Would it be possible to make a graph (consisting of at least 4-5 points) where y-axis is noise emitted (dBA), while x-axis is C/W? Probably, ambient temperature should be held at a reference value (20 degrees or 40 degrees?).
One single such curve would tell all that is relevant to us about a cooler:
*maximum cooling capability
*cooling capability at a given noise level
*noise level at a given cooling capability
*predictions about performance if run outside normal spec (extremely low fan voltage for instance).
Further, many such measurements of different coolers could be stacked together in one plot, for comparing any number of fan, cooler or integrated fan/cooler solutions.
I have posted a pic in another forum:
http://forum.hardware.no/index.php?s=&a ... 70966&st=0
I know that spcr have a very evolved test methology, and I am not suggesting to change it, only presenting the same data slightly different.
with regards
Knut Inge
Presenting CPU coolers differently in reviews?
Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee
While I think thats a good idea, I don't think its a viable one. It would be difficult enough to keep the ambient temperatures level but to keep the same level of background noise would be basically impossible given spcrs' budget.
Personally I think it would weaken the precise nature of the reviews. I'm sure many people already overlook the ambient test conditions and introducing comparable graphs would compound the issue for newcomers, possibly leading them to make poor decisions.
Just my tuppence.
Personally I think it would weaken the precise nature of the reviews. I'm sure many people already overlook the ambient test conditions and introducing comparable graphs would compound the issue for newcomers, possibly leading them to make poor decisions.
Just my tuppence.
But:
Logging ambient and CPU temp, fan and CPU voltage precisely is doable? Keeping CPU power constant is also possible? Then we have the prequisites for estimating the rise in CPU temp above amient temp that a given cooler produce for a given CPU/load for a given fan voltage.
Within reasonable limits, changes in ambient temp should not matter that much then, am I right? (In other words, a given setup condition will give X degrees celcius above ambient temp regardless of whether the room is 10 deg or 40 deg). This seems to be a condition for using C/W anyways.
Ambient noise is a problem no matter test method. It will produce a noise floor that is a lower limit to measured fan noise, as well as introduce uncertainty in measurements. By logging ambient noise between fan measurements, reliability of measurements may be indicated.
with regards
Knut Inge
Logging ambient and CPU temp, fan and CPU voltage precisely is doable? Keeping CPU power constant is also possible? Then we have the prequisites for estimating the rise in CPU temp above amient temp that a given cooler produce for a given CPU/load for a given fan voltage.
Within reasonable limits, changes in ambient temp should not matter that much then, am I right? (In other words, a given setup condition will give X degrees celcius above ambient temp regardless of whether the room is 10 deg or 40 deg). This seems to be a condition for using C/W anyways.
Ambient noise is a problem no matter test method. It will produce a noise floor that is a lower limit to measured fan noise, as well as introduce uncertainty in measurements. By logging ambient noise between fan measurements, reliability of measurements may be indicated.
with regards
Knut Inge
Why is that? My motivation is finding a presentation that can compare integrated fan/cooler solutions to different coolers with different fans. Of course, every inversion is not feasible, and the tester would have to make a judgement call on what combinations to include. But within those limitations, you should know exactly how silent you can make your current CPU, and with what cooler/fan combo...cotdt wrote:i think it would be a great idea! of course it would have to be a well-selected fan and the same fan would be used for everything.
K.
-
dukla2000
- *Lifetime Patron*
- Posts: 1465
- Joined: Sun Mar 09, 2003 12:27 pm
- Location: Reading.England.EU
The graph you are looking for has been tried by BeHardware in this review. Except they did temp delta rather than C/W but in theory equivalent as long as W is constant.
I found those curves interesting, but then got critical and decided I would prefer if they had used the same fan on each heatsink (practically impossible with 92mm and 120mm, let alone Zalman). Then I worked out they had only used their overclocked results to draw the curves: reasonable as long as the axis is delta C: your suggestion for C/W would enable all data to be used. Last, they had decided shapes for the curves and I couldn't work out how: maybe some best fit algorithm?
But isn't there enough data in the SPCR reviews to draw SPCR versions of these curves anyway?
I found those curves interesting, but then got critical and decided I would prefer if they had used the same fan on each heatsink (practically impossible with 92mm and 120mm, let alone Zalman). Then I worked out they had only used their overclocked results to draw the curves: reasonable as long as the axis is delta C: your suggestion for C/W would enable all data to be used. Last, they had decided shapes for the curves and I couldn't work out how: maybe some best fit algorithm?
But isn't there enough data in the SPCR reviews to draw SPCR versions of these curves anyway?
I think it's an interesting idea, and could be useful so long as its limitations are spelled out with the graphs. Instead of in the reviews themselves, it seems like a better fit in the Recommend Heatsinks article. In a perfect world we could do something similar to what pH did at procooling with their Interactive Waterblock Test Resultsgraph.
Ambient temperature variations and differences in CPU wattage wouldn't affect the results, as both are factored into the °C/W number.
We would have to careful about giving it too much weight though: the variety of test platforms and the vagaries of dBa measurements means that there would be a pretty substantial amount of fuzziness in the graph results. It wouldn't be reliable enough to compare coolers that are within a few percentage points of each other, but would probably illustrate general trends.
Instead of graphing °C/W vs dBa, it might be better to graph °C/W vs CFM. The reason being that CFM is an easier number to come by - most of our HSF tests don't include detailed dBa readings. We could use manufacturer data and formulae to work out the approximate CFM for pretty much any fan under any voltage. And since CFM and noise are generally tied together, the graph results would be just as useful.
Ambient temperature variations and differences in CPU wattage wouldn't affect the results, as both are factored into the °C/W number.
We would have to careful about giving it too much weight though: the variety of test platforms and the vagaries of dBa measurements means that there would be a pretty substantial amount of fuzziness in the graph results. It wouldn't be reliable enough to compare coolers that are within a few percentage points of each other, but would probably illustrate general trends.
Instead of graphing °C/W vs dBa, it might be better to graph °C/W vs CFM. The reason being that CFM is an easier number to come by - most of our HSF tests don't include detailed dBa readings. We could use manufacturer data and formulae to work out the approximate CFM for pretty much any fan under any voltage. And since CFM and noise are generally tied together, the graph results would be just as useful.
-
Devonavar
- SPCR Reviewer
- Posts: 1850
- Joined: Sun Sep 21, 2003 11:23 am
- Location: Vancouver, BC, Canada
I agree, some kind of graph could be very useful, as long as we can figure out how to produce it reliably and accurately.
There are a lot of practical issues here that need to be thought through. I also have reservations about how closely two graphs can be compared. There are a huge number of variables, all with error tolerances that are not that small.
Also, it is very difficult to consistently hold noise or temperature constant. Trying to consistently hold temperature rise constant would take ages because each tweak requires >15m to stabilize.
Noise is more practical, except that heatsinks with fans that can't easily be swapped (Zalman) don't necessarily have the same range of noise of other fans. This is partly why we don't always use our reference fans. That said, at the low noise levels that we test at, noise quality is often a more important factor than volume, and a graph can't capture quality.
I disagree with Rusty about CFM. Of all the measurements we do, I would say that CFM is most prone to error.
Any suggestions for how to deal with these?
There are a lot of practical issues here that need to be thought through. I also have reservations about how closely two graphs can be compared. There are a huge number of variables, all with error tolerances that are not that small.
Also, it is very difficult to consistently hold noise or temperature constant. Trying to consistently hold temperature rise constant would take ages because each tweak requires >15m to stabilize.
Noise is more practical, except that heatsinks with fans that can't easily be swapped (Zalman) don't necessarily have the same range of noise of other fans. This is partly why we don't always use our reference fans. That said, at the low noise levels that we test at, noise quality is often a more important factor than volume, and a graph can't capture quality.
I disagree with Rusty about CFM. Of all the measurements we do, I would say that CFM is most prone to error.
Any suggestions for how to deal with these?
C/W versus dBA sounds good! but i think the same reference fans should be used on all heatsinks. the better fans have almost zero bearing noise so i don't see a problem with the fan selection.
this might even make results more accurate as a computer can do a plot to fit the data that may be closer to the actual results than the actual data. this is frequently observed in the science labs that i participate in.
this might even make results more accurate as a computer can do a plot to fit the data that may be closer to the actual results than the actual data. this is frequently observed in the science labs that i participate in.
What I'd personally like to see is testing with a single standard 80mm fan attached to a standardized cylindrical duct. By tweaking it to particular rpms, the CFM may be reasonably approximated (with variations due to differences in backpressure).Devonavar wrote:I disagree with Rusty about CFM. Of all the measurements we do, I would say that CFM is most prone to error.
Any suggestions for how to deal with these?
If a CPU cooler has a fixed stock fan, remove it!
This way, all CPU coolers could be rated using the same fan, and the error in CFM may be reduced.
Im not sure if this is possible or helpful.
Cfm readings are hard to do the only reliable way would be to use a duct and take pito tube readings and sum. Their way your looking at about an hour per data point.
Secondly fan geometry and heatsink geometeries are different. Some fans may owrk better with different heatsinks. Interaction between the blades and the heatsink fins are going to cause problems and noise.
Scaling dBa is tricky as its a very perceptual to each person kind of thing.
if you say dBA=f(fan cfm)
and then go through the links
cfm=f(rpm)
rpm=f(power)
power=f(voltage)
it gets tricky and errors start coming in allover the place.
In the end im not really sure if it helps at all everyone has their fav brand of fan and the vagueness of heatsink descisions (is it too big, does it help with mobo cooling, weight) makes everything fuzzy. The procooling link works because waterblocks are quite similar and easy to quanitfy. They cool proportional to flow rate and they are a set restricivness (dp= k*q^2 for the most part) could probably define all blocks by three numbers price, pressure drop constant and cooling constant.
Cfm readings are hard to do the only reliable way would be to use a duct and take pito tube readings and sum. Their way your looking at about an hour per data point.
Secondly fan geometry and heatsink geometeries are different. Some fans may owrk better with different heatsinks. Interaction between the blades and the heatsink fins are going to cause problems and noise.
Scaling dBa is tricky as its a very perceptual to each person kind of thing.
if you say dBA=f(fan cfm)
and then go through the links
cfm=f(rpm)
rpm=f(power)
power=f(voltage)
it gets tricky and errors start coming in allover the place.
In the end im not really sure if it helps at all everyone has their fav brand of fan and the vagueness of heatsink descisions (is it too big, does it help with mobo cooling, weight) makes everything fuzzy. The procooling link works because waterblocks are quite similar and easy to quanitfy. They cool proportional to flow rate and they are a set restricivness (dp= k*q^2 for the most part) could probably define all blocks by three numbers price, pressure drop constant and cooling constant.