FLATNESS - anyone interested ?

[BillA]

this post is a duplicate of one on OC, but taking the liberty of posting here also for the non-adventuresome
- original thread .HERE

(and I'm hoping to get N8 involved, hello N8 . . . )
(and perhaps even a JoeK - JoeC 'contest' ?, more at the end of the post)

this is all about flatness
a bit of explanation: I bought a "Flatscope" which is an optical flat built into a reflex viewer with a mercury lamp and a green filter
- as such each visible fringe = 0.000010749" (~ 0.000273mm) and 'measurement' is made counting the fringes, or fractions, in terms of their deviation from linearity
(its not about how many fringes, its how curved they are; a bit oversimplified but that's the gist of it)

so I started looking at wbs, my present interest
(note that the surface must be reflective, more on this later; and that the images are severely pushed to enhance the contrast)
some examples - of my handicraft, no blame to the mfgrs here

the first wb bp I looked at, I was shocked (not my lapping)


a 'not good' of mine


here the die area can be seen towards the lower center


and one that is actually pretty good


too many images I’m told, have to split the post
to be continued . . .

[BillA]

flatness, Part 2

then I took a look at the heat die, not too enthused so I re-lapped it

but found after a bunch of testing that there was no discernable difference

on to the BIG question: what does the top of a CPU look like ?
but I had none to ck, called JoeC and he sent me a dead Duron and a P3
but both had non-reflective surfaces, time to lap which I did using ONLY 1µ film (the idea being to polish, not flatten)
- a big chore; the green 'coating' on the Duron seems to be a silicon oxide somehow processed to yield that finish (anyone know ? - not think, eh - know ?)
- and the P3 was just as difficult, looks like gun bluing - but MUCH harder

the Duron

pretty flat really

and the real shocker, the P3

it has a point on it like a church steeple (which I do not believe is any accident)

what can be said about all this ?
nothing, a sample size of one, on different CPUs from different mfgrs is not a basis for anything other than idle speculation

but if I had more CPUs to look at . . . . .

AN OPEN REQUEST TO OCers: I need your dead, un-lapped of course, CPUs - those sold w/o heatspreaders

I'd like to do an article on this subject -> but I need to see something more that a couple to have any validity at all

corner cracks ok, I can inspect
my address is HERE
please contribute

and the ‘contest’ ?
well, I was thinking that if I received more dead CPUs from procooling than OC – perhaps I should breakdown and write an article for JoeK ??

[Joe]

Hey Bill, I could send you a P3 1Ghz, and a Duron 800Mhz both work still, and I would like the P3 back eventually but they are just sitting here

if i looked hard I could probably find an old Slot A Athlon

[BillA]

lol, sh*t Joe, that Duron is a better CPU than any I'm using
its hell being a testing freak, no toys just meters and more meters

I do appreciate the offer, but lapping CPUs is generally fatal - even, I suspect, the super deliberate way I do it

I used to work in a wafer fab and the color changes on that Duron really started to look like a photoresist image - scary

if you wish go ahead and send them, I'll care for and return the P3 - but cannot guarantee it will still be alive

and THANKS

others ?
don't be bashful, step right up
(just follow your leader, lol)

[Joe]

lemme think about the P3 then. I have a gathering with a few co workers today, one of them was talking about wanting that CPU/Mobo I will see if he still does, if not then its all yours to play with.

[koslov]

I have four dead Tbirds (not mine, originally) and a Pentium Pro.

I'd send them right away, but I am stuck trying to find someone to do calibration work on my Digitec 5810, you wouldn't know anyone who does such a thing, would you?

[BillA]

koslov
I'm sensing there is a quid pro quo thing going on here ??

you need to find the combination of 2 things, or 3 perhaps
1) you need the procedure that is in the manual (why not offer to PAY the guy that has it for a copy ? - otherwise google, I bought the one I sold with the meters for $35)
2) you adjust the high and low ends over and over using 2 decade resistors of the appropriate range (and 'accuracy', eh ?) and a good meter
3) someone to put those 2 things together - ????

I don't meet #s 1 and 2, sorry

but I still do WANT your dead 'uns

[pHaestus]

I have a digitec manual and could probably be convinced to scan a copy. Time is an issue; would take an hour or two of work time (which I try to keep as short in duration as possible).

Bill I have a couple of dead XPs and a Duron around here; I'll get them packed up and sent Monday. Well, assuming I can find them all... I think one has mangled pins but the core is fine. Would that matter?

If you really want a bunch of dead CPUs then try HardOCP and their forums. A ton of casualties over there; in fact it's a vast resource of people with more money than sense (or ability).

[BillA]

but pHaestus, they don't like me over there, lol
actually that's not true, I think I'm ok with mota

bent pins are fine

HEY, get a 'concession' before sending off that manual
(joking, of course - like hell)

BTW, lot of killers here too
you fellows have already clobbered the OC donations, and we just started

[koslov]

Bill: Ok, I'll send the CPUs, but we had better get a good article with practical advice(like what we *can* do about bumpy CPUs!). Oh, and 1 of the Athlons is actually a Duron, but I think it still works so I will be keeping that as a backup. Just the 3 Tbird 1.4s and the Pentium 90 then. If you think the Duron has a chance of making it back alive (at least 50/50), I'll send that too.

pH: Ok, I know it would take a lot of your time then, but how much $$$ would make it worth your while?

BTW: I did not bust these cores personally! I get a lot of broken equip from friends who busted stuff trying to do it themsleves. All cores are actually in pretty good condition, one is flawless but refuses to boot for some reason.

[BillA]

if its alive, and you want it back, don't send it

and THANKS

why so concerned about what we think about your (er, friends) killing chips ?
its cool

[pHaestus]

All these dead chips lying around means that at least the o/cers here are honest enough (or destructive enough) not to try and RMA them and make AMD pay for their ****ups.

Bill: I went and looked about and I fear I threw away those dead CPUs when I moved. Sorry. I'll dig around in some less likely spots this evening but it doesnt look good.

I threw away or sold a bunch of heatsinks and fans and parts and such before packing; they say if a box doesn't have $100 in stuff in it then it is cheaper to throw away and buy at new home than to move it...

[koslov]

Hey, I take pride in not toasting expensive parts! It's just that people generally tend to associate chipping cores with noobishness, and tend to associate chipping FOUR with mental disability! Just wanted to clarify so no one would get the wrong impression, I am not some Athlon sadist!

[BillA]

now, now
no one was confusing a necrocpuologist with being "some Athlon sadist!"

[Hixup]

ive got an aold p2 200 mhz that ill send ya!

[BillA]

ok
and THANKS

[Joe]

Originally posted by Hixup
ive got an aold p2 200 mhz that ill send ya!

Umm theres no such CPU as a P2 200 Mhz

[N8]

Some more info to help out:

Cleanliness is ULTRA important in taking flatness measurements with an optic flat. One piece of dust from the air will interfere with the air gap and therefore the interference fringes.

Same with any surface roughness on the optic flat or the surface (hence why you need a polished surface.)

If the inteference fringes (rings) are close together (like in most of your photos), this is 'usually' an indication that there is some dust inbetween the surfaces, OR the surface is not very flat to begin with. The farther apart the rings, the more of the surface inbetween the rings is the same level (flatter.)

True, the straightness (or curvature) of the lines is indicative of the flatness, BUT you can get false readings from one piece of dust.

I go through a lot of cans of air at work checking flatness (monochromatic light source, optics flats.) You can also get optic flats that are flat to different degrees. The ones I use at work are either 1/20 wavelength or 1/10 wavelength.

Pefectly straight lines mean perfectly flat. The more curved, the more out of flat the measured surface is.

Bill, I have a really informative handout that I can scan or send or fax to you on measuring flatness using optic flats, maybe by tomorrow night (I have it at work.) I will probably scan it for best quality and email it to you.

Let me find some pix and post them here in a little bit.

[Aardil]

Got a dead athalon 1700+ I can send to ya, only slightly Abused.
Will probably work for your flat test but I wouldnt put any power to it.

Had a dead 800 too but cant find it, you dont suppose I actually threw something away?
NO WAY its gotta be on this desk somewhere.
I will try to locate it and send it along too.

Aardil

[BillA]

good to hear from you N8, not too many into this
by all means, send the info - I have the Van Keuren stuff and always looking for more
(very little on the web ??)

indeed cleanliness quite is the deal, and with my daughters cat it is a battle I will never be done with
the line spacing with most of these is really related to their lack of flatness
the 1/10 wavelength flat is 3"sq so any 'bad' spot jacks the whole thing up

my e-mail is on the site linked to above (2ed post @ bottom)

Aardil,
many Thanks

EDIT: N8
if it appears to you, or you even suspect, that I'm doing something incorrectly -> please voice your thoughts
I'd hate to be making (yet more) 'dumb' mistakes for months on end 'cause I was clueless
Thanks

[N8]

Just some pix I already have up on the server: These are an Alpha 8045 heatsink. The first pic is the factory surface under an optic flat. It had a lot of machining marks and was not a flat and/or polished enough surface to show any interference fringes.

The second pic is after lapping and several polishing steps down to 1 micron diamond.

The third pic is this polished surface under the optic flat. You can see the interference fringes (rings). As BillA stated, these rings are very hard to take photos of. You can see that most of the center area of the heatsink is within one ring and the rings get closer together toward the edge (the edges are rounded off, although to a 'very' slight amount.) Getting a large surface like this to minimal rings is hard to do (tooting my own horn.)

[BillA]

flat indeed !!
- helium lamp ?

I'm thinking I know how to avoid the crown in the center
I'll see if I can document it in the next several days

[koslov]

Just felt compelled to ask: Why are you measuring CPU flatness if lapping them will kill them anyway? Or is this just an experiment to add to the "good to know but not necessarily practical knowledge" database?

[BillA]

at issue is the bp flatness, for wbs or hsfs - same issue
since the bp 'mates' against the die, a starting point in the evaluation is the assessment of the die flatness:
concave ? convex ? flat ? what kind of ranges ?
irregular topography ? (look at the P3 above !)

THEN one looks at the effect of different flatnesses on the die temp
(everything else being QUITE controlled of course)
and from this one can hopefully conclude how much flatness 'makes sense'
- no point in chasing 2A° if the discernable effect threshold is 10A°
(for instance)

just learning more about what we are trying to do

[N8]

Originally posted by koslov
Just felt compelled to ask: Why are you measuring CPU flatness if lapping them will kill them anyway? Or is this just an experiment to add to the "good to know but not necessarily practical knowledge" database?

Lapping your processor will not always kill them. You are more likely to kill it with a bare AMD core as compared to a P4 with the IHS (integrated heat spreader.)

When I first got my P4 1.8a last May, it had a HUGE bulge in the IHS. This was even after running it for a while with the tremendous pressure of the also bulging stock P4 heatsink. They were basically only touching each other in the center of the IHS.

I lapped and polished the P4 IHS to flatten it out. Here is a pic where you can see that I had lapped a LOT of the IHS off down into copper, but only in the center. I had to remove a lot of material to get this thing flat.

Not only did I not kill the cpu, but it runs at 2600 MHz on air with a somewhat quiet 80mm fan (MCX4000 heatsink), 39-40 degrees C at 1.65v. Pretty good for one of the earlier 1.8a's.

I do plan on removing the IHS completely in the future to see what kind of temp difference it makes.