Intel's Little Secret ~ Re-marking Own CPUs

[Roy]

Hi, y'all! I've been away from the forums for a while, so maybe someone already has talked about this. It took me a while to figure out Intel's clever marketing gambit.

After spending millions on the Coppermine architecture for the new 133MHz FSB, it dawned on someone that a huge upgrade market would be missed if 100MHz models were not offered. Since all units are multiplier (ratio) locked, they could fudge only on FSB frequency, but in the underclocking direction! So they quietly re-marked some high core speed models at lower numbers.

Here's how it works. A few weeks ago I found myself ready to "go Coppermine". After looking at the numbers and prices, it was obvious that a modest 100MHz choice stood a better chance of overclocking, what with mobo and SDRAM speed considerations. There's a lot less room above 133MHz.

So I got a PIII-650/100 Slot 1 and started to play. [BTW ~ the "E" suffix does not apply to models from 650 up, they're all 0.18 micron technology.] It went 6.5x126=819 without balking. I guessed my ram was maxed out and went for some nice Mosel-Vitelik tCK=7ns from The Memory Man. Good guess, 6.5x133=867 was no problem.

That's when it hit me. What I have is a PIII-867/133 in sheep's clothing! What's more, it overclocks from there to 6.5x144=939 and never blinks. I saw 6.5x155=1007 for a few minutes, but at that point I don't know if my ram or the processor said "stop!"

So here's what I think - a PIII-750/100 really is a PIII-1000/133. I'm trying to justify going for a PIII-850/100, just to see if it will go 8.5x133=1133 and beyond. It well may be that Intel is just sitting on this one until AMD announces something in the stratosphere.

Oh ~ I should add that the MSI 6163Pro mobo I bought last summer has lived up to my expectations. It just took longer than expected to prove it. And I'm cooling with a flea-market dual fan HS carefully optimized. Typical temps from a probe on the heat plate right next to the CPU are in the low 90s F.

Happiness IS! http://sysopt.earthweb.com/forum/smile.gif

[Jv44~Siggi]

So how do they stop diagnostic software from spotting it's true identity? Re-marking the case is one thing but can they actually re-programme chips to falsely report their speed?

[Roy]

Well ... what I meant was that the basic die is the same.

I have to assume there's room in there for some EPROM to hold the ID info. That would be part of the marking process.

[Butler]

Sounds pretty far out to me...why would Intel be sitting on 1Gig chips, when the major computer manufacturers are begging for those proc's and are only getting them in batches of 40 or 50. It's still nearly impossible to get ahold of a gig proc, even looking to Dell, Compaq...etc. etc. Intel wouldn't be gaining anything by hoarding their most pricey processor, it just doesn't make sense. You've just got a good chip that OC's well (in my humble opinion)

Butler

[Roy]

Overclocking by 25% is common, but 44% without agressive tricks is not.

It's thoughts like Butler's that will push me over the edge and justify the cost of finding out. I'm getting those twinges again! http://sysopt.earthweb.com/forum/wink.gif

[AuraEdge]

The 1Ghz P3's out right now arent the same as all the other CuMines. They have had subtle chances made in them, such as narrowing gateways within the Processor, in order to make them more stable. This is partially the reason for the large overhead cost on the Ghz. Its because they hadda put some more work into it.

They way that Chip manufactures batch thier Processors is very similar. They test the rate at which they can run, and divy them up into 'piles' for each processor speed. These 'piles' will be based on demand and percentage of clock rate at that clock speed. If, say they get too many 700's (80%), they can throw some into the 500 or 600 piles, because not 80% of people will want to pay for a 700Mhz CPU. In this way, they can sell more CPU's, and even if they dont make as much profit per CPU, they can make more money in the end. After thats set, they lock down the Multi, either the speed divided by 100 or 133, percentage dependent on demand, and ship em in a box, labeled accordingly.

Im pretty sure they make the speed markings after they lock the CPU's, which would then be the initial marking, not a re-marking.

Also, notice that although u reached 6.5x133, did u use a higher voltage than the default 866? (1.65 or 1.7 i think it is..). Intel absolutely cannot ship a processor that requires more than the default voltage for that speed to run that speed. This is one of the fields were we overclockers can take advantage.

I dont know how accurate my representation of the CPU multilocking goes, or the sorting, but thats they way I invision it as happening. Again, this is just my representation, and I dont want any flames if I screwed up somewhere along it...but corrections would be more than welcomed.

[redly_one]

44% is EASY with Coppermine. Hell, 46% was cake in my case. 550E @ 803 with stock heatsink and good components. I've heard of people having even better results, 825 and even more with the 550E. 600E is also good.

[This message has been edited by redly_one (edited 06-08-2000).]

[Roy]

Thanks, Edge. I used the "re-marking" expression to get people's attention. Your description of the selection process matches my understanding. I hadn't heard about the 1Gig "mod", certainly the big 'I' has QA issues that govern the high-end product.

Indeed, I am running a small voltage bump ~ just for good measure. I'll do some runs at the stock 1.65V just to find that answer. Right now though, my CPU temp is a cool 84F with a Vcore of 1.70.

I'm also going to push the settings (chipset 3.3V too) past the conservative 10% mark and give 1Gig another run. Stay tuned.

EDIT

I'm back from that V bump run. 1007 lasted about 10 minutes, until I tried a serious graphics app. Vcore was 1.80, somehow I was thinking 1.60 was stock. Maybe I'll try again @ 1.85 or 1.90. Chipset V was 3.6. CPU temp stayed under 90F. Rain helps! Windows Registry got scrambled when the BSOD interrupted the fun.

redly_one ~ The lower multipliers tend to overclock better, percentagewise, but the actual OC core speed is not as high.

EDIT II

I've been running at 939 for an hour at stock voltages, including some heavy usage loading. No problems, still 84F.


[This message has been edited by Roy (edited 06-08-2000).]

[oblivion]

I just sold my 550e that did 850 at default.....hope my cbo is the bomb now

[krusty the klown]

Roy,

I personally would be cautious about using 1.9V, but that's just me - it's your chip at the end of the day!

What is the chip core temperature? The P3's have a thermal diode inside the core, which can be read by e.g. SiSoft Sandra and by the BIOS itself. Obviously, you cannot read the core temperature from BIOS whilst the system is under 100% load, bt the BIOS reading seems more accurate. Whichever way, reading the core temp will be far more accurate than placing a probe near the core.

What cooling are you using? I reckon that at ~1.8V +, stock cooling is inadequate - for two reasons: the cr@ppy thermal pad and the heatsink itself is only designed to cope with stock speed & voltage. A decent cooler and thermal paste is definately recommended.

Because the core is so tiny, there is a possibility of slight misalignment i.e. not getting the heatsink exactly flush against the core - a few people have had problems here, but if you're aware of it in the first place, you know to check this out.

The coppermine is a gem of a chip isn't it? My slot1 550E is humming along nicely @ 852MHz (she will do 733 with stock cooling and default core voltage). I am using a GlobalWin VOS32 heatsink - overkill, but effective - and my core temps do not get above 36ºC on a hot day, with the case on running @ 100% CPU usage for hours. This heatsink has dropped the temperature from 47ºC (too hot, IMO) @ 825MHz, +7.5% core volt increase. So that's an 11ºC drop and I'm now @ 852MHz with a 10% increase in core voltage.

Hmmmm.... personally, I reckon you should hang fire with the 850, unless you get an exceptional offer for your 650. ~1000MHz is about the theoretical limit for that technology, so I've heard on this BBS. I reckon with a decent cooling setup, you have a fair chance of hitting 1GHz with the 650, as it posted and got into windoze - if it ran for a few minutes at this speed, then gave in, heat could well be your problem and these chips seem susceptible to heat, so treat 'em gently http://sysopt.earthweb.com/forum/wink.gif

Happy clocking http://sysopt.earthweb.com/forum/smile.gif

[Roy]

While krusty the klown's thoughts are sensible, most of his questions and concerns were already answered in what I wrote.

"Still Fine @ 939!"

[krusty the klown]

Oh well...

I would rather trust the on-die temperature monitor.....?

EDIT: I'm curious to know how the temperature measures by the thermal diode in the chip compares to that measured remotely.

http://sysopt.earthweb.com/forum/redface.gif MORE EDITING: I guess it was just too early in the morning (UK).... I just saw the alarm bells..... coppermine....1.9V.....lockup...... D'OH

??? I didn't even get drunk last night, either http://sysopt.earthweb.com/forum/wink.gif

[This message has been edited by krusty the klown (edited 06-09-2000).]

[AuraEdge]

I believe the default for the Ghz CPU (7.5x133) is 1.8v.
One thing Ill take from whatever Intels doing..I wouldnt be afraid to go UP to the voltage of the highest CPU in the same family (I mean coppermine family in this case).

This bring an interesting note to the Celly 2's. They're built on CuMine cores, but are volted at 1.5v. I wouldnt hesitate to go up to 1.8v w/ the Celermines for this reason. Ive ordered a Celly 566 recently...
We'll see where that takes me! The board I got right now is a wank overclocker (SY-7VCA only allows +.16v...WAY short of what I need)
I may replace it with a CUBX or P3V4X, or maybe one of those BE6-2's with ATA-100.
Since Ill be running a Celery, theres no real need for 133Mhz Support.

[Roy]

Ah Ha! Thanks, Edge. I've missed quite a lot while focusing on another area of computing - high-end sound processing. It was the rendering time of half-gig files that pushed the "need for speed" button.

1.8V is a comfortable 9.1% increase over the default. My test at 1.9 was only 15.15% boost. Maybe I'll figure out how to fit a Peltier into the package and go for 2.0V ~ the max my MSI mobo allows. They would not have set that limit without input from Intel as the Coppermine was developing. Of course, 21.2% is outside the 15% I usually consider an agressive max, but agressive cooling should make up for that.

The higher multiplier continues to pull at my sensibilities, all else being the same. Architechture, that is.

[Roy]

I had a BX6-2 capable of monitoring the CPU thermal diode. There was no external probe to compare with. I wish the MSI provided the function, the CMOS readout is from the probe, not the diode. MoboMon ditto. My probe is touching the plate that carries heat away from the pretty blue surface of the PIII, less than 1mm away. It's been reading in the low 80sF since I dropped to the default Vcore=1.65V.

I have to accept that the rock solid 939MHz is the best my setup will do. The next available FSB - 155 - compromises the SDRAM and BX chipset sooner than the CPU.

Again, a higher multiplier and lower FSB are my next options. There's a PIII-850/100 in Illinois for $573. Hmmm. http://sysopt.earthweb.com/forum/wink.gif

[krusty the klown]

I believe the VIA 133+ chipsets allow you to run the mem at FSB-PCI speed (dunno about the BX - never had one!). My 550E clox stable @ 155MHz FSB (Soyo SY-6VBA133) and I'm running PC100 mem @ 117MHz (VIA). BUT the VIA chipsets have really poor memory throughput compared to the BX - 6 of one, half a dozen of the other!!!! I just couldn't stand the thought of ditching 128MB of mem and getting PC133!! Another nice feature of the VIA is the AGP/2 divvy - again, you're compromising mem throughput to run peripherals at a slower speed, so you're just pushing the CPU.

Good luck with the 850...... that's a lot of bux.....

[krusty the klown]

Sooooooo, the giga P3's run at 1.8V........ has Intel taken a bite out of of the overclocker's pie????????

[Roy]

I've done another run, this time at 1.95V/3.6V. The results were the same.

I think the BX chipset is the problem, a scrambled Registry has not been a symptom in any other overclocking I have done.

[800XL]

Intel has been doing this sort of thing for years. The 300A -> 450Mhz was another obvious one.

The Pentium III 600 Katmai core default voltage is 2.05v instead of 2.0v like the 550Mhz has. Another one of Intel's more known 'overclocks'.

Other than the speed rating, the only difference between a Coppermine 133Mhz FSB and a 100Mhz FSB with the same multiplier is the value of a couple of pins, BSEL0 and BSEL1. The right combo of short or open on those pins tells the motherboard whether to run at 100Mhz bus or 133. That change is done on the chip package, not the die, so it makes this "rebadging" very easy. The chip can even be speed graded then sit in a warehouse for a while before being officially clocked. If the demand for say a 650 is high and 866s are in low demand, the lot can be marked for 650 and set for 100Mhz. FCPGA chips are a bit more difficult, but not entirely different.

[Roy]

Thanks, 800XL, your explanation essentially confirms my analysis.