SPCR's 2010 CPU Heatsink Test Platform

Want to talk about one of the articles in SPCR? Here's the forum for you.
ekerazha
Posts: 218
Joined: Thu Oct 16, 2008 10:42 am
Location: Italy

Post by ekerazha » Wed Jan 27, 2010 8:03 am

K.Murx wrote:
ekerazha wrote:The executable LinX uses is patched to run on not-Intel systems too. I've tested it on my AMD Turion laptop and it does outclass Prime95 on this platform too.
...err... "patched"? How?
As far as I know, there is no way except a recompile to make the code use the full capabilities (and thus the full power/maximum heat generation) of the CPU. If the code uses the full capabilities of the Intel CPUs (because compiled for them), and not of AMD CPUs (because only "patched"), the relative load on AMD CPUs might be less, thus leading to wrong power dissipation numbers.

This is exacerbated by the fact that the libraries behind the Linpack benchmark are Intel's all the way to the bottom (BLAS).
I did patch it by myself too, it's a simple check at the beginning (so a simple jump instruction to bypass the check).

Maybe it's not fully optimized for not-Intel platforms (like AMD), but as I've already said ;) it still outstrips Prime95 and everthing else as heat generation etc. on AMD platforms too.

K.Murx
Posts: 177
Joined: Tue Mar 17, 2009 10:26 am
Location: Germany

Post by K.Murx » Wed Jan 27, 2010 8:25 am

ekerazha wrote: Maybe it's not fully optimized for not-Intel platforms (like AMD), but as I've already said ;) it still outstrips Prime95 and everthing else as heat generation etc. on AMD platforms too.
I do not dispute that.
But it might be, that Intel CPU's are running at 100% with this program, and that AMD CPU's only at 95% because the (compiled for Intel) program does not utilize all instructions on the AMD platform.
It would be thus fine to use this to generate heat for a heatsink test. But if it would be e.g. used to compare the power use of AMD and Intel CPU's (which SPCR uses iirc Prime95 for), this could lead to underestimation of the AMD power use.

QuietOC
Posts: 1407
Joined: Tue Dec 13, 2005 1:08 pm
Location: Michigan
Contact:

Post by QuietOC » Wed Jan 27, 2010 8:49 am

K.Murx wrote:HDT 1283: $23 after MIR / 22 Euros
The HDT-S1283 has a few issues. I don't like working with it. I have one on a 3.7GHz Athlon II 240, and it will only mount with airflow up into my PS. The stock fan mounting and dismounting is a big pain, and the fan has to be removed first to remove the heatsink. Also the tight fin spacing means it gets clogged up with dust pretty quickly which also requires removing the fan. My old Scythe Ninja has never become dust clogged.
MikeC wrote:You know, I have to ask what's wrong with the data we've already compiled on the P-D950? It all there already on a CPU that has a thermal profile smack dab in the middle of the current pack. Consider the 45W AMDs at the bottom (and they actually draw less than 45W in real apps) and the 130W i7s at the top, with 65W & 95W Intels, 80~95W AMDs around the middle: The 70~75W reacl power draw of our P-D950 is pretty mainstream.
Just a minor nitpick: most "65W" Core 2 Duos are lower load power than the 45W AMDs. The budget "Wolfdale-3M" chips (E7x0/E5x00) are especially low power--this is the same die as found in the 10W and 5.5W TDP CULVs. Even Apple dumped expensive mobile Core 2 Duos to use E7x00s in the latest iMacs.

My Athlon II X2 240 though is pretty darn close to my Pentium Dual Core E5200, and the AMD platform is hugely better (G45<740G not to mention the 785G.) And there's the whole fiasco of Intel reselling their old 65nm southbridges as "new" chipsets.
Last edited by QuietOC on Wed Jan 27, 2010 9:19 am, edited 3 times in total.

lodestar
Posts: 1683
Joined: Fri Aug 05, 2005 3:29 am
Location: UK

Post by lodestar » Wed Jan 27, 2010 9:02 am

One of the problems with not testing in a case is that at least one CPU cooler can never be tested by SPCR, and yet in my opinion it is one of the best 'quiet' coolers around. See http://www.quietpc.com/gb-en-gbp/produc ... kama-angle.

ekerazha
Posts: 218
Joined: Thu Oct 16, 2008 10:42 am
Location: Italy

Post by ekerazha » Wed Jan 27, 2010 10:19 am

K.Murx wrote: It would be thus fine to use this to generate heat for a heatsink test.
This is the point here.
But if it would be e.g. used to compare the power use of AMD and Intel CPU's (which SPCR uses iirc Prime95 for), this could lead to underestimation of the AMD power use.
I'd just say different, you could have that more optimized instructions lead to less power use... or you could have not noticeable differences :)

K.Murx
Posts: 177
Joined: Tue Mar 17, 2009 10:26 am
Location: Germany

Post by K.Murx » Wed Jan 27, 2010 2:03 pm

ekerazha wrote:
K.Murx wrote: It would be thus fine to use this to generate heat for a heatsink test.
This is the point here.
Well, I am not sure if MikeC would want the organizational hassle to have one program for heatsink tests and another one for power estimation.
I'd just say different, you could have that more optimized instructions lead to less power use... or you could have not noticeable differences :)
Well, if someone with access to a recent Intel and AMD CPU and a power meter would like to volunteer, I could write a similar benchmark, based on FLAME and probably the Goto-BLAS. As far as I can see, the benchmark consists of simply solving a random NxN system of linear equations, which should be rather trivial to code (given a LAPACK compatible library). Then we could compare power (and benchmark) figures :)

[edit] Actually, it would be simpler if that person just ran Prime95 and the Linpack benchmark...
Last edited by K.Murx on Wed Jan 27, 2010 2:22 pm, edited 1 time in total.

danimal
Posts: 734
Joined: Mon Jun 08, 2009 2:41 pm
Location: the ether

Post by danimal » Wed Jan 27, 2010 2:09 pm

K.Murx wrote: I do not trust any other website.
that is an emotional response, something that is not based on science or factual information, so it is not relevant to your assertion that the HDT 1283 is a better cooler than the mugen 2 is.
K.Murx wrote:That's why I am here, and the published results of SPCR state that on a Pentium D the HDT is better at the same noise level.
umm, no... if i recall correctly, the spcr comparison was done with an obsolete cpu chip and a mis-matched nexus fan, which was a test scenario that will seldom be duplicated in the real world.

the comparison aspect of the spcr test was therefore irrelevant.

in your zeal to dismiss my criticism about your use of the euro, you must have forgotten to address the apparently missing 1366 & 1156 hardware on the HDT 1283 cooler? perhaps you think that xigmatek will be mailing these parts to everyone for free?

K.Murx
Posts: 177
Joined: Tue Mar 17, 2009 10:26 am
Location: Germany

Post by K.Murx » Wed Jan 27, 2010 2:35 pm

danimal wrote: that is an emotional response, something that is not based on science or factual information, so it is not relevant to your assertion that the HDT 1283 is a better cooler than the mugen 2 is.
The reason I almost exclusively trust SPCR is because other websites tend to test coolers with the stock fan, on different CPU, under different conditions, don't measure noise, or whatever. So why bother searching?
But if you have a source/link, by all means, post it!
danimal wrote:umm, no... if i recall correctly, the spcr comparison was done with an obsolete cpu chip and a mis-matched nexus fan, which was a test scenario that will seldom be duplicated in the real world.
As far as I know the CPU used has always been the same (Pentium D 950). I am not aware that there was any issue/difference with the fan, could you please provide a source?
So far, the comparison aspect seems very valid to me.
danimal wrote:in your zeal to dismiss my criticism about your use of the euro, you must have forgotten to address the apparently missing 1366 & 1156 hardware on the HDT 1283 cooler? perhaps you think that xigmatek will be mailing these parts to everyone for free?
No. It is just that I use AMD parts, which don't screw customers with frequent socket changes ;) So if you are on intel and a value oriented person (unlikely), you need to spend an additional $9 - but that is still $3 less for the (according to the SPCR results) better, lighter cooler.

ekerazha
Posts: 218
Joined: Thu Oct 16, 2008 10:42 am
Location: Italy

Post by ekerazha » Wed Jan 27, 2010 3:09 pm

K.Murx wrote: the organizational hassle to have one program for heatsink tests and another one for power estimation.
If you want better tests it should be done.
Well, if someone with access to a recent Intel and AMD CPU and a power meter would like to volunteer, I could write a similar benchmark, based on FLAME and probably the Goto-BLAS. As far as I can see, the benchmark consists of simply solving a random NxN system of linear equations, which should be rather trivial to code (given a LAPACK compatible library). Then we could compare power (and benchmark) figures :)
If you can write a platform agnostic application like the Intel LINPACK benchmark but without platform specific optimizations, which can generate as heat as the Intel LINPACK banchmark, it would be great (write it using a cross-platform graphical toolkit like QT4 or as a simple console application so I can write a graphical frontend or as a dynamic library so I can link the library from a graphical interface :D ). Note it should be multi-threaded.

MikeC
Site Admin
Posts: 12285
Joined: Sun Aug 11, 2002 3:26 pm
Location: Vancouver, BC, Canada
Contact:

Post by MikeC » Wed Jan 27, 2010 3:34 pm

ekerazha wrote:
K.Murx wrote: the organizational hassle to have one program for heatsink tests and another one for power estimation.
If you want better tests it should be done.
Forget it. :roll:
Find some other sucker web site for such OT bs. This is utterly useless to the vast majority of computer users -- including most spcr visitors.

eckre
Posts: 17
Joined: Fri Jan 14, 2005 4:26 pm

Yawn.... Frostytech is better

Post by eckre » Wed Jan 27, 2010 3:47 pm

Meh, I haven't found anything at ALL as near as good and scientific as Frostytech's methods. Nobody else can touch them when you want to do some real hard core comparing. I only wish they reviewed more fans quicker.

http://www.frostytech.com/

danimal
Posts: 734
Joined: Mon Jun 08, 2009 2:41 pm
Location: the ether

Post by danimal » Wed Jan 27, 2010 10:50 pm

K.Murx wrote:The reason I almost exclusively trust SPCR is because other websites tend to test coolers with the stock fan, on different CPU, under different conditions, don't measure noise, or whatever. So why bother searching?
But if you have a source/link, by all means, post it!
i already did post it.

now it's up to you to back up your claims that it's not valid.
K.Murx wrote:As far as I know the CPU used has always been the same (Pentium D 950). I am not aware that there was any issue/difference with the fan, could you please provide a source?
So far, the comparison aspect seems very valid to me.
what is "valid" about using a nexus fan that is three degrees hotter than the stock mugen 2 fan?

Volts Mugen2Fan NexusSPCRfan
12v 33 36
10/9v 35 38
9/7v 38 40
http://www.silentpcreview.com/article961-page5.html

the xbit labs testing proved that as well:
"The tower-heatsinks – Prolimatech Megahalems and ThermoLab BARAM – were topped with a fan from Scythe Mugen 2 with the rotation speed varying between 250 and 1300RPM."
http://www.xbitlabs.com/articles/cooler ... up_18.html
using the same mugen 2 fan, the mugen 2 and the Prolimatech Megahalems were both at 42/72 degrees.
K.Murx wrote:No. It is just that I use AMD parts, which don't screw customers with frequent socket changes ;)
not all of us are willing to settle for an underpowered amd cpu :wink:
K.Murx wrote:So if you are on intel and a value oriented person (unlikely), you need to spend an additional $9 - but that is still $3 less for the (according to the SPCR results) better, lighter cooler.
yes, both the dollar and euro numbers that you posted earlier were wrong.

the weight of the cooler is irrelevant, because it's not a liability, it's an asset... according to your failed logic, there should be no cooler at all, because it weighs less :roll: :D

K.Murx
Posts: 177
Joined: Tue Mar 17, 2009 10:26 am
Location: Germany

Post by K.Murx » Thu Jan 28, 2010 12:07 am

*checkss xbitlabs and frostytech*
Both site test with different fans on different heatsinks. Thus it is not a valid comparison. I mean, I am still at SPCR, right? It's about noise here - and that means that at least I want to know what heatsink is the best at a certain amount of (fan-)noise generated.
frostytech has not reviewed the Mugen 2 (thanks for wasting my time), and Xbitlabs has not only tested with different fans but on different CPU's as well. Are you kidding me or trying to troll me?
danimal wrote: what is "valid" about using a nexus fan that is three degrees hotter than the stock mugen 2 fan?
Everything. Because if you switch fan you can't compare the heatsinks. If MikeC would switch the "refrence fan" to the Scythe Mugen 2 one, he would invalidate all old results.
danimal wrote:yes, both the dollar and euro numbers that you posted earlier were wrong.
Excuse me? I could walk down to the store and get both heatsinks at the quoted euro price. Presuming you are in North America, newegg would ship you the coolers at this price.
Yes, I omitted the Intel mounting issue. That does not make the quoted prices "wrong", and as pointed out before, even if you include the mounting bracket, the US$ price would still be lower (no I won't check the Euro one...).
danimal wrote: the weight of the cooler is irrelevant, because it's not a liability, it's an asset... according to your failed logic, there should be no cooler at all, because it weighs less :roll: :D
So, if you have two almost identically performing heatsinks, you choose the heavier one?
Weight by itself is not an asset. If (hypothetically) two heatinks performed identical, and one weighed 500 grams, the other 1000 grams, why would you buy the heavier?
That would put more stress on your mainboard, have a higher risk when you move the case around, etc. Those things do not seem to be desirable to me.

ekerazha
Posts: 218
Joined: Thu Oct 16, 2008 10:42 am
Location: Italy

Post by ekerazha » Thu Jan 28, 2010 1:40 am

MikeC wrote:
ekerazha wrote:
K.Murx wrote: the organizational hassle to have one program for heatsink tests and another one for power estimation.
If you want better tests it should be done.
Forget it. :roll:
Find some other sucker web site for such OT bs. This is utterly useless to the vast majority of computer users -- including most spcr visitors.
Clearly it is not OT (it's strictly related to the testing methodology), however I understand you are still irritated with me because of the poor impression you made in the Fortress FT02 topic, so that's ok.

dev
Posts: 43
Joined: Sun May 10, 2009 12:47 am
Location: Here

Post by dev » Thu Jan 28, 2010 6:27 am

When I'm going to use a heated piece of metal instead of a cpu frostytech is going to be the first site I'm going to get my info from. Until then the 'testing' on that site is utter nonsense.

Mike I think the current methodology is great and people with a decent grasp on things will figure out which cooler is efficient for any platform. I mean if one heatsink can handle the hottest cpu available Im pretty sure it will do great on anything less.

danimal
Posts: 734
Joined: Mon Jun 08, 2009 2:41 pm
Location: the ether

Post by danimal » Thu Jan 28, 2010 12:44 pm

K.Murx wrote:*checkss xbitlabs and frostytech*
Both site test with different fans on different heatsinks.
that entire paragraph is another one of your gross mis-generalizations.
K.Murx wrote:
danimal wrote: what is "valid" about using a nexus fan that is three degrees hotter than the stock mugen 2 fan?
Everything. Because if you switch fan you can't compare the heatsinks. If MikeC would switch the "refrence fan" to the Scythe Mugen 2 one, he would invalidate all old results.
are you dyslexic? what part of "nexus fan is three degrees hotter" do you not understand?

stop avoiding the subject.

what part of "nobody in the real world is going to pay extra $$$ to swap the superior scythe fan for the inferior nexus fan" do you not understand?
K.Murx wrote: Yes, I omitted the Intel mounting issue.
yes, you totally mis-stated the pricing, i'm glad that we can agree on something.
K.Murx wrote:So, if you have two almost identically performing heatsinks, you choose the heavier one?
irrelevant... as i already proved, we don't have two almost identically performing heatsinks here.

MikeC
Site Admin
Posts: 12285
Joined: Sun Aug 11, 2002 3:26 pm
Location: Vancouver, BC, Canada
Contact:

Post by MikeC » Thu Jan 28, 2010 1:26 pm

MikeC wrote:
danimal wrote:there is a rather thorny problem with the same reference fan being used on all of the coolers, because it is inferior to the stock mugen 2 fan at higher rpms, by about three degrees, as your earlier tests proved:
http://www.silentpcreview.com/article961-page5.html
Not really. At the same SPL/rpm, there is no appreciable cooling difference between those fans. We will not change the reference fan unless and until we find a 120mm fan that's clearly quieter at the same airflow level. There are some contenders in the huge Scythe lineup, but we haven't proved it for ourselves yet. Never enough time.
danimal and K.Murx -- You guys need to calm down. Stop yelling at each other. No more warnings before a ban.

I cite my own comment above because this is totally relevant: The fan on the Mugen2 has a higher default speed (at 12V) and is noisier at that speed than our Nexus. But when it is set to the same speed, the noise and cooling effect is identical to the Nexus.

I'm not sure why you don't understand what this means, daminal.

To clear the air: What do we mean by a "reference fan"? We mean a point of reference, a benchmark, something that we can compare other things to or measure other things by. It does not mean "the best", which so many people seem to think it means. It's true that at one point, the Nexus 120mm fan was about the quietest fan $ could buy, but I don't believe this is true any more. There are lots of other fans that are nice sounding, some of which might even have higher airflow at the same noise level as our Nexus. But it's probably gong to STAY as our reference fan, because 5+ years of testing data with this fan is too much to spoil.

This is the same thinking behind our use of the Pentium D 950 all these years -- it's a consistent reference point. In fact, after much thought and discussion about the whole range of HS testing issues, we've decided to bring back the Pentium D for testing of non-massive heatsinks. We're going to go with a different motherboard that has better VRM cooling (and hence more likely to be more robust and stable) and proves to be considerably more efficient. However, as long as the same CPU is used, the heat "seen" by the heatsink will be the same as before, on the now retired mobo. With the same reference CPU & fan, reviews on this platform can be fairly compared to our big collection of existing HS tests.

So, there will be two tiers of HS -- those for most CPUs, and those for super hot processors -- the >100W models. We know the HS will be divided by size, and a HS will be tested on one or the other platform, not both. Not all the fine details have been worked out; once they are, an addendum will be made to the HS test platform article.

cb95014
Posts: 109
Joined: Wed Jun 27, 2007 9:39 pm
Location: The insane State of California

Post by cb95014 » Thu Jan 28, 2010 2:00 pm

MikeC wrote: You guys need to calm down. Stop yelling at each other. No more warnings before a ban.
Thank you! One of the things that I really like about the SPCR forums is the decorum. If I want to read flames, there are plenty of other sites. :lol:

K.Murx
Posts: 177
Joined: Tue Mar 17, 2009 10:26 am
Location: Germany

Post by K.Murx » Thu Jan 28, 2010 3:04 pm

cb95014 wrote: Thank you! One of the things that I really like about the SPCR forums is the decorum. If I want to read flames, there are plenty of other sites. :lol:
Well, it that case I assume my tone was inappropriate and apologize to the other readers inconvenienced. I guess I got carried away a bit.
I'll just leave the argument "as is", everything necessary has been said, anyway.

Now, with that out of the way, back to the benchmark/load generator issue:
To recapitulate, the Intel Linpack benchmark provides a heavier load than prime95 on both AMD and Intel CPU's. However, the benchmark originates from Intel's MKL library, which in my experience [I'm mostly using the eigenvalue routines] is heavily optimized for Intel's CPUs and provides suboptimal performance on AMD CPU's.
At least to me, that leaves unclear whether the power figures obtained with this tool are reliable.
I would offer to code up a similar benchmark to the Intel Linpack program (solve a NxN dimensional linear equation system, several passes, etc.). This would be compiled once as a "platform agnostic" reference implementation, using the netlib reference BLAS and LAPACK implementation, and gfortran 4.4 -mtune=generic.
Then I would provide binaries optimized for each platform, which means compiled with ifort/*a-certain-non-free-for-educational-use-compiler*, FLAME and the Intel/AMD version of the Goto-BLAS. These are the compiler/library combinations that deliver the best performance (yes, better than MKL), at least for the eigenvalue problems I usually deal with.

However, I do not have (private) access to neither recent Intel CPUs nor a power meter, and do not intend to buy either. So I would need someone that could actually run the benchmarks and report results. That would obviously include running obscured binaries from me - while I will naturally provide the source and Makefile, the person would need to jump through quite some setup steps to actually build the benchmark. I just happen to have done that from my last library and compiler hunting session and it is not exactly what I would call "fun" ;)

But before I start writing that, I would like to see someone provide power figures for prime95 and the Intel Linpack benchmark on both AMD and Intel CPU's. If the differences there seem consistent then I guess I could save the effort (and would suggest to consider using the Linpack benchmark in the future).
Last edited by K.Murx on Thu Jan 28, 2010 3:56 pm, edited 1 time in total.

dhanson865
Posts: 2198
Joined: Thu Feb 10, 2005 11:20 am
Location: TN, USA

Post by dhanson865 » Thu Jan 28, 2010 3:56 pm

MikeC wrote:We will not change the reference fan unless and until we find a 120mm fan that's clearly quieter at the same airflow level.

But it's probably gong to STAY as our reference fan, because 5+ years of testing data with this fan is too much to spoil.

This is the same thinking behind our use of the Pentium D 950 all these years -- it's a consistent reference point. In fact, after much thought and discussion about the whole range of HS testing issues, we've decided to bring back the Pentium D for testing of non-massive heatsinks. We're going to go with a different motherboard that has better VRM cooling (and hence more likely to be more robust and stable) and proves to be considerably more efficient. However, as long as the same CPU is used, the heat "seen" by the heatsink will be the same as before, on the now retired mobo. With the same reference CPU & fan, reviews on this platform can be fairly compared to our big collection of existing HS tests.

So, there will be two tiers of HS -- those for most CPUs, and those for super hot processors -- the >100W models. We know the HS will be divided by size, and a HS will be tested on one or the other platform, not both. Not all the fine details have been worked out; once they are, an addendum will be made to the HS test platform article.
+1 to the two tiers idea. I'm all for it.

My only question is how many years can you keep using the "reference fan" before the bearings wear out? Do you have an ample supply such that you can swap out fans over the years and still feel comfortable that the baseline hasn't changed or have you already done so (swapped fans) and decided thats a solved problem?

Oh and +1 to the whole calm it down concept as well. This doesn't need to be so intense.

cmthomson
Posts: 1266
Joined: Sun Oct 09, 2005 8:35 am
Location: Pleasanton, CA

Post by cmthomson » Thu Jan 28, 2010 4:57 pm

QuietOC wrote:The HDT-S1283 has a few issues. I don't like working with it. I have one on a 3.7GHz Athlon II 240, and it will only mount with airflow up into my PS. The stock fan mounting and dismounting is a big pain, and the fan has to be removed first to remove the heatsink. Also the tight fin spacing means it gets clogged up with dust pretty quickly which also requires removing the fan. My old Scythe Ninja has never become dust clogged.
All fair points. The original Ninja is an awesome HS if you can mount it on your MB (I still use it). But its fan mounting is also a pain, so no plus there.

I continue to recommend the 1283 because its performance is almost identical to the original Ninja, it can be mounted on many motherboards, and it is dirt cheap.

All that said, I haven't had occasion to try a Megahalem, so maybe all this yak yak is about a favorite but end-of-life product.

Johnsy
Posts: 26
Joined: Thu Jan 28, 2010 4:46 pm
Location: U.K

heat/power coefficient

Post by Johnsy » Thu Jan 28, 2010 5:18 pm

The biggest casualty of the new platform is the notion that a universalized cooling coefficient (deg.C/W) can be derived for a specific heatsink employing a specific fan at a specific speed.

Take the Megahalems for example. With a heat source of ~78W (D950), we get a coefficient of 0.22 deg.C/W when the fan is spinning at 720rpm (7V). If we take the i7-965 to be ~112W, employing that coefficient should yield a delta of just under 25 deg.C (112 x 0.22). As we see, however, the actual tested figure is far higher - in this case 42 deg.C.

The equivalent coefficient derived from this test would be 0.375 deg.C/W.

MikeC
Site Admin
Posts: 12285
Joined: Sun Aug 11, 2002 3:26 pm
Location: Vancouver, BC, Canada
Contact:

Re: heat/power coefficient

Post by MikeC » Thu Jan 28, 2010 5:40 pm

Johnsy wrote:The biggest casualty of the new platform is the notion that a universalized cooling coefficient (deg.C/W) can be derived for a specific heatsink employing a specific fan at a specific speed.

Take the Megahalems for example. With a heat source of ~78W (D950), we get a coefficient of 0.22 deg.C/W when the fan is spinning at 720rpm (7V). If we take the i7-965 to be ~112W, employing that coefficient should yield a delta of just under 25 deg.C (112 x 0.22). As we see, however, the actual tested figure is far higher - in this case 42 deg.C.

The equivalent coefficient derived from this test would be 0.375 deg.C/W.
Not really true because we really don't know exactly how much power/heat is dissipated by each CPU. All of our power numbers are approximations. This idea would be best tested on a CPU simulation device like that used by Frostytech, but on real CPUs, w/o real power measurements, we really can't confirm/verify. So for all practical purposes, deg. C/W is not usable -- no one we know has that kind of CPU power measurement equipment, and there are significant variants in the power profile of every CPU made (ie, a TDP of 130W only tells us that this is the higher power measured by Intel on a wide range of i7 processors.

Johnsy
Posts: 26
Joined: Thu Jan 28, 2010 4:46 pm
Location: U.K

Post by Johnsy » Thu Jan 28, 2010 6:32 pm

How exact does it need to be?

Taking a set of reasonable notional values for the i7, along with the measured temp. rise at 7V, shows that the lower the load actually is, the worse any derived coefficient will be.

42/86 = 0.48
42/100 = 0.42
42/110 = 0.38
42/120 = 0.35
42/130 = 0.32
(All in deg.C/W)

To recreate the best of these coefficients (which assumes the i7 to be at it's max TDP of 130W) where the temp. delta is 17 deg.C (D950 with 7V fan), the D950 would have to have actually been producing only around 53-54W.

I agree with your decision to abandon the deg.C/W figures, but for different reasons.

MikeC
Site Admin
Posts: 12285
Joined: Sun Aug 11, 2002 3:26 pm
Location: Vancouver, BC, Canada
Contact:

Post by MikeC » Thu Jan 28, 2010 7:35 pm

Johnsy --

A couple of things:

1) The 78W for the PD950 was measured using a different tool. It measures 86W w/ our current tool.
2) But this number is not the CPU's power; it includes the VRM losses. My guess is ~80% efficiency, which means CPU power is at ~62W.
3) The 112W estimate for our i7 is really just that -- less an estimate and more speculation, given the complexities of i7-1366 power.

I'm not defending the C/W relevance; just saying that the info we have on our i7 and our PD950 aren't enough to dismiss it altogether.

We did just take delivery of a CPU die simulator much like Frostytech's and are in the process of gathering the components needed to make it operational for a whole nuther project -- testing fans. (More on that later.) But we could test the relevance of C/W with that device, because we will be able to dial in precise amounts of heat into the die.

Can C/W from one test be used to predict HS behavior with a hotter or cooler CPU, and to what range -- ie, you really can't expect the C/W of a HS to stay the same between a 50W CPU and a 100W CPU -- but how about a 50W CPU vs a 75W CPU? It's a question that's interesting to me because C/W relevance was long upheld by some heavies in the web tech world of HS testing. (Not sure if it still is...)

Of course, even if the C/W relevance can be verified, it's still not useful in the real world, because you need the exact power dissipation of the CPU on which the HS was tested, and the power of the CPU that you're trying to predict the same HS's performance on.

Sorry. I'm dithering on....... :oops:

Johnsy
Posts: 26
Joined: Thu Jan 28, 2010 4:46 pm
Location: U.K

Post by Johnsy » Thu Jan 28, 2010 8:38 pm

Can C/W from one test be used to predict HS behavior with a hotter or cooler CPU
If it can't, then what's it for?
and to what range
And there, I think, is the rub. I personally have always valued C/W figures, but the testing on the new platform has caused me to consider more closely something I previously took as read.

Obviously any heatsink will have an upper limit to the amount of heat it is able to practically dissipate, and equally obviously, C/W figures will only be valid up to a point somewhere below that maximum.

Notwithstanding the fact that we don't have exact figures for power dissipation on either platform, employing a reasonable hypothetical range shows considerable discrepancies in C/W between the two. I suppose what I'd like to know is whether the breakdown is gradual and linear - in which case, it could perhaps be factored in - or whether it occurs more abruptly near the limit of the heatsink's practical capacity - in which case a range would need to be specified alongside C/W.

This isn't a question I expect SPCR to answer, of course (though I look forward to your findings with the hotplate), but hopefully other readers may find the discussion stimulating.

danimal
Posts: 734
Joined: Mon Jun 08, 2009 2:41 pm
Location: the ether

Post by danimal » Thu Jan 28, 2010 10:30 pm

MikeC wrote:I cite my own comment above because this is totally relevant: The fan on the Mugen2 has a higher default speed (at 12V) and is noisier at that speed than our Nexus. I'm not sure why you don't understand what this means, daminal.
i can see that higher speed is a logical explanation for why the mugen 2 fan cools better than the nexus does.

why did scythe put a higher rpm fan on the high-impedance mugen 2 cooler? because it requires more air pressure to realize it's full cooling capacity.

"When a standard fan is included, we run two sets of tests, one with our reference fan, and one with supplied fan."
http://www.silentpcreview.com/article1018-page2.html

since spcr did not run the i7 cooling test with the supplied fan, the mugen 2 was never able to realize it's full cooling capacity... if testing for noise was the goal, the supplied fan would have been tested, and the final rankings would have reflected fan noise, instead of cooling capacity only.

it sounds like the new testing procedure will address that.

Cistron
Posts: 618
Joined: Fri Mar 14, 2008 5:18 am
Location: London, UK

Post by Cistron » Fri Jan 29, 2010 1:26 am

MikeC wrote:Can C/W from one test be used to predict HS behavior with a hotter or cooler CPU, and to what range -- ie, you really can't expect the C/W of a HS to stay the same between a 50W CPU and a 100W CPU -- but how about a 50W CPU vs a 75W CPU? It's a question that's interesting to me because C/W relevance was long upheld by some heavies in the web tech world of HS testing. (Not sure if it still is...)
Very interesting question. My gut feeling says that it will vary with different heat sources, as warmth transport is based on temperature differences (in an ideal world) - being more effective with greater differences.

anato
Friend of SPCR
Posts: 4
Joined: Wed Sep 17, 2008 5:08 am
Location: Finland

Only two sticks of ram?

Post by anato » Fri Jan 29, 2010 2:47 am

In the article SPCR's 2010 CPU Heatsink Test Platform you specify that there are only two sticks of ram used in i7-965 "2GB QiMonda DDR3 memory. 2 x 1GB DDR3-1066". Is this a mistake or intentional?

Nehalem LGA1366 i7 has triple channel memory architecture so why only use two ram modules? I thinks this will affect the at least memory controllers power usage and can affect processors performance.

Zenphic
Friend of SPCR
Posts: 107
Joined: Sun Jan 20, 2008 12:54 pm
Location: Quebec, Canada

Post by Zenphic » Fri Jan 29, 2010 6:38 pm

The Cooler Master Hyper 212 Plus would have been a nice addition to this article; a very popular cooler nowadays. :)

Post Reply