

[Edited by raydioactive on 11-29-2000 at 03:41 AM]