Lurfys
Thu Mar 20 07:44:53 PDT 2008
On Thu, 20 Mar 2008 18:55:00 +1100, "Mark Dormer" <markd@mvpsx.org>
wrote:
>"Lurfys Maw" <invalid@invalid.invalid> wrote in message news:19s3u3djc2a88vfvjikd3jikkc8fm8ombi@4ax.com...
>> Are there any guidelines about how hard a CPU can be run and for how
>> long and/or how to tell if a CPU is overheating?
>>
>> The other day, I downloaded BOINC to run on an old laptop that I
>> mostly use for testing shareware. It's an HP Omnibook 6000.
>>
>> BOINC (Berkeley Open Infrastructure for Network Computing) is the
>> software that allows various compute-intensive projects such as SETI
>> (Search for ExtraTerrestrial Intelligence) to ships small pieces of
>> the computations out to PCs around the world to sop up their idle
>> cycles. (Probably everyone already knew that.)
>>
>> BOINC allows me to set various parameters that control how much of my
>> PCs resources it will use. One parameter controls the percentage of
>> the processor. I can set it from 10% to 100% in 10% increments.
>>
>> It looks like BOINC achieves the throttling back by starting and
>> stopping the calculations. The Task Manager Performence graph is kinda
>> fun to watch. Each little grid on the graph appears to be about 5
>> seconds wide. BOINC completes about two cycles each grid.
>>
>> I have done a little testing at various settings.
>>
>> At 100%, the fan comes on tight away and runs continuously. The bottom
>> of the laptop gets fairly hot to the touch -- not so hot that I can't
>> stand it, but close.
>>
>> At 90% & 80%, the fan comes on and runs continuously, but the laptop
>> doesn't feel quire as hot.
>>
>> At 70%, the fan is on almost all the time. It occasionally shuts off
>> for a few seconds. I just tested it and it did not shut off at all for
>> 10 minutes.
>>
>> At 60%, the fan is on about 90% of the time.
>>
>> At 50%, the fan is on about 65% of the time and the laptop is cooler
>> to the touch.
>>
>> At 40%, the fan is on about 30% of the time.
>>
>> At 30%, the fan comes on occasionally for a few seconds.
>>
>> At 20% and 10%, the fan never comes on.
>>
>> Is it safe to run BOINC continuously (24/7) at 100%? If not, what is
>> the highest level that can I run it continuously without damaging the
>> processor?
>>
>> Would it be better to measure the actual processor temperature? If so,
>> what is the best way to do that? My hand is not a reliable
>> thermometer.
>
>I have run BOINC on a laptop for weeks on end with no issue.
>Well it costs money and wears out faster but that is to be expected.
I realize that any use of the laptop incurs wear. What I am trying to
determine is whether any of the higher BOINC usage settings will incur
excessive wear or premature failure -- significantly beyond normal
usage.
I believe heat is one of the primary, if not the primary, CPU killers.
That's why I posted the fan percentages. When the fan is running, the
CPU is over the threshhold the manufacturer thought was safe. The fan
is there to bring it back down to a safe level. Right?
My intuition tells me that if the fan comes on briefly, the machine is
operating within safe levels. I would think that using the 30% setting
or less should cause little more than normal wear on the machine. Is
that a safe assumption?
At the other end of the spectrum, the 90-100% settings may be causing
considerably more than normal wear. If the fan is running
continuously, the CPU is never below the manufacturer's threshold and,
since it never shuts off, we cannot be sure how much over the
threshold it is. How likely is it that these settings are causing much
more than normal wear?
My working hypothesis is that given the measurements above, the 10-50%
settings should not cause excessive (more than normal) wear. The
80-100% settings probably are causing excessive wear. But what about
the 60-70% settings?
Talking about "wear" is probably misleading. Since the CPU has no
moving parts to "wear out", executing instructions may not inherently
cause "wear". Heat is probably the dominant, if not the only
consideration.
The fan never comes on with either 10% or the 20% setting. The CPU is
probably warmer at the 20% setting than at the 10% setting, but does
not reach the fan threshold. It is working twice as hard. If it were a
gear, it would wear out twice as fast. My hunch is that CPUs running
at the 20% setting would not, on average, fail anything like twice as
quickly as those running at the 10% setting.
I would like to see a graph showing life expectancy as a function of
operating temperature. Does life expectancy decrease linearly with
rising temperature or it there a threshold at which life expectancy
drops steeply.
My intuition tells me that as operating temperature increases, life
expectancy decreases more slowly at lower temperatures than at higher
temperatures. A graph with life expectancy on the vertical in hours
and CPU temperature on the horizontal in degrees would be some type of
concave down curve with a slope that is slightly negative at low
temperatures and becomes more and more steeply negative with higher
temperatures. I also might expect a discontinuity at or just above the
fan threshold where life expectancy drop much more steeply as
components start to melt.
A quick search turned up this link:
http://www.overclockers.com/tips30/
About 5-6 screens down, there is a formula
CPU Life = Normal Life Hours / [((273 + New Temp) / (273 + Normal
Temp)) ^ M]
and a graph. The graph plots life expectancy vs temperature for three
CPUs: a "hardy" CPU, an "average" CPU, and a "weak" CPU.
Surprisingly (to me), the curve is concave up. This says that
increasing the CPU temp from 25C to 30C (77F-86F) reduces life
expectancy more than increasing it from 70C to 80C (158F-167F).
It looks like an exponential decay function, which would say that life
expectancy has some lower limit, which is nonsense. Once the CPU
melts, life expectancy is zero.
If I read the graph correctly, increasing the CPU temp from 25C (77F)
to 30C (86F) decreasing lfe expectancy from 30K to 27.5K for the hardy
CPU and 23K for the weak CPU. But increasing the temp from 70C (158F)
to 75C (167F) only decreases the life expectancy from 15K to 14K for
the hardy CPU and from 2K to 1.5K for the weak CPU.
My guess is that this graph is theoretical rather than empirical. I
would wager than emperical data would be very different.