When is 100% not 100%?

The FPU unit would be far more active at 200FSB as there would be a lot less
waiting around for main memory. So the heat output of the CPU would rise.

 

I have recently changed my FSB/multiplier on my CPU. It used to run at
133FSB X 16 multi, 1.825vcore for 2.13GHz. Athlon XP1800+ Tbred-B. I had no
choice as one of my memory modules (Dual-channel mode) was PC2100 (2 X
256MB). I now have two PC3200 modules and have changed the settings to
200FSB X 10.5 for 2.1 GHz and have been able to drop the vcore to 1.80 volts
and still have it run stably. I see a slight improvement in both CPU and
memory scores in PCMark 2002. (It won't do 11X even at 1,850 vcore, the
highest setting available to me without pin-mods, I think I have reached the
capabilities of my CPU without radical cooling, I'm on air).

However, running at 100% (SETI and/or Prime95) my CPU temp had gone up
nearly 5°C compared with 100% load previously with the CPU running slightly
faster and with a higher vcore. I find this paradoxical. I have since
rectified the problem at the expense of extra fan noise.

What would explain this phenomena? Isn't 100% in fact 100%? Would the
(presumed) faster CPU through-put account for this?

Just curiousity, I actually expected lower CPU (diode) temps, particularly
as I was able to lower vcore.

Thanks for any input.

 

Could it be that the on chip cache is now running at 400MHTZ instead of
266MHTZ causing the chip to run a little warmer ? This would also increase
the through put of the chip even at the same speed if it was a bottleneck
previously.

 

I've found that running Prime95 along with other programs will result in
less heat production than with Prime95 on it's own. Presumably the context
switching reduces the load.

I've always assumed heat production is directly related to voltage and MHz,
never really thought about it, but it also seem likely that if you end up
doing calculations faster, you would expect power consumption to increase.

I would have expected that too, but on memory intensive tasks it would seem
that the faster memory bus reduces the latency significantly enough to
enable more calculations and therefore more power consumption.

Ben

 

Did ya clean the processor's heatsink and fan? and the north bridge or
is it south controller's fan ?? I noticed huge temp drops with a simply
monthly cleaning of a few minutes with a cotton swap

 

Yep. I don't have a 'sink on the southbridge but I clean the CPU/northbridge
'sinks regularly.

Thanks.

 

No. The on-chip cache runs at CPU speed, not 400 or 266. So in fact the
cache is running slower.

Thanks,

 

I've found the same.

That seems to be it. I guess if the CPU is benchmarking better (and it is)
then it must be drawing more current? It just took me by surprise, I've
never heard this discussed or mentioned and found it strange.

Thanks for your input.

 

-
~misfit~ stood up at show-n-tell, in
AksAb.21101$, and said:

I believe that by dropping your core voltage, the processor must increase
it's work. Back in my K6-2 overclocking days, it was widely seen that
increasing the vcore, just a bit, would actually decrease max temps...

 

OK, that sounds feasible. It's just the increase in heat far outwieghed the
increase in CPU performance (% for %).

Thanks

 

Ok, food for thought. I may try upping the vcore a little again. Previously
to running a 200Mhz FSB, every time I increased vcore I got an increase in
temp.

Thanks,

 

Seems logical, more voltage means the electrons travel faster or
something like that? ;p

 

I can't think of a single scientific theory that could predict the CPU
would 'need to work less hard' with more vcore.

 

-
David Maynard stood up at show-n-tell, in , and
said:

Neither can I.

 

I was hoping for your input into this thread David. What do you think is the
cause of my CPU running 5°C+ hotter when it's actually running at a slightly
lower frequency (2.1GHz instead of 2.13) with less vcore (1.80v instead of
1.825v)?

The FSB has been increased from 133MHz to 200MHz (finally replaced my old
PC2100 RAM with two 256MB sticks of PC3200 in dual-channel mode to go with
my new nForce2 mobo).

It's nothing to do with a rise in case temps. My CPU is force-fed air from
outside the case. I'm leaning towards the hypothesis that it's just getting
more work done due to the higher through-put/FSB speed. 100% before was 100%
of what it could do within the limitations imposed by the data available.
Now more data is available to the CPU it can do more work. 100% now is
probably 105 or 110% of the work it could do before. Therefore more heat is
produced.

Make sense?

 

Well, it's hard to tell because there are a lot of variables. Like, are you
100% sure there's no change to the average temp of the air coming in? AndI
don't mean just room temp; changing case orientation (or something else in
the room) can make a significant difference (changing airflow patterns,
less airflow, stagnation from nearby structures, etc.).

The FSB increase, and hence a faster feed to the processor, would sure seem
to be a source of more processing taking place. That's a potential 50%
increase in memory bandwidth; just from the FSB. I'm not sure how much more
you're getting from the dual channel. The vcore change represents only
about a 2.8% reduction in power.

But when you say "new nforce2 mobo" that makes me wonder what all has
changed from the 'old' reading.

Here's a thought for you. Assuming nothing else changed, perhaps simply the
act of unplugging and plugging in new memory caused enough of a disturbance
to affect the heatsink (like maybe the thermal compound contact).

What was the original temp?

But yes, I would suspect it's the increase in memory bandwidth increasing
CPU utilization.

Btw, I noticed the cache speed comment. What that misses, though, is that
the FSB buffers, which have to drive the external bus, consume more power
than internal devices, so an FSB increase would increase the power
consumption for those buffers. But I couldn't find any numbers to use for
that in the AMD data sheets.

 

ICs are all about capacitance... you need to charge each gate to make it
switch, the more voltage you have the faster you can charge it, once it's
charged you clock it and wait for the next set of gates to be charged. If
you do not charge the gates enough (either because you're clocking too fast
or charging too slowly) then you will get errors.

Power consumption (and therefore heat production) is directly related to how
many gates need charging per unit time, and the charge you put on them.
More voltage = more charge (I think thats a linear relationship, I could be
wrong it was a while ago). It would seem a bit strange that an increase in
voltage would decreases temps.

Ben

 

-
Ben Pope stood up at show-n-tell, in
bqv6qd$25v0fs$-berlin.de, and said:

Yes, I agree. I was, merely, stating an observation that seemed widespread
with the K6-2+'s... Of course, this is from memory, nothing more. Perhaps
I should have been a bit more specific, in my prior comment. It seemed (or
so it was reported) that once overclocked, a very small (0.1-0.2V) increase
in core voltage dropped the temperature (which had, consequently, increased
due to the initial overclocking) by about 1-2degC. Obviously, this can be
attributed to measurement error/variation, etc... However, from what I
remember, it was pretty consistent. And, at a certain point, raising the
core had detrimental effects (for obvious reasons).

I'm of the opinion that not everything in the universe is
known/fathomed/figured out Then, again, it's been over ten years since
I've studied PChem/Quantum Mechanics and Physics. So, perhaps, I should've
just kept my fingers shut.

I've jumped the AMD ship, in the past year, so it's really academic to me
anyway. And, as usual, I'll leave the disclaimer that my memory is not what
it used to be, yada, yada...

 

I understood that it was from observation and I do believe you, I was just
thinking aloud really to see if I could come up with a reason... the best
one so far is that the increased current drain from the processor reduces
the voltage on the sensor, so when reading it back the temperature seems
lower. (not a very good theory

This is a public forum, say what you like

Ben

 

I have observed this increase in temperature with a higher fsb compared to
upping the multiplier (and have mentioned it in other posts) - I just put it
down to the fact that by increasing the fsb you are making memory related
things run faster as well as fsb things and subsequently they give off more
heat. - technical or what