Opteron Overclocking?

Has anyone tried to overclock an Opteron based systems?

Adrian Richards wrote:
Has anyone tried to overclock an Opteron based systems?

Yes. (not personally, can;t remember where) They scale well 'cos the
memory controller is onboard.

Ben
--
I'm not just a number. To many, I'm known as a String...

"Adrian Richards" wrote in message
...
Has anyone tried to overclock an Opteron based systems?



Is it really worth melting something as new / expensive as an opteron? too
expensive as yet to be worth overclocking

HamMan wrote:
"Adrian Richards" wrote in message
...
Has anyone tried to overclock an Opteron based systems?



Is it really worth melting something as new / expensive as an opteron? too
expensive as yet to be worth overclocking

As far as I know, they run pretty cool thanks to SOI, a heatspreader,
relatively low clock and only 1.55V.

I reckon you could pump a whole load of voltage in there and not damage it.
Maximum rated thermal dissipation is 80W, but since AMD plan not to revise
this figure to make upgrading easier (no new HSF required) it's probably
considerably lower than that right now.

Ben
--
I'm not just a number. To many, I'm known as a String...

"Ben Pope" writes:

As far as I know, they run pretty cool thanks to SOI, a heatspreader,
relatively low clock and only 1.55V.

How is it any thanks to a heat spreader? A64 is like an AXP with an
extra layer of aluminum and TIM between the chip and the HSF. And AMD
chooses the type and thickness of one of the two TIM layers (making
TIM reviews even less useful). These would tend to make the CPU run
hotter than it would otherwise.

CPU heat spreaders should be called "CPU protectors", because they're
there mostly to protect the edge of the chip from being chipped by the
HSF and probably to discourage people from messing with bridges.

Gary W. Swearingen wrote:
"Ben Pope" writes:

As far as I know, they run pretty cool thanks to SOI, a heatspreader,
relatively low clock and only 1.55V.

How is it any thanks to a heat spreader?

Getting the thermal interface good is less important as you have much more
area.

A64 is like an AXP with an
extra layer of aluminum and TIM between the chip and the HSF. And AMD
chooses the type and thickness of one of the two TIM layers (making
TIM reviews even less useful). These would tend to make the CPU run
hotter than it would otherwise.

It's not that much like an XP is it really?

CPU heat spreaders should be called "CPU protectors", because they're
there mostly to protect the edge of the chip from being chipped by the
HSF and probably to discourage people from messing with bridges.

Agreed that they protect the chip. As to whether they were put there to
discourage people from playing with bridges is probably debateable. I don't
think it was any accident that XPs are so easy to overclock.

Ben
--
I'm not just a number. To many, I'm known as a String...

"Ben Pope" writes:

Gary W. Swearingen wrote:
How is it any thanks to a heat spreader?

Getting the thermal interface good is less important as you have much more
area.

Getting it good is just as important (more, I think) because of the
added thermal resistance of the extra layer of TIM and package lid.
You're hoping that AMD gets the chip-to-lid thermal interface good on
the small chip surface and your lid-to-heatsink thermal interface job
is made more difficult because you now have more area to cover with a
good layer of TIM. The lid doesn't do much heat spreading anyway
because the heat sink does it so much better than the thin lid.

A64 is like an AXP with an
extra layer of aluminum and TIM between the chip and the HSF.
....
It's not that much like an XP is it really?

Like I said, it's like an XP with an extra layer of TIM and the
lid/spreader. It's almost twice as big in area as the XP which
probably helps spread the head some, but that's a different issue; it
would still be better to put the heatsink on the "bare" chip. (From
some thermal plots I've seen, small parts of the chip can be much
hotter than others, leading me to believe that the chip considerably
less heat-conductive than aluminum or copper, but I don't really
know.)

Gary W. Swearingen wrote:
"Ben Pope" writes:

Gary W. Swearingen wrote:
How is it any thanks to a heat spreader?

Getting the thermal interface good is less important as you have much
more area.

Getting it good is just as important (more, I think) because of the
added thermal resistance of the extra layer of TIM and package lid.
You're hoping that AMD gets the chip-to-lid thermal interface good on
the small chip surface and your lid-to-heatsink thermal interface job
is made more difficult because you now have more area to cover with a
good layer of TIM.

Oh come on... twice the area gives you twice the ability to transfer heat.
Which means that pe sq mm you don't need as good a thermal interface to
dissipate the same energy.

The lid doesn't do much heat spreading anyway
because the heat sink does it so much better than the thin lid.

So only the middle bit of the lid heats up? I doubt it.

A64 is like an AXP with an
extra layer of aluminum and TIM between the chip and the HSF.
...
It's not that much like an XP is it really?

Like I said, it's like an XP with an extra layer of TIM and the
lid/spreader.

So it's like an XP...

It's almost twice as big in area as the XP which

But twice as big, and uses SOI which drsatically reduces the capacitance of
the transistors in the chip, reducing power requirements and heat
production.

So when we're talking about the thermal characteristics, it's not that
similar, is it?

probably helps spread the head some

So now area is important in transferring heat?

but that's a different issue; it
would still be better to put the heatsink on the "bare" chip.

Only if you do a better job of the thermal interface than AMD.

OK, I concede. Maybe the heat spreader doesn't always allow an increased
thermal dissipation from the chip. Without hard evidence, neither of us
know for sure.

Ben
--
I'm not just a number. To many, I'm known as a String...

"Ben Pope" writes:

Gary W. Swearingen wrote:
"Ben Pope" writes:
Gary W. Swearingen wrote:
How is it any thanks to a heat spreader?

Getting the thermal interface good is less important as you have much
more area.

Getting it good is just as important (more, I think) because of the
added thermal resistance of the extra layer of TIM and package lid.
You're hoping that AMD gets the chip-to-lid thermal interface good on
the small chip surface and your lid-to-heatsink thermal interface job
is made more difficult because you now have more area to cover with a
good layer of TIM.

Oh come on... twice the area gives you twice the ability to transfer heat.
Which means that pe sq mm you don't need as good a thermal interface to
dissipate the same energy.

When you said "more area", I thought you were talking about the top of
the lid (the "heat spreader"), since I was discussing your comment
on those. Now you're talking about "twice the area" which leads me to
think that you're thinking about the chip (die) area. I agree with
the gist of your statement above on that subject. (But I can quibble
with that too: 1) You're going to dissipate the same energy, no matter
how good or bad your thermal interface is; the important issue is how
hot your chip (actually, the hot spots of the chip) are going to be.
2) If 95% of the heat is coming from a small portion of the chip where
most of the calculations are occuring, then it barely matters how big
the chip is. That's an exageration, of course, but you should get the
point. Most of the heat from such hot spots flows up through the thin
chip, the TIM, any lid and extra layer of TIM, and into the heat sink,
and much less first flows sideways through the chip and/or the lid.)

Yes, of course a much bigger chip producing about the same heat can be
said to run cooler; you won't (and haven't) gotten any argument from
me on that. But the question was whether a lid helps it run cooler.

Now, I must admit that a lid CAN help a CPU run cooler in some
circumstances that shouldn't matter to overclockers: Like with no heat
sink or with a VERY crummy one or maybe even with a copper or silver
lid with a cheap (thin) aluminum heat sink. The copper "cores" of
some aluminum heat sinks is a kind of heat spreader, but you'll notice
that they tend to be much thicker than CPU lids (though that might
just be eye candy). But AMD lids look like aluminum and overclockers
use good heatsinks, often with copper bottoms, at least. (It would be
interesting to know the thermal resistance of the copper-aluminum
interface in Cu-Al heat sinks.) But someone desiring the coolest
possible chip would not use an extra TIM layer, if he had control over
the complete design and didn't have to use a very cheap heat sink or
worry about chip chrushing.

The lid doesn't do much heat spreading anyway
because the heat sink does it so much better than the thin lid.

So only the middle bit of the lid heats up? I doubt it.

And I didn't say or even imply that. I implied that the bulk of the
heat flows into the heat sink without flowing through the parts of the
lid that don't touch the lid. Some of the heat does, of course; how
much depends on the thickness of the lid.

There are two ways to look at this that should help you see my point.
With both, assume that the lid and heat sink are made of the same
material and that we're only discussing one type of CPU chip.

1) Consider lidded and unlidded CPUs with heat sinks. If some
imaginary TIM that had the same thermal resistance as the lid and sink
was used between the lid and sink, then there would be little or no
difference between a lidded and unlidded CPU; the thermal resistances
are nearly the same for both. It's as if the lid was the bottom of the
heat sink. But now replace the imaginary TIM with real TIM and you
should see that you've made the situation worse. It would be better
to not have to use that layer of TIM, as with the unlidded CPU.

2) Think of an unlidded CPU with a mounted heat sink. Now imagine
that you cut the heat sink a bit above the CPU chip and use the
resulting slab as the lid of a lidded CPU. You'll need to add another
layer of TIM between the lid and sink, and the result can only be
worse than if you had stuck with the unlidded CPU.

BTW, while the non-middle bits of the lid do heat up, they don't heat
up as much as the middle bits. Again, think of the lid as just a
lousy bottom of your heat sink. Even if it was a good bottom of a
heat sink, the parts of the sink (and lid) away from the chip are
cooler than those near it; that should be fairly obvious. Think of
the heat flowing out of the chip surface up through the
TIM/lid/TIM/sink in a nearly hemispherical pattern.

The lid design effectively slices a layer off the bottom of the sink
and then puts it back on with TIM, with several bad side effects: The
chip TIM is chosen by AMD instead of you, and it can't be replaced
when degraded through a combination of overclocking and time. The
thin layer is possibly changed from copper to aluminum. An extra
layer of TIM is introduced. The extra layer has a large area which
makes it harder to get a thin even application of TIM and makes it
essentially impossible to apply the same amount of pressure to both
layers of TIM (so the outer layer is not as well pressed as one would
desire, without putting too much pressure on the chip, despite the
lid) (unless the CPU lid is strong enough to avoid transferring much
force to the chip, which I doubt).

A64 is like an AXP with an
extra layer of aluminum and TIM between the chip and the HSF.
...
It's not that much like an XP is it really?

Like I said, it's like an XP with an extra layer of TIM and the
lid/spreader.

So it's like an XP...

Yes, it is. They both run hotter than they could without a "heat
spreader". You must not be used to this common English figure of
speech (called a "simile") in which two essentially unlike things are
compared in some "like" characteristic(s). It's done all the time,
like in "John is like Einstein with no cerebrum." Is John like
Einstein? Yes (and no). I hope that helps.

It's almost twice as big in area as the XP which

But twice as big, and uses SOI which drsatically reduces the capacitance of
the transistors in the chip, reducing power requirements and heat
production.

Yeh, yeh. And many more transistors, both in the caches and in the
high-temp computing parts of the chip which have more bits. But none
of this has anything to do with the issue of "heat spreaders".

So when we're talking about the thermal characteristics, it's not that
similar, is it?

But I wasn't talking about thermal characateristics; I was talking
about whether heat spreaders help CPUs run cooler. Like I said in the
same sentence, the size of the chip is a different issue. In this
discussion, I consider the AXP and A64 thermal characteristics to be
effectively identical to each other and to any other heat-producing
chip.

probably helps spread the head some

So now area is important in transferring heat?

I didn't say anything there about transferring heat; I was referring
to the fact that the sources of heat are more spread out on the
A64. (The max heat generation of the two CPUs are not much different.)

but that's a different issue; it
would still be better to put the heatsink on the "bare" chip.

Only if you do a better job of the thermal interface than AMD.

Nope. Both me and AMD can do good jobs with our two interfaces and
still wind up with a hotter-running chip than yours with a single poor
interface (unless it's TOO poor, of course).

OK, I concede. Maybe the heat spreader doesn't always allow an increased
thermal dissipation from the chip. Without hard evidence, neither of us
know for sure.

Without hard evidence or convincing analysis, both of us will have to
rely on our own analysis or on the handwaving of AMD's and Intel's
marketing people. Pick one.

On Sat, 04 Oct 2003 00:05:58 GMT, (Gary W.
Swearingen) wrote:

..
Without hard evidence or convincing analysis, both of us will have to
rely on our own analysis or on the handwaving of AMD's and Intel's
marketing people. Pick one.

I've gotten a real kick out of how the "free" press is so silent on
AMD's 64bit release.

Almost as funny as the day the Linux Convention opened in SF and the
business section of the local paper was almost 100% Microsoft
articles.


Here's a post by Roger Hamlett over at the Asus newsgroup on his
experience with Opteron.

On Sat, 27 Sep 2003 13:59:02 +0100, "Roger Hamlett"
wrote:


...

There are two very impressive features of both the Opteron, and the
Athlon64. The first is how cool they run!. At first I thought something was
wrong, because the cooling fan doesn't start with the machine, and only came
on after several minutes. The second is that the run 32bit applications
extremely well.
I tried the Linux64 implementation (for a database), and VMware, running XP
as a 32bit application on this. gave me the best of both worlds (64bit
address range, and 8GB RAM on the Opteron), for the database, and could
still run all the 32bit applications (which in general don't need the larger
address space).

Best Wishes