Real world c/W figures
Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee
Real world c/W figures
Has anyone else gone and tried to calculate observed c/W figures for their quiet/silent cooling solutions? Here are my figures:
2.40GHz @ 1.10V:
85.3W TDP * (1.10V / 1.40V) ^ 2 = 52.7W actual
(35c load - 18c ambient) / 52.7W actual = 0.323c/W
2.40GHz @ 1.40V:
(59c load - 18c ambient) / 85.3W actual = 0.481c/W
My cooling setup is a passive SI-128 cooling the CPU. I have two D12SL-12 intakes blowing towards the CPU heatsink and no exhaust fans.
I think most of my observed numbers seem reasonable. Does anyone have any ideas why I'd be getting this descrepancy? I'd also be curious to see what kind of numbers others were getting.
2.40GHz @ 1.10V:
85.3W TDP * (1.10V / 1.40V) ^ 2 = 52.7W actual
(35c load - 18c ambient) / 52.7W actual = 0.323c/W
2.40GHz @ 1.40V:
(59c load - 18c ambient) / 85.3W actual = 0.481c/W
My cooling setup is a passive SI-128 cooling the CPU. I have two D12SL-12 intakes blowing towards the CPU heatsink and no exhaust fans.
I think most of my observed numbers seem reasonable. Does anyone have any ideas why I'd be getting this descrepancy? I'd also be curious to see what kind of numbers others were getting.
yeh, because temperature sensors are f.o.s. ... I've noticed also that HSFs seem to be pretty linear up to a 'point' where they can't really handle much more heat and the temps go up quite a bit, no idea how true this is, but its just my observation. Out of interest what are you measuring the CPU temps with? CoreTemp or the motherboard sensors ?
I'm just reporting whatever sensor RM Clock happens to be reading. I have no idea if this is on die or on the motherboard. But I thought all CPUs since Pentium 4s had on die thermistors? So what would the point be for thermistors on the motherboard?TMM wrote:yeh, because temperature sensors are f.o.s. ... I've noticed also that HSFs seem to be pretty linear up to a 'point' where they can't really handle much more heat and the temps go up quite a bit, no idea how true this is, but its just my observation. Out of interest what are you measuring the CPU temps with? CoreTemp or the motherboard sensors ?
Also, I don't understand how heatsinks could be linear only to a point in most cases? Unless we're approaching the melting point of a particular heatsink or some other strange property, I don't know why it would deviate from linear.
I suspect that you're using the wrong ambient temps. You need to use the in-case temperature, not the in-the-room ambient temp. There's no good on-mobo sensor to measure that though. You'll need to stick a thermometer in there, near where the HSF is drawing air into itself, and see what temps you get. I'd bet it's around 25°. That will shift the calc's quite a bit.
Plus, your TDP's are probably wonky. If the TDP is off for the stock vcore, multiplying it by the reduced vcore ratio makes the resulting number off by a different percentage than the stock TDP was. (What is the CPU anyway?)
Plus....TMM is sorta right. Heatsink performance is linear...so long as it is a conventional solid-state heatsink. Heatsinks with heatpipes are not as linear, since the heatpipe mechanism's efficiency varies with the difference in temperature between its hot and cold ends. As the hot end gets hotter, or the cold end gets colder relative to each other, the heatpipe becomes more efficient, up to a point.
Plus....while modern CPU's do all have on-board temperature diodes, the motherboard's IC chip frequently messes with that number before it gets reported, so what RMclock is seeing may not be what the CPU is actually reporting.
Plus, your TDP's are probably wonky. If the TDP is off for the stock vcore, multiplying it by the reduced vcore ratio makes the resulting number off by a different percentage than the stock TDP was. (What is the CPU anyway?)
Plus....TMM is sorta right. Heatsink performance is linear...so long as it is a conventional solid-state heatsink. Heatsinks with heatpipes are not as linear, since the heatpipe mechanism's efficiency varies with the difference in temperature between its hot and cold ends. As the hot end gets hotter, or the cold end gets colder relative to each other, the heatpipe becomes more efficient, up to a point.
Plus....while modern CPU's do all have on-board temperature diodes, the motherboard's IC chip frequently messes with that number before it gets reported, so what RMclock is seeing may not be what the CPU is actually reporting.
Responses inline.
Rusty075 wrote:I suspect that you're using the wrong ambient temps. You need to use the in-case temperature, not the in-the-room ambient temp. There's no good on-mobo sensor to measure that though. You'll need to stick a thermometer in there, near where the HSF is drawing air into itself, and see what temps you get. I'd bet it's around 25°. That will shift the calc's quite a bit.
The two 120mm intake fans that provide the only active cooling for the heatsink are at most 1" away from the heatsink, so I'd think using the room ambient temperature would be fairly accurate for these calculations.
Plus, your TDP's are probably wonky. If the TDP is off for the stock vcore, multiplying it by the reduced vcore ratio makes the resulting number off by a different percentage than the stock TDP was. (What is the CPU anyway?)
The CPU I'm using is an Orleans A64 3800+. Even if the TDP figure I got from AMD's website is off, I'm using the same figure in both calculations, so it should be a consistent error that propagates through to the results. The voltage difference term I used is pretty straighforward, since my testing was done at 1.40V and 1.10V, so I assume that it should be at the very least a valid ratio of TDPs from the two trials.
Plus....TMM is sorta right. Heatsink performance is linear...so long as it is a conventional solid-state heatsink. Heatsinks with heatpipes are not as linear, since the heatpipe mechanism's efficiency varies with the difference in temperature between its hot and cold ends. As the hot end gets hotter, or the cold end gets colder relative to each other, the heatpipe becomes more efficient, up to a point.
Good point on the heatpipes being a cause of non-linear function for the CPU heatsink.
Plus....while modern CPU's do all have on-board temperature diodes, the motherboard's IC chip frequently messes with that number before it gets reported, so what RMclock is seeing may not be what the CPU is actually reporting.
Your TDP value is not correct, AMD gives TDP of AM2 A64 3800 as 62W:
http://www.amdcompare.com/us-en/desktop ... 3800IAA4CN
http://www.amdcompare.com/us-en/desktop ... 3800IAA4CN
Wow, thanks for finding that, jaganath. I have gone ahead and recalculated my c/W figures with that value:jaganath wrote:Your TDP value is not correct, AMD gives TDP of AM2 A64 3800 as 62W:
http://www.amdcompare.com/us-en/desktop ... 3800IAA4CN
2.40GHz @ 1.10V:
62W TDP * (1.10V / 1.40V) ^ 2 = 38.3W actual
(35c load - 18c ambient) / 38.3W actual = 0.444c/W
2.40GHz @ 1.40V:
(59c load - 18c ambient) / 62W actual = 0.661c/W
As I figured, the stock c/W figure is still 49% higher than the undervolted figure, regardless of the TDP used. Looking at the calculations, it seems the only way to explain this huge descrepancy is either that the temperature readings are terribly inaccurate or that the CPU heatsink is very non-linear in this heat range.
Ah. i tryed RM Clock on my system and it reads the same as CoreTemp (link), so it should be reading the actual on die temp for you aswell. All Athlon64s/Opterons, most Pentium-M and all CoreDuo/Core2Duo support on die temps - but whether a piece of software reads it or the motherboards sensor is another matter.ciz28 wrote:I'm just reporting whatever sensor RM Clock happens to be reading. I have no idea if this is on die or on the motherboard. But I thought all CPUs since Pentium 4s had on die thermistors? So what would the point be for thermistors on the motherboard?TMM wrote:yeh, because temperature sensors are f.o.s. ... I've noticed also that HSFs seem to be pretty linear up to a 'point' where they can't really handle much more heat and the temps go up quite a bit, no idea how true this is, but its just my observation. Out of interest what are you measuring the CPU temps with? CoreTemp or the motherboard sensors ?