Testing large number of processors

[A Jones]

I have a large inflow of processors that I'm acquiring
from liquidators. A significant percentage of these
are defective, and I'm reselling them to people who
expect all working processors, so they must be
tested.

I need some advice on how to go about this:

1. How do I provide heat dissipation during testing?
To be safe, do I need to use apply thermal paste or a
thermal pad to every one? That'll get expensive.
Do I need to clamp down the heatsink on every one?
That'll get very tedious. Is there any safe shortcut?

2. If I put an incompatible processor in a motherboard,
can I damage the chip and/or the motherboard?

3. How extensive a test do I need to perform? If a
processor posts or boots, how likely is it that it will have
some subtle defect that will take more extensive testing
to find?

Thanks

 

[Al Dykes]

I have a large inflow of processors that I'm acquiring
from liquidators. A significant percentage of these
are defective, and I'm reselling them to people who
expect all working processors, so they must be
tested.

I need some advice on how to go about this:

1. How do I provide heat dissipation during testing?
To be safe, do I need to use apply thermal paste or a
thermal pad to every one? That'll get expensive.
Do I need to clamp down the heatsink on every one?
That'll get very tedious. Is there any safe shortcut?

2. If I put an incompatible processor in a motherboard,
can I damage the chip and/or the motherboard?

3. How extensive a test do I need to perform? If a
processor posts or boots, how likely is it that it will have
some subtle defect that will take more extensive testing
to find?

Thanks

Since you'll be selling these CPUs as used with return privilages I
suggest you'll net the most cash in you just give people a money back
guarantee.

As you suspect, testing to the point you can give a better guarantee
would be very expensive and involve testing to timing and voltage
margins.

 

[A Jones]

Since you'll be selling these CPUs as used with return privilages I
suggest you'll net the most cash in you just give people a money back
guarantee.

As you suspect, testing to the point you can give a better guarantee
would be very expensive and involve testing to timing and voltage
margins.

That'll work up to a point, but many liquidation lots have 50% or
more dead processors. If a customer has to return half the processors
I send them, they won't be a customer very long.

A reasonable strategy might be to resell all the processors that post.
I'm assuming that if a processor posts, chances are pretty good that
it is fully functional. Do people here agree?

 

[GSV Three Minds in a Can]

Bitstring <[email protected]>, from the

A reasonable strategy might be to resell all the processors that post.
I'm assuming that if a processor posts, chances are pretty good that
it is fully functional. Do people here agree?

Yes; That's probably more testing (in terms of cycles/time, if not
parametrics) than Intel or AMD gave it originally. Testing processors
for all possibly failure modes is ludicrously expensive, Joe Consumer
would rather have it $50 cheaper with a 1% false-pass and a money back
warranty.

Joe Military/NASA/Etc. has a different viewpoint and deeper pockets.

 

[A Jones]

Bitstring <[email protected]>, from the


Yes; That's probably more testing (in terms of cycles/time, if not
parametrics) than Intel or AMD gave it originally. Testing processors
for all possibly failure modes is ludicrously expensive, Joe Consumer
would rather have it $50 cheaper with a 1% false-pass and a money back
warranty.

Joe Military/NASA/Etc. has a different viewpoint and deeper pockets.

OK, that makes sense. I still have to work out heat dissipation during
testing. I've heard that a P4 can be damaged in a matter of seconds
if it doesn't have proper heat dissipation. Do I need to put thermal
paste clamp down a heatsink on every one of these?

 

[GSV Three Minds in a Can]

OK, that makes sense. I still have to work out heat dissipation during
testing. I've heard that a P4 can be damaged in a matter of seconds
if it doesn't have proper heat dissipation. Do I need to put thermal
paste clamp down a heatsink on every one of these?

For 'just POST' (and maybe boot into Windows) you could probably get
away with no paste but just a HS. I've done it with no fan before now.
8>.

The P4s, and modern Athlons, all thermally throttle (the Athlon probably
needs a motherboard which co-operates). They'll lock up before they
self-destruct (usually) anyway. The scare stories are about old Athlons
running with no HS at all .. not anything halfway modern with a lousy
one.

 

[A Jones]

For 'just POST' (and maybe boot into Windows) you could probably get
away with no paste but just a HS. I've done it with no fan before now.
8>.

The P4s, and modern Athlons, all thermally throttle (the Athlon probably
needs a motherboard which co-operates). They'll lock up before they
self-destruct (usually) anyway. The scare stories are about old Athlons
running with no HS at all .. not anything halfway modern with a lousy
one.

Ok, that's good to know. However, I find that the mechanism for
clamping and unclamping a heatsink on just about any Intel processor is
incredibly awkward. I'd almost rather put some thermal paste and rest the
heatsink on without clamping it -- if that works. It'd be more expensive
of course, but it'd be longer before I go nuts and the guys with the
straight jackets have to come.

What about Pentium IIIs? I'm getting fairly large numbers of them. The
slot 1's are okay because they usually have heatsinks attached. The
socket 370's usually come with no heatsink, so they're a problem. I
think they have no throttle mechanism. How fast can you damage one of
those?

 

[Robert Redelmeier]

Yes, but AFAIK both Intel and AMD spend considerable time
in testing (15+ minutes power-on). They use expensive gang
testers, starting with bed-o-nails (and cooling block?) before
the wafer is even sawed. After packaging, they still need
to speed test for binning, and I don't think that can be done
in under 5 minutes even with preheated blocks.

Yes, but even for Dell a dead CPU costs more than the device's
purchase price. A subtly defective one costs _far_ more.
(Pentium FDIV bug).

For 'just POST' (and maybe boot into Windows) you could
probably get away with no paste but just a HS. I've done
it with no fan before now.

No fan is much easier than no paste. The fan is only needed after
the block heats up, and that takes minutes. For a temporary TIM,
I'd suggest Vaseline (microcrystalline petroleum wax) if there
are exposed electrical points on the CPU topside, or K-Y Jelly
(thickened isotonic saline) if there are no conductivity issues.
The water in the latter is an _excellent_ thermal conductor.

CPUs do a BIST (Built-in self test) and the BIOS tests them
further, but you won't catch everything. It might not
catch a fried address line like A30.

There's _a lot_ to CPU testing, and Keith can comment
more knowledgably than me.

-- Robert

 

[Robert Redelmeier]

Ok, that's good to know. However, I find that the mechanism
for clamping and unclamping a heatsink on just about any
Intel processor is incredibly awkward. I'd almost rather
put some thermal paste and rest the heatsink on without
clamping it -- if that works.

It will, so long as the heatsink can fit dead flat
and isn't torqued around by the HSF or it's cable
or any of the attachment hardware.

What about Pentium IIIs? I'm getting fairly large numbers

Die back like Athlons. Supposed to shutdown on overheat,
but if the heatsink is in edge contact, some parts may
see overheat damage just before shutdown.

-- Robert

 

[Del Cecchi]

Yes, but AFAIK both Intel and AMD spend considerable time
in testing (15+ minutes power-on). They use expensive gang
testers, starting with bed-o-nails (and cooling block?) before
the wafer is even sawed. After packaging, they still need
to speed test for binning, and I don't think that can be done
in under 5 minutes even with preheated blocks.

I would be willing to bet that there is some preliminary test in there to
get rid of the bad ones first, perhaps even with a reduced set of pins and a
cheap tester.

Yes, but even for Dell a dead CPU costs more than the device's
purchase price. A subtly defective one costs _far_ more.
(Pentium FDIV bug).


No fan is much easier than no paste. The fan is only needed after
the block heats up, and that takes minutes. For a temporary TIM,
I'd suggest Vaseline (microcrystalline petroleum wax) if there
are exposed electrical points on the CPU topside, or K-Y Jelly
(thickened isotonic saline) if there are no conductivity issues.
The water in the latter is an _excellent_ thermal conductor.

Saline? You are going to put salt on there? How are you going to get it
off again?

CPUs do a BIST (Built-in self test) and the BIOS tests them
further, but you won't catch everything. It might not
catch a fried address line like A30.

There's _a lot_ to CPU testing, and Keith can comment
more knowledgably than me.

-- Robert

He just needs something to get the fallout down to a level low enough to not
piss off the customer that knows they are buying junk in the first place.
Are you old enough to remember PolyPacks?.

del cecchi

 

[A Jones]

CPUs do a BIST (Built-in self test) and the BIOS tests them
further, but you won't catch everything. It might not
catch a fried address line like A30.

There's _a lot_ to CPU testing, and Keith can comment
more knowledgably than me.

I'm really only looking for practical issues here. The
testing that Intel and AMD do before shipping the processor
is not necessarily relevant to my situation.

Likewise, if it's common for a single address line to burn out
in a chip that still posts, I need to consider it. If it's
something that might happen only occasionally, I can't
afford to test for it. I'm not doing research on these
chips, or certifying them for use in the space shuttle; I'm just
trying to make a living.

 

[Robert Redelmeier]

I would be willing to bet that there is some preliminary
test in there to get rid of the bad ones first, perhaps
even with a reduced set of pins and a cheap tester.

Yes, there probably is a quickie test after packaging to
test for open/shorted pins.

He just needs something to get the fallout down to a level
low enough to not piss off the customer that knows they
are buying junk in the first place. Are you old enough to
remember PolyPacks?.

I'm more than old enough (just shy of 30h), but don't remember
anything by that name.

-- Robert

 

[Robert Redelmeier]

Saline? You are going to put salt on there?
How are you going to get it off again?

With a tissue, of course. Isotonic is only 0.8%w.
The nanogram residue won't do anything to plastic
or silicon but could to copper.

-- Robert

 

[Robert Redelmeier]

I'm really only looking for practical issues here.
The testing that Intel and AMD do before shipping the
processor is not necessarily relevant to my situation.

True. After all, these CPUs once _did_ pass.

Likewise, if it's common for a single address line to burn
out in a chip that still posts, I need to consider it.

Check the pinout. I consider any of the outside pins
likely candidates for ESD.

-- Robert

 

[Del Cecchi]

With a tissue, of course. Isotonic is only 0.8%w.
The nanogram residue won't do anything to plastic
or silicon but could to copper.

-- Robert

Nanogram? I don't think so. And look up Sodium and threshold instability.
Goes back aways.

del

 

[Del Cecchi]

Yes, there probably is a quickie test after packaging to
test for open/shorted pins.


I'm more than old enough (just shy of 30h), but don't remember
anything by that name.

-- Robert

It was a company that sold bags of parts, sometimes "unmarked untested" for
low prices through ads in magazines like popular electronics. They have
apparently vanished into the mists some time ago.

del

 

[Tony Hill]

OK, that makes sense. I still have to work out heat dissipation during
testing. I've heard that a P4 can be damaged in a matter of seconds
if it doesn't have proper heat dissipation. Do I need to put thermal
paste clamp down a heatsink on every one of these?

Generally speaking, a P4 processor will run without thermal paste for
about 5-10 minutes before it starts overheating (maybe less time if
your heatsink is already warm from the last chip tested). Even then a
sufficiently powerful fan and decent heatsink you can keep it cool
enough that the system won't crash (though the chip might start
thermal throttling). So, for your purposes, if you're just doing a
POST test (or even booting into Windows) you should probably be fine
without the paste.

On the other hand, if the heatsink isn't clamped down, the processor
will usually overheat and shut itself off in less than 15 seconds.
Most chips will shut themselves down without being damaged, but if
you're dealing with large volumes then you probably will run into some
that are damaged. In short, not clamping down the heatsink probably
isn't an option here.

Now I'm afraid that I can't say TOO much about the testing of CPUs,
but I can add my little $0.02 worth. I deal almost exclusively with
computers from major OEMs. Now I'm sure that these OEMs don't put a
huge amount of emphasis on testing CPUs, though I know that they at
least ensure the system boots to Windows. With a relatively minimal
amount of testing, I can say for certain that the CPUs that do make it
out to the field almost never die. Yeah, it does happen occasionally,
but FAR less frequently than just about any other component in a
computer. Sadly though, motherboards DO die a lot, and *OFTEN* people
will blame a motherboard (or memory, or power supply) problem on a
CPU. I heard a stat a while back that said that 95% of all CPUs
returned as defective worked flawlessly. The moral of this story is
that you should expect a reasonably high number of returns regardless
of how well you test your chips.

 

[A Jones]

No fan is much easier than no paste. The fan is only needed after
the block heats up, and that takes minutes. For a temporary TIM,
I'd suggest Vaseline (microcrystalline petroleum wax) if there
are exposed electrical points on the CPU topside, or K-Y Jelly
(thickened isotonic saline) if there are no conductivity issues.
The water in the latter is an _excellent_ thermal conductor.

So let's say I go with Vaseline. Then there's no controversy about
whether the saline is a problem. Can I use Vaseline as a cheap
substitute for thermal paste that's adequate for a post or
even a boot to Windows?

If so, this is starting to sound like an attractive shortcut:
1. I coat the processor with Vaseline.
2. I rest a clean heatsink carefully on top of the processor
so it rests flat, but I don't clamp it down.
3. I do a post or even a boot to Windows.
4. If it boots, I sell it. If the buyer has a problem, I refund
or replace.

Does anyone have a practical objection to this approach?

 

[Robert Redelmeier]

Nanogram? I don't think so.

Don't think about numbers -- calculate!

How much gel is left after wiping? Max 5um on the flat top
and 15um on the edges? That gives 800 nL or 800 ug. At 0.8%w
NaCl, that is 6400 ng. Call it low microgram if you wish.

Think if it a different way. You apply about one drop (50uL).
Good cleaning should remove 98+% leaving 1000 nL or 8000 ng.

And look up Sodium and threshold instability.
Goes back aways.

I did and got no good hits on Google. Can you be
more specific? Certainly sodium can contaminate printed
semiconductors, but it's got to get there. How is it going
to migrate through 1+mm resin or 0.5 mm hi purity silicon?

-- Robert

 

[keith]

It was a company that sold bags of parts, sometimes "unmarked untested" for
low prices through ads in magazines like popular electronics. They have
apparently vanished into the mists some time ago.

Yep, "unmarked untested", but we know it has (or once had) 16 pins. They
were famous for selling "Digital IC - 16 pins - $.50". I did see some with
curious 7-digit part numbers in the form 'NNANNNN' (like 39F1234). ;-)

Their parts usually failed the floor-sweepings test. In reality they were
best used as thumb tacks.