I must confess; I am not a cooling aficionado.
This is a skill I have been able to afford to neglect for more than a year simply because somebody else on this website is: Joe Citarella.
My concept of cooling can be expressed in one sentence, “Colder is Better.” If Joe says, “this will make it cold,” I say “Good,” he makes it so, and I’m free to worry about other things.
Until a couple weeks ago, things seemed cold enough, checked every once in awhile, it looked OK to me, so be it.
Probably not much different than most of you, right?
Then a few things changed in my life. I started tossing in higher-speed processors and new motherboards. Suddenly, I found out things weren’t so cool anymore.
I didn’t know if I was a website writer monitoring my computer temperature, or a short-order cook checking the temp of the grill.
The temperatures I was getting from the water-cooled system scared the s— out of me. I went to an Alpha as a temporary expedienct, and it only scared the hell out of me. Went to the Glaciator, and my finger was telling me one thing, and the thermometer was telling me something else.
Even worse, the temperatures were all over the place; I’d run one item one day, run the same thing the next day under the same conditions, and get completely different results.
Sounds like a job for Dr. Joe, but he was making a house call on the other end of the planet.
So I waited, and made do, and read a whole lot about heatsink and measurements, and things like that.
Dr. Joe showed up with his measuring satchel, poked the water system around and pronounced it dead, and then we started to take readings:
Here’s the machine:
AMD 1333Mhz AYHJA, running at 1500Mhz (10X150) at 1.85V
MSI K7Master, no cooling modifications outside of additional silicone thermal grease on Northbridge heatsink
512Mb Crucial PC2100
IBM 45Gb 75GXP hard drive
Matrox G450 Dualhead video card
Enermax EG451P-VE 431-Watt power supply
InWin Q500A full tower case, case open on both sides
Win2K, SP2
Board temperature upon initial warmup 25C (for later tests, probably 27-30C)
Test programs used: Seti@Home and/or CPU Stability Test
AMD761 cooling tweak (using HLT and Stop Grant on an ACPI-disabled machine) enabled.
We basically ran three tests:
Test One: Just How Accurate Is That Thermistor?
The K7Master uses an thermistor which measures temperatures at the center of the back of the CPU. The thermistor touches the CPU.
The sensoring device is one of the Via sensor, and both the MSI program (PC Alert) and Motherboard Monitor use the same sensors. They gave essentially identical results.
I asked Joe to hook up a thermistor right next to the MSI’s. The closest the two temperatures ever got was 5C. The furthest away it ever got was 12C. 6-7C was the norm, and the MSI thermistor was always too high when we were gunning it.
The highest temperature we recorded with the thermometer hooked to the back side of the CPU was 51C. The MSI thermistor topped off at around 57C.
We then let the CPU cool down. Temperature readings from the MSI thermistor have gotten as low as 23C, 25C is typical.
Using the thermometer, CPU temperature after extended idle did not dip below 30C.
So the MSI device isn’t very accurate from either end, and not even consistently inaccurate.
Test Two: How Bad Is That Alpha, Really?
I asked to test the Alpha under the same circumstances. Under earlier circumstances, sometimes, it looked like the Alpha was doing just as well as the Glaciator. Sometimes it didn’t.
Well, this time around, we couldn’t even get the machine to boot through to Win2K at 1500/1.85V (it was warmer and more humid than it was during earlier tests). We watched temperatures during boot, and there was no comparison between the Alpha and the Glaciator during boot; the Alpha was consistently about 10C higher.
We scaled back to 1400Mhz/1.75V, and we were seeing this time around very strange readings, again, much higher than for the Glaciator. MSI thermistor temperatures were hitting 66C this go-round, again, significantly higher than the thermometer reading. Even more disturbing for those who would like to know just what was going on, the temperature difference between thermistor and thermometer didn’t follow the pattern set by the Glaciator.
Test Three: A CPU Core Reading
Finally, Joe put in another Glaciator, the one he had used for previous testing, with a hole bored through the copper core and a thermocouple glued above the CPU.
Running at 1500/1.85V, running both SETI@Home and CPU Stability Test simultaneously, we got a temperature of 47C, with a board temp of about 27C. The MSI thermistor reported 10C higher than that.
Isn’t that a lot higher than what was reported in the article?
The reading we got would translate to a C/W of 0.20. Depending on how it was tested, the C/W in the review article ranged from 0.13 to 0.18. Joe considers the C/W from the heater (which in this case is 0.18) to be the “best” number for comparative purposes.
Our reading is not directly comparable to the review reading for a few reasons:
So while the temperatures we got might be a bit more realistic for the average person, that doesn’t mean they’re more accurate than Joe’s measurements.
How To Read Joe’s Numbers
The purpose of Joe’s tests is to establish a fairly precise pecking order among heatsinks.
Fairly precise, not absolutely exact. It like an IQ test. If one person has an IQ of 112, and the other has one of 111, the guy with the 112 isn’t smarter. You need about a ten point difference between IQ scores for it to mean anything.
For this Heat Quotient (HQ) test, the significant range is .02 between exactly the same measurements.
If you look at Joe’s review, pick the lowest number, and say, “Wow, I’m going to get 34C running at 1333Mhz,” that’s not how you should be doing it.
If you look at a couple of Joe’s reviews, and conclude, “Wow, the Millenium blows the Swiftech away,” again, you have not read it right.
If you look at a couple of Joe’s reviews, look at the C/W numbers and conclude, “The Millenium is better than the Taisol,” now you’re starting to get the idea.
What Is Not Important When You Don’t Get The Same Temperatures Joe Does
The primary reason your motherboard is likely to be actually hotter than Joe’s is simply because it’s in a closed case.
Unless you have so many fans in there that the National Weather Service is tracking the storm in your case and given it a name, it’s probably like 30-35C, maybe even higher.
Whatever that temperature might be, that’s your starting-off point. If it’s 35C, and Joe’s got 22C, the temperature you get is going to be at least 13C higher than Joe’s. Any heatsink you buy is going to be at least 13C higher just for that reason.
What Is Important When You Don’t Get The Same Temperatures As Joe
The purpose of the whole exercise is to keep your CPU alive and well. The CPU doesn’t care why it’s at a certain temperature; like you, it just cares if it is getting too hot. Like you, it quits when it gets too hot.
Like you, if it gets too hot too long too many times, it dies.
How hot is too hot?
AMD has guidelines for this. You can find them here. AMD measures the temperature a certain way. That is exactly how Joe (and few others) measure the temperature.
When you convert the formulas into English, AMD basically says 70C is the maximum sustainable temperature when the CPU is going full-throttle.
Now that’s 70C measured the way AMD (and Joe) measures it. That’s not 70C the way Motherboard Monitor or just about any other normal means measures it.
Unfortunately, trying to figure out how hot things really are is much like looking at a old rusty CocaCola thermometer at an older country store. That thermometer is older than you and possibly older than your father. And it’s usually in the shade.
You essentially have a five-cent thermistor that you’re depending upon to determine life or death for your CPU.
This is not too good when you are generating heat that per square inch is approaching that found in nuclear reactors and Space Shuttle reentry (Factoid courtesy of Andy Lemont.)
How good would you feel if you lived near a nuclear reactor or were on the Space Shuttle, and you found out the NRC or NASA chose the shielding just because it looked cool? Or just because it was pretty cheap?
Competent people solve problems, smart people prevent them
The best way to prevent heat problems is to stop them from occurring in the first place, and all that takes is a little time, a little math, and maybe a little money.
All you need to do is figure out how much heat you have to get rid of, find the coolers that will get rid of enough heat to keep you within the AMD guidelines, and give yourself enough of a cushion to account for variations.
Tomorrow, we’re going to give you an easy way to figure this out, so you can feel pretty safe about your system, so long as you put the heatsink on right, and the fan works.
I Get Good Temperature Readings, I’m OK
That’s precisely the problem. You think you’re OK. Whether you actually are or not is an entirely different story.
Unfortunately, the average means of taking temperatures not only aren’t accurate, they’re not even consistently inaccurate.
There’s are a ton of reasons why the temperature you’ll get is not likely to be correct: it’s not measuring the right spot, the thermistor can be inaccurate, the measuring program isn’t calibrated right, to name just a few.
We’re not being anal-retentive about this. We don’t think it’s an emergency if MBM reads 53C when the real temperature is 49C, or vice versa.
What we’re concerned about are those cases where an inaccurate temperature reading makes you think you’re safe when you aren’t.
I go to various forums and I see people with the same motherboard under more or less the same conditions reported temperatures that are 20C different from each other.
Now the high temperature guy is already worried. It’s the low temperature guy who’s in trouble. I’ve seen temperature readings sometimes that simply cannot be for what he’s using. Either he’s a lying dog, his idea of a full load isn’t, or
his temperature readings are way out of whack.
A Second Opinion
Just relying on MBM or that five cent thermistor all alone is not good enough.
We’re not saying you have to buy a thermometer and start drilling holes or weaving thermistors into strange places. We know most of you won’t do that. Nor do you have to.
What you absolutely should do, though, is do a minute’s worth of math to figure out how hot your CPU is supposed to get when it’s going 100% at your speed and voltage. Get at least a notion of what the worst real temperature can be. If that turns out OK,
then you can go back to not worrying with a little more certainty. If it doesn’t turn out OK, then you’ll know and can do something about it.
Step One: How Much Heat Does Your CPU Put Out?
Here’s a chart that will give you an approximation of how many watts your CPU will put out at 100%:
Speed | 1.75V | 1.80V | 1.85V | 1.90V | 1.95V | 2.00V | 2.05V | 2.10V | 2.15V | 2.20V |
1000 | 54 | 57 | 60 | 64 | 67 | XX | XX | XX | XX | XX |
1100 | 60 | 64 | 67 | 71 | 75 | XX | XX | XX | XX | XX |
1133 | 63 | 67 | 71 | 74 | 78 | XX | XX | XX | XX | XX |
1200 | 66 | 70 | 73 | 78 | 82 | XX | XX | XX | XX | XX |
1300 | 68 | 72 | 76 | 80 | 84 | XX | XX | XX | XX | XX |
1333 | 70 | 74 | 78 | 83 | 87 | 92 | 96 | XX | XX | XX |
1400 | 72 | 76 | 80 | 85 | 89 | 94 | 99 | XX | XX | XX |
1466 | 75 | 80 | 84 | 89 | 94 | 98 | 103 | 109 | 114 | 119 |
1500 | 77 | 82 | 86 | 91 | 96 | 101 | 106 | 111 | 116 | 122 |
1550 | 80 | 84 | 89 | 94 | 99 | 104 | 109 | 115 | 120 | 126 |
1600 | 82 | 87 | 92 | 97 | 102 | 107 | 113 | 118 | 124 | 130 |
1650 | 85 | 90 | 95 | 100 | 105 | 111 | 116 | 122 | 128 | 134 |
1700 | 87 | 93 | 98 | 103 | 109 | 114 | 120 | 126 | 131 | 138 |
Step Two: How Well Does Your Heatsink Handle It?
The next step is to find the C/W of the processor. C/W is a measurement on how well your cooler gets rid of heat. The lower the number, the better. Go here to find out what that is, along with a brief explanation on how C/W works.
Step Three: How Hot Is It Inside Your Case?
If you have some sort of temperature reading from MBM or better, certainly use that.
If for whatever reason, you literally have no idea what your case/board temperature is, consider the following:
If your computer is wide open, figure it’s about 3-5C higher than room temperature.
If your computer has the cover on, and you put in two good fans to suck air in from the bottom and suck it out at the top, figure it’s 10C hotter than room temperature.
AMD presumes a case temperature of 42C, which would be likely for an OEM box with poor air circulation in a temperate environment.
Step Four: The Math.
Take the wattage from the chart and multiply it by the C/W. Take that number and add the case temperature. This will give you an approximation of how hot your CPU will get at full throttle (in C).
If that number is less than 60, you should be OK. If it is over 60, you probably need to do something about it.
Why 60?
AMD’s spec indicate an effective maxiumum sustained temperature of 70C. (The 90 and 95C numbers are peak heat levels; it can survive that for a little bit, it can’t live like that).
From our experience, we really doubt an AMD processor could stably handle 70C all the time.
There’s also a number of factors we have not and cannot account for which could make the true temperature higher than our rough calculation.
Coolers react differently with different motherboards. Temperatures behind the CPU tend to be hotter than those measured at the core,
and there’s some evidence that that affects stability, too.
We suspect the AMD wattage statistics could be off a bit (after all there’s been several revisions of the chip), C/W when you’re not user a heater to measure it, but a mobo, can be off a few C, since you don’t get 100% CPU utilization, but a bit less than that.
So we gave ourselves a 10C cushion to account for these. You may want to give yourself more than that; we don’t think you should give yourself less. Keep in mind that a few degrees colder than expected will never be bad for the processor and may help.
We’re not saying your system will work wonderfully at 60C, in at least some cases, we’d doubt it. If you come in just below 60C, and your system is stable, that may be good reason to leave well enough alone. If for whatever reason, you’re looking for a new or better cooler, you shouldn’t be aiming for 59C, you should be aiming for the lowest temperature possible.
As you’ll see, if you crank it up pretty good, even the best cooler is going to still going to leave your CPU a bit toasty.
Math In Action
Example 1:
I have a processor running at 1500Mhz at 1.85V. I am using a Millenium Glaciator. I have an open case, I know my case temperature right now is about 28C.
I look in the chart, and find my maximum wattage to be 86 watts.
So it’s: 86 watts X .18 C/W = 15.5C + 28C = 43.5C. That a good deal less than 60, so I’m OK.
Example 2:
I have a processor running at 1500Mhz at 1.85V. I am using a Thermaltake Super Orb. I have a closed case, and my case temperature is 34C.
Let’s assume everything stays the same, except I close up my case (getting a case temperature of 34C), and I’m using a Thermaltake Super Orb.
86 watts X .35 C/W = 30C + 34C = 64C. That would not be OK.
Example 3:
I have a processor running at 1500Mhz at 1.85V. I live in a hot place and the computer is not in an airconditioned room. The room temperature gets up to 30C, and my case temperature is 40C. Assume I’m using the Swiftech MC-462A with the Delta fan.
The math is: 86 watts X .18 C/W = 15.5C = 40C = 55.5C. Even with a top-of-the-end air cooler, I’m getting awfully close to that 60.
Example 4:
I am king of the hill! I have a processor running at 1650Mhz at 2.20V. My machine keeps crashing, and I don’t know why. I’m using an Alpha PAL6035, and I have nine fans in my box in a room with the A/C going full-blast. My case temperature is 25C. MBM says I have a CPU temperature of 50C, that’s good enough, isn’t it?
134 watts X .37 C/W = 50C + 25C = 75C. That’s why you’re crashing.
My numbers look OK
Good! You may go. 🙂 Hopefully, this little exercise has at least reassured you a bit.
This can’t be right! My temperature are way lower than that under load!
If by “much lower” you mean by more than 5C, there’s only two possibilities:
Download and run this program. When you run it, choose the “CPU Warming Only”
option. Let it run for fifteen minutes. If the temperature you get is a lot closer to the math number, then your load program wasn’t heating up the CPU that much.
If it still registers a low temperature, your temperature measurement is bad. Use the math instead to gauge where you stand.
My temperatures are higher than the number I came up with
If they’re a good deal higher, make sure you put the heatsink on right. If you used el cheapo Radio Shack silicon, you may want to upgrade to a silver thermal grease.
If the temperature is being measured by a thermistor touching the back of the CPU; you could expect an actual temperature of say, 5C above the CPU core temperature. If you’re near that 60C level by the math, and you’re getting a reading 10C or more above what the math indicates, send me a note. Some heatsinks effectively
channel more heat to the back of the CPU than others; we’re not exactly sure why, but Joe’s measured that.
OK, OK, I believe you. My temp is too high, now what do I do?
There are three elements to your problem. We’ll talk about two of them now.
Either:
The case/board temperature is very high
If the temperature is high because it’s ninety degree in the shade, even the best cooler can do just so much. Try to get it to a somewhat cooler spot. An airconditioned room is a “DOH,” but moving it to any
cooler spot can help. We had someone once literally dig a hole in the ground; if that’s too much, a submerged or even semisubmerged basement may help.
If it’s practical, taking the case off can help.
If the problem isn’t the outside temperature, figure out why the inside temperature is so high. If the cables in your machine look like Medusa launching a revival, they’re probably blocking any cooling air flow. Tidy them up.
The tidiest arrangement in the world won’t help, though, if you got fans (or fan) that wouldn’t cool a mosquito. At the very least, have a good powerful fan sucking cool air in from the bottom of the case, and one sucking hot air out of the top.
The heatsink isn’t very good
Let’s say you like to play basketball, and your friend says, “Let’s play basketball tonight.”
You say “OK” but instead of the playground, he takes you to the NBA finals, and the next thing you know the only thing between Shaq and a slamdunk is you. How do you think you’d do? Would a fan yelling “You’re the greatest” make you any less roadkill?
An overclocked, overvolted 1.5Ghz neighborhood TBird is the Shaq of CPUs. Your Alpha or Orb were good or maybe even great in the PIII playground, might have hung on with a Duron.
But this is the NBA (Nasty Blastfurnace Association), and they can’t play in this league. They were never meant to, and you cheering them on isn’t going to change the laws of physics one bit.
Look at the math. If the numbers are bad, they aren’t lying.
If your numbers are bad, but you just can’t bring yourself to pay more than $10 or $15 for a heatsink, think of your CPU as Oliver Twist saying, “Please, sir, I want some more” and you saying no.
Think of your CPU as being like Tiny Tim. If the numbers are bad, and you don’t take care of him, he is likely to sicken and die. Even Scrooge ponied up after realizing that. Unlike Scrooge, though, you’ll have to buy Tiny Tim’s replacement should he pass.
We’re not talking big bucks here. So maybe you made a $15 or $20 mistake with the last cooler you bought. God won’t send you to Hell for that, maybe not even your wife. 🙂
Don’t endanger your CPU and system just for that.
If you truly have a financial rather than a mental block with paying $40 for a new cooler, you might want to consider a better fan for it. A high-speed fan can help a mediocre heatsink quite a bit.
However, what you save in cash from depending on a 7000-rpm fan for cooling you’ll pay for in noise. Imagine a perpetual mini-vacuum cleaner, and you’ll get the idea.
When you choose a somewhat cheaper heatsink that seems to do almost as well as the big boys; you’re almost certainly going to get that mini-vacuum cleaner along with it.
The big boys (which, right this moment, I’d call the Millenium and Swiftech, in no particular order) are the big boys because they substitute good design and mass for fan speed. They don’t need 7000-rpm fans to work very well. Not going to call them whisper-quiet, they’re not, but that’s
the best you’re going to do right now if you need big-league cooling. You want whisper-quiet, water.
At the beginning of this section, I said there were three elements to the problem. Here’s the third.
Pump Down The Volume
There’s one sure-fire way to get rid of excess heat. Don’t make it.
You can lower the voltage or MHz, or you can do less strenuous things with your computer.
That’s becoming harder to do, though. Even Web browsing can be occasionally strenuous. Some of those particularly annoying ads can heat your CPU up better than Prime95; I had one this morning.
Slowing things down is probably a good temporary expedient when your computer is in a really hot environment and you can’t take countermeasures. Better a few less fps than taking the risk of zero fps.
If that’s not the case, you’re kind of out of options. Unless you believe the Cooling Fairy is going to come along or something.
Why We Wrote This
We wrote the piece because we don’t think a lot of people know or understand what they’re getting themselves into running at high speeds, and now it’s not a matter of overclocking a bit less. It’s a matter of possible CPU failure.
There’s never been CPUs like these. Even at relatively moderate overclocking levels, these things are putting out over twice the heat of a PIII at 1Ghz (which is just about the hottest pre-Willy Intel CPU ever made). At the very edge, it around
four times as much. This is nasty stuff.
I see a lot of posts in forums now where people who are running their systems at speeds where cooling has to be a serious concern have no idea they should be concerned at all.
So we wrote this to give people some fundamentals, some insight into why they get the temperatures they get, and whether or not that a normal or desirable temperature. We gave a brief simple test to let people figure out if they’re OK or not, and if not, what to do about it.
This is an area where people either know ten times more than what’s in this article, or practically nothing outside of a vague notion that 50 or 60C isn’t too good.
This is meant for the second group.
Be the first to comment