Glaciators, The Emperor Speaks! :) [Archive]

taz1004

06-18-01, 09:56 AM

Article is great and certainly something to consider but what other choice do users have other than MBM? I have hard time believing inaccuracies up to 20'C but let's say it's true and Glaciators is the best heatsink. If MBM reading goes above 70'C, should the user just assume that MBM is inaccurate and it's really only 50'C and don't worry about it? Or should they all start drilling holes in their heatsink and place another thermistor in it?

How is it possible that in this day and age there could be 20'C inaccuracies with temp readings? 20'C difference can kill a CPU. Unless chipsets with better temp monitoring system is in place, heatsink manufacturers have to deal with current system and it is their responsibility to design a heatsink that will not only perform, but also report correct (or close enough) temp reading for that chipset. We can't just take the heatsink manufacturer's word and assume that it's actually working well when MBM reports 70'C. Weather it reports higher or lower temp, misleading temp reading is a design flaw in the heatsink nevertheless. Otherwise, heatsink manufacturer will keep blaming chipset for inaccurate sensor and chipset manufacturers will blame heatsink and the users will get screwed in between.

Reminds me of Ford and Firestone. Regardless of whose fault it is, Ford Explorer owners will avoid Firestone... not the other way. People will not go out and by 4 sets of Firestone first and say "hmm, Explorer wont work with my tires so I'll buy a Chevy" It is responsibility of Firestone to make their tires work with Explorer. Likewise, designing a heatsink that reports correct temp for the chipset it is designed for should be one of the design goal of the heatsink manufacturer.

Car manufacturers will not design a car based on certain brand of tires and chipset manufacturers will not design a thermal sensor for certain brand of heatsinks.

--BrianC

castle lager fan

06-18-01, 12:25 PM

So using watercooling will be the only way? No "hot air" blowing about! :-)

Can someone test the accuracy of MB's temp readings with watercooling and then compare with HSF. A benchtest with a constant heatsource will also help settle the issue?

Just asking as it seems we will be going in circles for a long time.

Tambo

06-18-01, 12:37 PM

very well said and thought out, taz.

Spewn

06-18-01, 02:24 PM

Keep in mind taz, if you're trying to sell an hsf, you're trying to sell it to the people who read sites like this. If you can make a heatsink that "fools" the system into thinking it's getting better temps than it actually is, and a bunch of hardware review sites hype up your heatsink because of that, that means you get business. As far as the firestone vs ford explorer analogy goes, it's a little different here. It's not that heatsinks don't WORK with the motherboard's, it's that they cause the system to report innacurate temperatures. Reporting an accurate temperature has nothing to do with how well the sink is actually doing it's job. It's up to motherboard/cpu manufacturers to provide us with a means of getting an accurate temp reading, regardless of what heatsink we're using.

Yomama

06-18-01, 02:44 PM

stroligo (Jun 17, 2001 02:20 p.m.):
Seriously, if you haven't seen it already, I posted an email from Andy LeMont explaining why you're getting the MBM readings you're getting, along with how MBM can be manipulated: you can design a heatsink that cools the CPU worse but gives you a better MBM number:

Article here (http://www.overclockers.com/tips510).

Probably later today, Joe is also going to post an article on the same subject.

Ed - I ususally concur with your findings, but this one is hard to understand. If I read the article correctly, you are impying that temparture misreadings may be caused by over-heating or "over-cooling" (by excess air from the heatsink) of the caps on the side of the CPU. To that extent it should be able to reverse this effect - at least temporarily by isolating the caps from the warm or cold air, e.g. by means of a piece of cardboard or plastic.

I still think that what we are seeing is an inherent design issue of all copper heatsinks; they can absorb a lot of heat, but unlike aluminum they retain the heat, ultimately requiring more airflow to dissipate it. I am kind of busy right now, but I bet with a strong fan blowing from the front onto the cpu, one would be able to relieve the heatsink, and get lower temps overall.

For now I have to believe that despite the higher readings, the system is more stable, as I was able to reduce the voltage by 0.05 -0.1V. OTOH I was not able to o/c further, but this may be just a matter of getting rid of some more heat.

Yo

taz1004

06-18-01, 02:56 PM

Spewn, I'm not trying to pick on your post but I knew someone was gonna say what you've said and I can't resist. :)

"It's not that heatsinks don't WORK with the motherboard's, it's that they cause the system to report innacurate temperatures"

I disagree. Temperature monitoring is a feature (important one at that) and if the heatsink and certain motherboard reports inaccurate temp, it is NOT working properly. Like I've said, heatsink and motherboard manufacturers can point fingers all they want but the fact is that they're not working together properly. And since heatsinks are designed around motherboards and not the other way around, if the heatsink is reporting misleading temp readings, it is fault of the heatsink designer.

"Reporting an accurate temperature has nothing to do with how well the sink is actually doing it's job."

But if there's no easy means of determining weather the sink is actually doing its job... well... then how the heck do you know if it's doing its job? My point was that the MBM with the chipset monitoring feature is the most accesible means of measuring temp right now and the heatsink designer is saying ignore that. If such inaccuracies exist in that big margin, how are we supposed to know if my CPU is burning up or not? And in case of Glaciator, I overclocked my CPU and it is now reporting 70'C. Should I go higher since MBM is incorrect? If so, how high should I go?

"It's up to motherboard/cpu manufacturers to provide us with a means of getting an accurate temp reading, regardless of what heatsink we're using."

Can't argue with that. But do such means of getting perfectly accurate temp reading exist? From that article, it sounds like Intel's way of measuring temp within the die itself isn't accurate enough. But are they close enough? I think so. Should heatsink manufacturers design based on this imperfection? I believe so.

--BrianC

taz1004

06-18-01, 03:16 PM

Also, consider the fact that this article is based on email from the designer of the heatsink in question. I would be surprised if it isn't biased.

--BrianC

Badger

06-18-01, 04:17 PM

I think it's quite clear from the various reports that the Glaciator just don't perform quite as well as some of the other designs out there. The evidence is quite overwhelming and quite frankly the argument that somehow the heatsink design can give a significantly higher MBM reading (consistently in comparison to a veriety of heatsinks over a wide range of mobo's) while actually cooling the chip better is pretty unbelievable. Further evidence is the fact that even on die temp monitoring is showing higher temps, Phil's theory that the analogue signal from the thermal diode can somehow be corrupted (and this corruption somehow makes the monitor read higher only when a Glaciator is installed !) is even harder to swallow.
The only factor that explains all the high MBM readings being reported is that the Glaciator just doesn’t cool as well as some of the other heatsinks. I can't grasp why people have not come to this rather obvious conclusion.

I'm not Glaciator bashing here, it looks like a well designed cooler, but when it comes to no compromise ultimate cooling it's clear it can't compete with the far noisier high air flow solutions offered by other manufacturers.

Richard

06-18-01, 04:47 PM

I'll say this much. I seriously doubt any other heatsink would have been given such special consideration.

There's a lot of hoo-ha going on right now about which heatsink is the "best." Well, I'm pretty sure my PEP66 isn't the "best" heatsink, but it gets the job done. My 1.2 GHz Tbird runs rock solid all day long. Even in a rather warm room (26.6c) it chugs along just fine. Case temps report at a constant 39c. (Although, I have a hard time believing that.)

Yeah, there are better heatsinks. I could slap a loud as heck Delta 38 on this puppy and get even better cooling. The question is, why?

Would a Millennium Glaciator make an already stable system more stable? No it wouldn't. Would I be able to squeeze out more MHz? Possibly. At the expense of more noise and $40.

I've tried a lot of different coolers ranging from stock AMD to copper waterblocks. So why am I using a PEP66? It's a good heatsink that doesn't sound like a Harrier Jumpjet, and I just feel comfortable with a nice portable setup.

Sometimes its important to take a step back and weigh the pros and cons of all these coolers. A little objectivity goes a long ways. If you've already got a Thermoengine with a Delta Black Label fan. Do you honestly think you're going to get a heck of a lot better performance out of a conventional heatsink? I don't care how well manufactured the thing is. There's a limit, and we're seeing it. If AMD doesn't do something to drastically reduce the amount of heat these puppies are putting out, we'll all be running the ridiculous hairdryer coolers in our cases.

Until then I say, "Get a grip."

Badger (Jun 18, 2001 04:17 p.m.):
I think it's quite clear from the various reports that the Glaciator just don't perform quite as well as some of the other designs out there. The evidence is quite overwhelming and quite frankly the argument that somehow the heatsink design can give a significantly higher MBM reading (consistently in comparison to a veriety of heatsinks over a wide range of mobo's) while actually cooling the chip better is pretty unbelievable. Further evidence is the fact that even on die temp monitoring is showing higher temps, Phil's theory that the analogue signal from the thermal diode can somehow be corrupted (and this corruption somehow makes the monitor read higher only when a Glaciator is installed !) is even harder to swallow.
The only factor that explains all the high MBM readings being reported is that the Glaciator just doesn’t cool as well as some of the other heatsinks. I can't grasp why people have not come to this rather obvious conclusion.

I'm not Glaciator bashing here, it looks like a well designed cooler, but when it comes to no compromise ultimate cooling it's clear it can't compete with the far noisier high air flow solutions offered by other manufacturers.

stroligo

06-18-01, 05:07 PM

[quote]taz1004 (Jun 18, 2001 10:05 a.m.):
---Article is great and certainly something to consider but what other choice do users have other than MBM?

If you have a clock that's broken and cannot keep track of time accurately, and you can't afford another one, do you say, "It must be OK?" That's essentially what you're saying.

---I have hard time believing inaccuracies up to 20'C

We have documented this again and again and again, and so have other people, and all of this long before there was a Millenium heatsink. Have you looked at the articles. What do we have to do short of coming to your house, drilling holes in your heatsink, and doing it in front of you?

Well, actually, I pretty much had Joe do just that. Not because I didn't believe him, but because I wanted to be able to tell people that I've seen it with my own eyes. The MBM/thermistor in my particular machine reads at least 10C higher than the real temp. I grant you that is probably more inaccurate than most.

---but let's say it's true and Glaciators is the best heatsink.

We aren't saying it is. The way we presented the numbers may have left you with that impression, and we're going to have to explain what those numbers mean more in the future. What we are saying is that it performs on par with the best, notwithstanding MBM numbers.

----If MBM reading goes above 70'C, should the user just assume that MBM is inaccurate and it's really only 50'C and don't worry about it?

A person should always check for other reasons why the temperature might be so high; improper placement, bad application of thermal grease, high system temps, other reports of high temperature readings.

However, that sort of situation (which is what I have with the Master) occurs less often that the opposite, when MBM is lower than the actual core.

---Or should they all start drilling holes in their heatsink and place another thermistor in it?

If they need an accurate temperature reading, something along those lines would need to be done. However, most people don't need that level of precision; we put together that calculation so people could at least have an alternate means of checking their situation.

stroligo

06-18-01, 05:08 PM

---How is it possible that in this day and age there could be 20'C inaccuracies with temp readings? 20'C difference can kill a CPU. Unless chipsets with better temp monitoring system is in place, heatsink manufacturers have to deal with current system and it is their responsibility to design a heatsink that will not only perform, but also report correct (or close enough) temp reading for that chipset.

How is it possible that there's world hunger? How is it possible that Windows crashes? How is it possible that Intel and Via chipsets have bugs in them. It is. Saying that there shouldn't be doesn't change the "is."

---We can't just take the heatsink manufacturer's word and assume that it's actually working well when MBM reports 70'C.

If you ever actually end up in that situation, there are relatively cheap solutions that will yield pretty accurate numbers, look in our archives.

---Weather it reports higher or lower temp, misleading temp reading is a design flaw in the heatsink nevertheless. Otherwise, heatsink manufacturer will keep blaming chipset for inaccurate sensor and chipset manufacturers will blame heatsink and the users will get screwed in between.

When something breaks in your house, for the next six months, do you say, "It shouldn't be broken." What does that do?

---Reminds me of Ford and Firestone. Regardless of whose fault it is, Ford Explorer owners will avoid Firestone... not the other way. People will not go out and by 4 sets of Firestone first and say "hmm, Explorer wont work with my tires so I'll buy a Chevy" It is responsibility of Firestone to make their tires work with Explorer.

The tires work fine with every other brand except Ford Explorer. Hmmmm?

---Likewise, designing a heatsink that reports correct temp for the chipset it is designed for should be one of the design goal of the heatsink manufacturer.

None of them do! The sensor measurement are inherently inaccurate, they are not precise. This is like having a guy with very poor sight judging the Miss America contest, and you're saying it's the contestant's fault.

Provided you've done a mathematical check like we've suggested, and your CPU is in no danger of blowing, under normal operations, MBM will usually be good enough to tell you if your system is ready to blow up. It is not good enough to judge a beauty contest between heatsinks.

How many times do we have to tell you: the measure is inaccurate. It's inaccurate for all of them, just in different and varying ways, some more, some less.

Badger

06-18-01, 05:25 PM

stoligo,
What makes you think drilling a hole in the HS and sticking a thermister in is going to give you more accurate readings than the sensor in the socket?
A thermister mounted in a hole in the HS will be effected by the HS temperature as much as by the core temp. Obviously the HS temp will always be below core temp (otherwise the heat would not flow from core to HS) so any temps measured by this method is likely to give low readings.
Maybe your MBM temps are more accurate than you thought !.
Anyway when it comes to HS comparisons the accuracy of measurement is not so much the issue as the repeatability. I would consider the repeatsbility of the standard socket temp sensor better than the drilled HS method due to possible variences in the way the thermister is mounted in the various heat sinks.

stroligo

06-18-01, 05:29 PM

Ed - I ususally concur with your findings, but this one is hard to understand. If I read the article correctly, you are impying that temparture misreadings may be caused by over-heating or "over-cooling" (by excess air from the heatsink) of the caps on the side of the CPU. To that extent it should be able to reverse this effect - at least temporarily by isolating the caps from the warm or cold air, e.g. by means of a piece of cardboard or plastic.

No, it's not just a matter of hot air. Heat gets conducted a bunch of ways throught the secondary heat paths. What you suggest might reduce the problem, wouldn't eliminate it.

---I still think that what we are seeing is an inherent design issue of all copper heatsinks; they can absorb a lot of heat, but unlike aluminum they retain the heat, ultimately requiring more airflow to dissipate it. I am kind of busy right now, but I bet with a strong fan blowing from the front onto the cpu, one would be able to relieve the heatsink, and get lower temps overall.

Will the Glaciator do a bit better job with a 7K fan? Sure it will. I know of somebody who did just that. The MBM number still wasn't as low as the other, somewhat inferior heatsink.

You just won't believe MBM is inaccurate, will you? What do we have to do that would prove it to you?

I suspect the answer from many is, "Nothing." Well, we know that's wrong. If you insist that MBM is an accurate measurement, then when you look at a heatsink review, don't look at our other numbers. Just look at the MBM number, since that's the only one that counts in your book.

Do you know what? If you had, you would have seen that the Glaciator doesn't do so well in MBM numbers. Our test shows that.

---For now I have to believe that despite the higher readings, the system is more stable, as I was able to reduce the voltage by 0.05 -0.1V.

If MBM were so accurate, how could this be?

taz1004

06-18-01, 05:54 PM

I love it when someone disects my analogies yet leave my conclusion out. Let me make it simple. Glaciator is reporting higher temp than others. Designer claims that sensor is inaccurate and it's actually running cooler. WHAT TOOL DO USERS HAVE TO VERIFY THIS AND HOW ARE THEY SUPPOSED TO MONITOR IT? Oh... I see... guess?

Many other heatsinks even with similar design are reporting cooler temp thru MBM. Is it likely that these heatsinks are reporting higher temp than actual as well? Or is it just the Glaciator that's reporting higher temp yet all others are reporting lower temp? No other heatsink has been accused of misleading temp before this point and like Richard said, no other heatsink has given this much benefit of the doubt that it might actually be performing better. If it did, please point it out to me cause I might've missed it.

"The tires work fine with every other brand except Ford Explorer. Hmmmm?"

Exactly my point. Ford should redesign Explorer so it works with Firestone right? Explorer has been designed long before Firestone and it is designed with standard tires.

--BrianC

stroligo

06-18-01, 05:55 PM

---I think it's quite clear from the various reports that the Glaciator just don't perform quite as well as some of the other designs out there.

MBM is not an accurate measurement.

---The evidence is quite overwhelming and quite frankly the argument that somehow the heatsink design can give a significantly higher MBM reading (consistently in comparison to a veriety of heatsinks over a wide range of mobo's) while actually cooling the chip better is pretty unbelievable.

Have you looked at any of the articles we or other people have written on this issue? If you don't believe us, look at this (http://http://www.burningissues.net/hardware/review/Review .htm). This documents the Golden Orb. As much as a 30C difference. between
MBM and more accurate CPU readings. How is that, how could this be?

----Further evidence is the fact that even on die temp monitoring is showing higher temps, Phil's theory that the analogue signal from the thermal diode can somehow be corrupted (and this corruption somehow makes the monitor read higher only when a Glaciator is installed !) is even harder to swallow.

Secondary heatflows do that. Take a look at this (http://www.overclockers.com/articles306).

---The only factor that explains all the high MBM readings being reported is that the Glaciator just doesn’t cool as well as some of the other heatsinks.

No, there is another explanation. MBM is inaccurate. We've been saying that for a long, long time.

---I can't grasp why people have not come to this rather obvious conclusion.

Because it's wrong. Joe has seen this again and again and again and again. I've seen the same with my own eyes, watching simultaneous readings and seeing how damn inaccurate it is.

---I'm not Glaciator bashing here, it looks like a well designed cooler, but when it comes to no compromise ultimate cooling it's clear it can't compete with the far noisier high air flow solutions offered by other manufacturers.

Tell me this. We've done all kinds of measurements over the course of time, drilling holes through thermistors. Why do you think we do that?

We do that because AMD says that's how it has to be done for temperature testing. I quote from the way to do it. AMD Thermal, Mechanical, and Chassis Cooling Design Guide (http://http://www.amd.com/products/cpg/athlon/techdocs/pdf/23794.pdf), "Thermocouple Installation for Temperature Testing", page 13:

"To make accurate thermal measurements, drill a hole into the heatsink to obtain the heatsink base temperature.

Center the heatsink base thermocouple directly over the die and 2mm above the die as shown in Figure 7"

We do that, and we get much different temperatures than MBM.

Is AMD stupid, too?

stroligo

06-18-01, 06:26 PM

---I'll say this much. I seriously doubt any other heatsink would have been given such special consideration.

I'll tell you what happened here. We got the heatsink well ahead of others. Few review copies went out. People somewhat misinterpreted the test results (and we could have explained that a bit more).

Due to the particular design of the Glaciator, it doesn't do as well at MBM as some other heatsinks. Most people think MBM is accurate (though they didn't look at our MBM numbers), and concluded that the heatsink sucked and we
BSed a product. What I'm really defending is our integrity and our test methodologies.

---There's a lot of hoo-ha going on right now about which heatsink is the "best." Well, I'm pretty sure my PEP66 isn't the "best" heatsink, but it gets the job done. My 1.2 GHz Tbird runs rock solid all day long. Even in a rather warm room (26.6c) it chugs along just fine. Case temps report at a constant 39c. (Although, I have a hard time believing that.)

I'll tell you flat-out that temperature's understated. Understated enough to imperil your CPU? No. Understated enough for you to replace it and get better performance? Maybe, maybe not.

---Would a Millennium Glaciator make an already stable system more stable? No it wouldn't. Would I be able to squeeze out more MHz? Possibly. At the expense of more noise and $40.

You're looking upon this the right way.

---I've tried a lot of different coolers ranging from stock AMD to copper waterblocks. So why am I using a PEP66? It's a good heatsink that doesn't sound like a Harrier Jumpjet, and I just feel comfortable with a nice portable setup.

No problem with that.

Sometimes its important to take a step back and weigh the pros and cons of all these coolers. A little objectivity goes a long ways. If you've already got a Thermoengine with a Delta Black Label fan. Do you honestly think you're going to get a heck of a lot better performance out of a conventional heatsink?

You won't. A bit better? Yes. Enough to justify the $40 all the time? If you have a crappy CoolerMaster, yes. The kind of heatsink described above? Probably not.

---I don't care how well manufactured the thing is. There's a limit, and we're seeing it.

Yep.

If AMD doesn't do something to drastically reduce the amount of heat these puppies are putting out, we'll all be running the ridiculous hairdryer coolers in our cases.

Yep.

---Until then I say, "Get a grip."

We're not trying to say, "everybody has to go buy a Glaciator."

AKDUDE

06-18-01, 06:27 PM

This has been a interesting thread so far. Its hard to understand why you guys don't beleive that onboard thermosistors and MBM are wrong. I don't have much experiance with the AMD cpus, as I just started buying them in Jan. However, with three different Via Boards, and 1 processor and heatsink, I got a 15c differance in temp between them. How accurate is that?
That little test alone makes me not believe anymore, and just take MBM with a grain of salt, and maybe as a earlier warning to fan failure.

I do have a question for Ed- How much differance do you see in temps taken with a small flat thermosister touching the die on top of the CPU, than one thats in the HSF itself?
I could see airflow getting under the HSF and disturbing the readings, but I can also think of ways to block that airflow easily.
AKDUDE

stroligo

06-18-01, 06:32 PM

Badger (Jun 18, 2001 05:25 p.m.):
stoligo,
What makes you think drilling a hole in the HS and sticking a thermister in is going to give you more accurate readings than the sensor in the socket?

Because AMD says so to get an accurate temperature reading. See the link above.

---A thermister mounted in a hole in the HS will be effected by the HS temperature as much as by the core temp. Obviously the HS temp will always be below core temp (otherwise the heat would not flow from core to HS) so any temps measured by this method is likely to give low readings.

AMD says to measure it that way. It's obviously not perfect, they're putting an on-die diode to measure temps in Palominos, but that is what they say to do.

---Maybe your MBM temps are more accurate than you thought !.

If it's not accurate enough for AMD, it's not accurate enough for us.

---Anyway when it comes to HS comparisons the accuracy of measurement is not so much the issue as the repeatability. I would consider the repeatsbility of the standard socket temp sensor better than the drilled HS method due to possible variences in the way the thermister is mounted in the various heat sinks.

AMD gives very precise instructions on where the hole is to be placed and the diode put. Look at the AMD techdoc with the link in a previous post.

Yomama

06-18-01, 06:38 PM

The emperor appears to be quite adamant today :)

stroligo (Jun 18, 2001 05:29 p.m.):
Ed - I ususally concur with your findings, but this one is hard to understand. If I read the article correctly, you are impying that temparture misreadings may be caused by over-heating or "over-cooling" (by excess air from the heatsink) of the caps on the side of the CPU. To that extent it should be able to reverse this effect - at least temporarily by isolating the caps from the warm or cold air, e.g. by means of a piece of cardboard or plastic.

No, it's not just a matter of hot air. Heat gets conducted a bunch of ways throught the secondary heat paths. What you suggest might reduce the problem, wouldn't eliminate it.

Y: Probably - but may be worth a try to prove/disprove your theory

---I still think that what we are seeing is an inherent design issue of all copper heatsinks; they can absorb a lot of heat, but unlike aluminum they retain the heat, ultimately requiring more airflow to dissipate it. I am kind of busy right now, but I bet with a strong fan blowing from the front onto the cpu, one would be able to relieve the heatsink, and get lower temps overall.

Will the Glaciator do a bit better job with a 7K fan? Sure it will. I know of somebody who did just that. The MBM number still wasn't as low as the other, somewhat inferior heatsink.

Y: Ultimately it's not the MBM temp, but the bios temp that we are talking about. I agree that this temp can be influenced by environment conditions, but in general it should not drift by more than a couple of degrees, after all the internal temp rose by only 4C, and the heatsink is not that hot either, so where should this warm-up come from? I think that the fan, though spinning at 5600rpm, may not be efficient enough. If I find the time I'll try a delta screamer or comparable.

You just won't believe MBM is inaccurate, will you? What do we have to do that would prove it to you?

I suspect the answer from many is, "Nothing." Well, we know that's wrong. If you insist that MBM is an accurate measurement, then when you look at a heatsink review, don't look at our other numbers. Just look at the MBM number, since that's the only one that counts in your book.

Y: Anything is possible, and i am open to learn. I am just a little surprised that you are making what appears to me a strong attempt to justify the use of this heatsink.

Do you know what? If you had, you would have seen that the Glaciator doesn't do so well in MBM numbers. Our test shows that.

---For now I have to believe that despite the higher readings, the system is more stable, as I was able to reduce the voltage by 0.05 -0.1V.

Y: This is true, but I also reapplied AS and it may just be coincedence. Bottomline here - it ran at 966 stable b4. Unless I can o/c higher this improvement is irrelevant.

If MBM were so accurate, how could this be?

Y: because heat may not be the issue. I can run this chip at 50C, and it still is stable, but whichever part of the CPU gives out first does not benefit from the heatsink. If I could get it to below 40C (like in Winter with an open window :)) I am able to reach 980 or more.

stroligo

06-18-01, 06:46 PM

taz1004 (Jun 18, 2001 05:54 p.m.):
I love it when someone disects my analogies yet leave my conclusion out. Let me make it simple. Glaciator is reporting higher temp than others. Designer claims that sensor is inaccurate and it's actually running cooler.

---We've tested that ourselves, and that claim is accurate. I've seen that test run before my eyes, and that claim is accurate.

---WHAT TOOL DO USERS HAVE TO VERIFY THIS AND HOW ARE THEY SUPPOSED TO MONITOR IT? Oh... I see... guess?

According to AMD, to get an accurate thermal reading, start drilling. We didn't say that, AMD did.

---Many other heatsinks even with similar design are reporting cooler temp thru MBM. Is it likely that these heatsinks are reporting higher temp than actual as well?

They usually, but not always, do, and there is reason for that. We've been explaining and documenting that for at least the last eight months. Please read this (http://www.overclockers.com/tips240) and this (http://www.overclockers.com/articles306) for some early examples of this, including those done by other sites.

---Or is it just the Glaciator that's reporting higher temp yet all others are reporting lower temp? No other heatsink has been accused of misleading temp before this point

Read the articles mentioned above, we and others have been pointing this out for a long time.

---and like Richard said, no other heatsink has given this much benefit of the doubt that it might actually be performing better. If it did, please point it out to me cause I might've missed it.

The issue is the Glaciator yields better CPU but worse MBM results than the others. We use the more accurate measurement prescribed by AMD, and that is the measurement we go by.

"The tires work fine with every other brand except Ford Explorer. Hmmmm?"

---Exactly my point. Ford should redesign Explorer so it works with Firestone right?

As I pointed out a couple days ago, there is obviously an interaction between an Explorer and a Firestone that is causing the problem, and both parties should stop blaming each other for 100% of the problem and try to figure out what that interaction is.

---Explorer has been designed long before Firestone and it is designed with standard tires.

Firestone made tires before the Model T. The tires work fine with other SUVs. They don't work so fine with Ford Explorers. On the other hand, Ford Explorers work fine with other tires. Again, it's not just one party's fault.

--BrianC

Phil

06-18-01, 06:52 PM

Badger (Jun 18, 2001 04:17 p.m.):
I think it's quite clear from the various reports that the Glaciator just don't perform quite as well as some of the other designs out there. The evidence is quite overwhelming and quite frankly the argument that somehow the heatsink design can give a significantly higher MBM reading (consistently in comparison to a veriety of heatsinks over a wide range of mobo's) while actually cooling the chip better is pretty unbelievable. Further evidence is the fact that even on die temp monitoring is showing higher temps, Phil's theory that the analogue signal from the thermal diode can somehow be corrupted (and this corruption somehow makes the monitor read higher only when a Glaciator is installed !) is even harder to swallow.
The only factor that explains all the high MBM readings being reported is that the Glaciator just doesn’t cool as well as some of the other heatsinks. I can't grasp why people have not come to this rather obvious conclusion.

I'm not Glaciator bashing here, it looks like a well designed cooler, but when it comes to no compromise ultimate cooling it's clear it can't compete with the far noisier high air flow solutions offered by other manufacturers.

You've misinterperated the way I meant corrupted. The signal is read as resistance, and cooling the tracks that this signal travels along will lower the resistance and heating them up with air from a heatsink fan will increase them. If you read the back door cooling article it seems like he has a very good way of lowering temps, but in reality he's just manipulating the thermisters readout

Daniel ~

06-18-01, 07:21 PM

taz1004 (Jun 18, 2001 02:56 p.m.):
Spewn, I'm not trying to pick on your post but I knew someone was gonna say what you've said and I can't resist. :)

"It's not that heatsinks don't WORK with the motherboard's, it's that they cause the system to report innacurate temperatures"

I disagree. Temperature monitoring is a feature (important one at that) and if the heatsink and certain motherboard reports inaccurate temp, it is NOT working properly. Like I've said, heatsink and motherboard manufacturers can point fingers all they want but the fact is that they're not working together properly. And since heatsinks are designed around motherboards and not the other way around, if the heatsink is reporting misleading temp readings, it is fault of the heatsink designer.

"Reporting an accurate temperature has nothing to do with how well the sink is actually doing it's job."

But if there's no easy means of determining weather the sink is actually doing its job... well... then how the heck do you know if it's doing its job? My point was that the MBM with the chipset monitoring feature is the most accesible means of measuring temp right now and the heatsink designer is saying ignore that. If such inaccuracies exist in that big margin, how are we supposed to know if my CPU is burning up or not? And in case of Glaciator, I overclocked my CPU and it is now reporting 70'C. Should I go higher since MBM is incorrect? If so, how high should I go?

"It's up to motherboard/cpu manufacturers to provide us with a means of getting an accurate temp reading, regardless of what heatsink we're using."

Can't argue with that. But do such means of getting perfectly accurate temp reading exist? From that article, it sounds like Intel's way of measuring temp within the die itself isn't accurate enough. But are they close enough? I think so. Should heatsink manufacturers design based on this imperfection? I believe so.

--BrianC

So you would rather have a heat sink that cools less well but reports with greater accuracy? The methodologies are written by Intel and AMD, Joe and Ed merely follow their protocol. The chip is what the Heatsink is design to work with, not the Motherboard. If the Motherboard is flawed in it's reporting, then a HS.designer can use that flaw to create more favorable reporting, while actually cooling less well.

What Joe and Ed are saying is that such reporting cannot be relied upon, not if you want to know what's really going on. Hey you know and I know that a lot of you guys know this stuff better than I do.

But as I see it Joe and Ed are saying if you want to know weather, and to what degree, you are with in safe temp range for a given chip, then you have to use the same procedure as the Chips manufacturer, or there is a lot of unaccountable variables that can and will mislead you.

seadog

06-18-01, 07:31 PM

I spent part of the weekend playing with new heatsinks and here's what I came up with:

Taisol 760092 w/Delta38 ....... idle - 38C ........ P95 load - 43C ....... max stable P95 - 1450

Glaciator ............................. idle - 40C ........ P95 load - 45C ....... max stable P95 - <1430 (didn't test any lower)

Thermalright SK-6 w/ Delta30 ....... idle - 37C ....... P95 load - 41C ....... max stable P95 - 1450

Interpret the results any way you like, these are simply the numbers I came up with off of my system.

Case - AntecSX830, 2 47cfm Panaflos intake, 2 34cfm fans out the back, Enermax psu w/dual fans. Case temps constant 23-24C. All readings from MBM.

Forget for a minute what MBM says about the temps. If you agree that, ALL THINGS BEING EQUAL, a chip's overclocking stability is significantly impacted by the core temp then the numbers speak for themselves. If anyone has another explanation, I'm all ears.

All testing done at 23-24C case temps, 1.91 Vcore with nothing changed but the HS. The Glaciator was re-attached over half-a-dozen times with varying amounts of ASII and eventually lapped in my search for the expected numbers. (Prior to lapping, the MBM numbers were 2C higher than what I have posted.)

Now, I don't really care which is the best HSF at this point. I already own all three. So flame away if you must ;D . I know others have achieved different results - every system is different. But thats my story and I'm stickin to it. :)

Hoot

06-18-01, 08:24 PM

Thanks Seadog for the nice array of HSFs, speeds and temps example. It must be tough deciding which one to go with, since they're all very nice units. you've done your homework well, including case ventilation. One thing though, which setup, to your ears, was the most tolerable in terms of noise?

Hoot

stroligo

06-18-01, 08:53 PM

I spent part of the weekend playing with new heatsinks and here's what I came up with:

Taisol 760092 w/Delta38 ....... idle - 38C ........ P95 load - 43C ....... max stable P95 - 1450

Glaciator ............................. idle - 40C ........ P95 load - 45C ....... max stable P95 - <1430 (didn't test any lower)

Thermalright SK-6 w/ Delta30 ....... idle - 37C ....... P95 load - 41C ....... max stable P95 - 1450

Interpret the results any way you like, these are simply the numbers I came up with off of my system.

Case - AntecSX830, 2 47cfm Panaflos intake, 2 34cfm fans out the back, Enermax psu w/dual fans. Case temps constant 23-24C. All readings from MBM.

Forget for a minute what MBM says about the temps. If you agree that, ALL THINGS BEING EQUAL, a chip's overclocking stability is significantly impacted by the core temp then the numbers speak for themselves.

--I don't agree. We're talking a couple C difference here, for a start, that's usually insignificant. That's one system, stick that in another case and another motherboard, you could get entirely different results, just like you would with a CPU.

Now what might be significant is if we saw a few dozen people do essentially what you did, and we consistently saw that level of difference.

seadog

06-18-01, 08:55 PM

One thing though, which setup, to your ears, was the most tolerable in terms of noise?

The Taisol was definitely the loudest with the signature "Delta whine". Between the SK-6 w/Delta 30 and the Glaciator it's basically a toss up. If I were to disconnect all the other fans, I think the Glaciator would come out on top in terms of noise. With all fans running I couldn't really hear any significant difference between the two

seadog

06-18-01, 09:09 PM

--I don't agree. We're talking a couple C difference here, for a start, that's usually insignificant. That's one system, stick that in another case and another motherboard, you could get entirely different results, just like you would with a CPU.

I'm afraid I'm a little confused by what you are saying here. I don't necessarily see the temp difference as significant, especially in light of the variations in MBM temp reporting that have been pointed out.

My question is that if it is not a heat issue, then what accounts for the loss of at least 20mhz in overclocking stability with the Glaciator?

Hoot

06-18-01, 09:56 PM

SeaDog, that's something like 1.3%. Though it undenyably runs slower, I have seen such small changes using identical components and just the difference from one dismounting/remounting of the HSF to the next, not to mention other subtle differences. Don't get me wrong. I am not discounting your results, nor what they imply. My in-socket thermistor is calibrated with several thermocouples and this whole issue has me chomping at the bit to get my Glaciator. I just hope it arrives before everyone but me has one, along with an opinion! ;D

Hoot

taz1004

06-18-01, 10:06 PM

"---Article is great and certainly something to consider but what other choice do users have other than MBM?
If you have a clock that's broken and cannot keep track of time accurately, and you can't afford another one, do you say, "It must be OK?" That's essentially what you're saying."

I was asking for other means of measurements. I asked a question. How do you derive a conclusion from a question?

"---but let's say it's true and Glaciators is the best heatsink.

We aren't saying it is."

Really? The Emperor Speaks?

"---Or should they all start drilling holes in their heatsink and place another thermistor in it?

If they need an accurate temperature reading, something along those lines would need to be done. However, most people don't need that level of precision; we put together that calculation so people could at least have an alternate means of checking their situation."

Are you absolutely positive that the calculation is accurate? How can it be? It depens on the system, thermal compounds, method used on applying compound, case fans. It seems even more unreliable than the MBM.

"How is it possible that there's world hunger? How is it possible that Windows crashes? How is it possible that Intel and Via chipsets have bugs in them. It is. Saying that there shouldn't be doesn't change the "is.""

You don't wanna hear my answer on these. In case you haven't noticed, it was just an expression.

"When something breaks in your house, for the next six months, do you say, "It shouldn't be broken." What does that do?"

Doesn't have nothing to do with what I said.

"This is like having a guy with very poor sight judging the Miss America contest, and you're saying it's the contestant's fault."

Beauty contest are never contestant's fault. If I'm the judge, yes, I make decision based on my vision and taste. It's matter of opinion and my opinion says this one doesn't look good. If my eye sights are bad, at least it's consistantly bad. I don't favor one contestant over the other like some do.

"Firestone made tires before the Model T. The tires work fine with other SUVs. They don't work so fine with Ford Explorers. On the other hand, Ford Explorers work fine with other tires. Again, it's not just one party's fault."

I didn't wanna spread into another subject but... Is the tire made before the Model T the tire in question? I don't think so. Tire in question is released well after Explorer. I'm not interested in determining who's at fault. I'm interested in solution. Is it cheaper for Explorer owners to get other tires or is it cheaper for them to keep the tires and get a new car? Does it make sense for them to replace tire at the place they bought them? or should Ford recall all Explorers and fix the problem and send the car back to owners in months?

Firestone designer didn't do their homework and excluded Explorer in their research. Every car companies have their own suspension system and Firestone is well aware of that fact. They will loose that market and it's their loss. I don't think there's soul out there that's gonna stay away from Explorer because it doesn't go with Firestone.

Do you even think car manufacturers have time and resource to deal with all these ridiculous brands and models of tires out there? They design a car for standard radial tires and that's it. If the tire manufact goes out of this spec, it's their problem. On the other hand, how many car makes do tire manufactures have to deal with? Only a handful. I hope this subject doesn't come back.

You asked for it. :)

--BrianC

seadog

06-18-01, 10:48 PM

SeaDog, that's something like 1.3%.

Hoot, I agree that the differences are extremely small. I don't think that anyone should regret the purchase of any of the 3 HSFs that I tested.

For me the loss of 20mhz is significant because, like all of us, I am trying to get the most out of my system. I did re-mount the hs numerous times and still was not able to run P95 stable at 1430. I don't have a definitive answer for this.

I have never represented these results as anything other than MY results on MY system. At the bottom of my initial post, I acknowledged that others have shown and may yet achieve different outcomes. I only posted the numbers because I believe that the more information we all have the better able we are to make informed buying decisions.

Hopefully, your temperature studies will shed some more light on the situation. :)

sfa ok

06-18-01, 10:54 PM

It seems quite clear to me that hsf's should be designed around a CPU, not a $0.10 mobo sensor. If you want more accurate readings without drilling, get a probe that sits next to the core (I'm not saying that this is super-accurate, but better than mobo). I've been frequenting this site for a while (a long time before I joined the forums), and I doubt that preferential treatment would be given to a hsf.

William

06-18-01, 11:44 PM

Taz, I don't see why you think reading your temps through the cpu ceramic is going to be so accurate and why you think sticking a thermocouple above the core inside some highly condictive piece of metal will be more inaccurate. I think the explination of the Glaciator MBM troubles is perfectly valid. Yes the thermocouple is the best we have, but why don't we trust the review sites? Especially the ones that do it to the directions of AMD. I would think AMD would know how to get the most accurate temps from their chips. Just take for example the Asus thermocouple which is typically as much as 10C off of the actual temperature. SFA is totally correct, why design a heatsink that gives better readings with MBM rather than a heatsink that gives poorer numbers in an inaccurate measurement system but produces better results. The flaws with the thermocouple are extremely well documented elsewhere and they have no mention of a specific HSF.

Spewn

06-19-01, 01:19 PM

Wow, I don't check for a day and this is what happens! Just back to the issue about it being a heatsinks job to allow the motherboard to report an accurate temperature. If heatsinks were designed to report temperature, and not just regulate it, they would come with thermocouples drilled into the bases, or at least a thermocouple and any wiring or whatever you'd need to attatch it to a thermometer. Do any? I don't know, but I highly doubt it. As far as temp readings go, yeah they can be important, but how important are they really? If you're cpu is hot enough that it may be damaging itself, wouldn't it be crashing before it started burning? That's the whole idea behind overclocking right? Keep the temp down to a point where the system runs stable, and you've overclocked successfully. Do you really need to know exactly what the temp is to do this? If your cpu is stable at a given speed, but you're worried that maybe MBM is innaccurate, and that your HS isn't *really* 60'C, try touching the base of the heatsink. If it's not REALLY hot, you're fine. That's the way I look at it anyway, I've had a K6-2 333 oc'd to 450 for...I dunno, about 3 or 4 years now. I have no idea what the temp is, but I know it runs stable, and the heatsink doesn't burn my hand, so everything's fine(not to mention the fact that the cpu hasn't died yet).

One thing that might pose a problem is changes in ambient temperature. Or at least, I'm sure someone will reply saying something along those lines. If you're so close to the edge of being stable that 1 or 2'C higher temps cause your system to crash, or you're worried it may cause your cpu to get burned over time, maybe you need be stepping back that oc in the first place? As far as I see it, accurate temps are for interest-sake only. Given that the whole basis behind all of this is "lower temps means higher oc'ing" if a heatsink allows a higher oc, it cools better. THAT is really all I'm interested in.

sequoia464

06-21-01, 07:01 PM

...the Thermosonic Thermoengine has just the opposite effect happening. When tested using the socket thermistor it seems to be a great sink, 2cooltec tested this similarly to the methods used here at Overclockers and the sink FAILED their tests. Not didn't perform well but failed. The reason they give is the path of the airflow, I'm sure quite a few of you have read about this.

Don't get me wrong, I'm bummed that I can't trust the temps that I see from the boards system. It doesn't seem like rocket science either to drill a hole and plug a sensor in the Glaciator either, might have to be my next project.

Stroligo, if you happen to read this post, perhaps you could point me to some info on this.

coldmop

06-21-01, 08:35 PM

The legendary ducted Superorb effect, swirling air being forced under the socket giving false readings. But the fop38 with its wicked fan and open sides was never suspect of the same effect. Why? Just once I'd like to see some tests with a paper shield clipped under the heatsink and covering the socket and surrounding components. Is that a real world test? No but it helps isolate the role of the heatsink. In truth you want air cascading off the HSF onto surrounding components especially the clock gen chip (those things are hot). Another funny thing, you get a nice immediate temp drop when you direct a fan at a waterblock...must be the fine fin arrangement. Didn't Sherlock Holmes mention something about discovery of the improbable.

Slake

06-21-01, 10:31 PM

I'm SO confused! Having read a lot of Ed's articles, as well as all of the intelligent posts in this thread, I concede that almost all of you are far more knowledgable than myself. The one thing I am very confident in saying is I will continue to rely on Ed's reviews and opinions to be more accurate than novices. I see the point and logic of those who say that on core measurement is bound to be more accurate than readings from any point not on the chip. Also that measurements taken this way even though not accurate are repeatable and should be a fair reference point to judge heat-sinks. Ed does this for a living, and as such is acutely aware of the myriad of influences that combine to produce various measurements. Most of his in depth "scientific" theories are a bit over my head. That's why I bacame a member here, to benefit from others knowledge. Ed's inegrity as well as that of Overclockers.com, as far as I am concerned is beyond reproach. Any thoughts that he may be swayed by a manufacturer are totally BOGUS. It's great to have access to all of you guys opinions but I believe Ed is correct in his assertions based on his years of expertise. If Ed is proven wrong I'm sure he will publicly acknoledge it. None of us is above human error.
I'm happy to have a ThermalRight SK6 on the way.
Accusations of me being a yes man or a Ed's butt-boy are graciously accepted. ;D

William

07-26-01, 02:48 PM

bump

M@€$†®Ö™

07-27-01, 03:55 AM

What is the point of picking this subject apart the gains are so small. You would never notice it. Or so I feel and I have messed with as many computers as most anyone here. Oh well that is my 2 cents in the barrel :)

Maestro

CalCoolage

07-28-01, 02:36 AM

sequoia464 (Jun 21, 2001 07:01 p.m.):
...the Thermosonic Thermoengine ... 2cooltec tested this similarly to the methods used here at Overclockers and the sink FAILED their tests. Not didn't perform well but failed.
.

It FAILED did it? The moral of this should be don't make the mistake of thinking a reviewer is competent just because he used a thermistor on the top of the CPU..

No the Thermoengine didn't fail. Mr. 2Cool failed. The fool didn't get it to contact his synthetic heat source firmly. He should have realized that and said so. Instead he maligned an excellent heat sink, and mislead the people who read his article. The thermoengine sink, and some others like the Thermaltake orbs, have a clip that does not press on the sink at a single point centered over the CPU slug. This makes the sink more prone to making uneven contact, pressing on one side of the slug more than the other, or maybe not even making contact except on one edge. Not making good contact is the ONLY reason ANY heat sink will "fail". Mr. 2Cool's makeshift clip mounts are not positioned in the same relation to the CPU slug as on a real socket. The clip works OK in real life even though it doesn't on the botched test jig.

I don't know if Mr. 2Cool ever admitted his stupidity or issued an apology. I doubt it.

There are difficulties with using a synthetic heat source the size of which is much larger than the CPU slug, which Mr. 2Cool did not deal with well, and therefore his test results should not be taken seriously.

Tests have been done on the Thermoengine with a top mounted thermistor. Although they do not show the Thermoengine to be as exceptional, it retains its excellent rank among the aluminum sinks that were its contemporaries, essentially the same ranking as it had in tests done with in-socket thermistors.

That stuff about "blowing directly on the thermistor" is a load of horse manure. The most likely reason for the extra coolness at the socket thermistor is that the ceramic chip carrier is getting cooled more.

Joe Citrella has said that getting the back of the chip cool does have some effect on CPU stability, so it is not true that the top mounted thermistor tells the whole story.

Nothing on the mobo represents an infinite resistance to heat. Therefore the path of heat flow is complex and can't be very well modeled using lumped resistances (as in electrical circuits) . Everything is thermally connected to everything near it.

It is not often mentioned, but the silicon substrate of a chip is quite a good condutor of heat, about half as good as aluminum if I recall correctly. Although the substrate is too thin to conduct much along the surface of the chip, its cross-section in the perpendicular direction is large, so it can conduct heat very well toward the back. Some ceramics are pretty good conductors of heat. Although I don't know if AMD is trying to get heat out through the back, I don't see why not. It people will recall, up to a certain point, Intel chips were mostly encased in ceramic and the heatsink only contacted the ceramic. Until the flip-chip (as in FCPGA) Intel mounted the bottom, or substrate toward the top of the chip, so heat really came out through the substrate, the bottom. Considering all this, it does seem as though heat conduction could be doubled if chips were designed so heat was conducted out both the top and bottom.

The heat dissapated by a chip is generated almost entirely when the transistors switch state. That's why heat dissapation is proportional to MegaHertz. The switching time is a small part of a whole cycle. Therefore the actual maximum temperature of the transistors must be much higher than the average temperature. An average is the only possible temperature a probe could possibly measure. Likewise transistors only account for a portion of the whole chip area, the rest of which is wiring and spacing. .Any probe measures an average over a large area. Again, the transistors maximum temperature must be much higher.