SPCR's 2010 CPU Heatsink Test Platform
-
- Site Admin
- Posts: 12285
- Joined: Sun Aug 11, 2002 3:26 pm
- Location: Vancouver, BC, Canada
- Contact:
javitxi --
The main reason to consider a lower power test platform is not to establish that the best quiet coolers on i7 are usually the best quiet coolers on anything else. That's obvious.
What's not so obvious is that some heatsinks that are perfectly good on other CPUs simply will not "pass" our stock i7 test... if by pass we mean obtain adequate cooling with low enough noise. Those low profile coolers with quiet fans that we discussed elsewhere might be at the verge of i7 throttling w/ temps in the 80s. Yet, on a typical mainstream CPU (with real TDP of 70W like our P-D950), those same coolers cool quietly w/o any problem.
Will all SPCR readers understand this when they read the review about i7 temps in the high 80s with the fan at full tilt? I doubt it.
The main reason to consider a lower power test platform is not to establish that the best quiet coolers on i7 are usually the best quiet coolers on anything else. That's obvious.
What's not so obvious is that some heatsinks that are perfectly good on other CPUs simply will not "pass" our stock i7 test... if by pass we mean obtain adequate cooling with low enough noise. Those low profile coolers with quiet fans that we discussed elsewhere might be at the verge of i7 throttling w/ temps in the 80s. Yet, on a typical mainstream CPU (with real TDP of 70W like our P-D950), those same coolers cool quietly w/o any problem.
Will all SPCR readers understand this when they read the review about i7 temps in the high 80s with the fan at full tilt? I doubt it.
Mike I say you givem time. They will adjust once 775 platforms become obsolete. Once the prices for i5/i3 settle down you'll see the majority of them migrating. And it will be easier for them to understand that a cooler good for an i7 cpu will be great for anything less.
One small quibble: when reaching a conclusion, about a cooler performing not that good at 1000 rpm or below, you might want to consider that certain products are not made for low rpm. The Zalman 10x Extreme is one of them.
One small quibble: when reaching a conclusion, about a cooler performing not that good at 1000 rpm or below, you might want to consider that certain products are not made for low rpm. The Zalman 10x Extreme is one of them.
tests should be done in a case. nobody is using open. results would be different inside a case. you can't even say results will hold relatively inside the case compared to open. you can't interpolate the results to inside case.jmke wrote:testing outside a case... makes passive cooling tests worthless, since inside a case with a single exhaust fan, you'll have more airflow and completely different end result.you should also include results with no fan,
they can pick up a reference case such as Antec 180 or Solo. this would give much better idea. you would know that if you have the same components you would get the same results.
-
- *Lifetime Patron*
- Posts: 541
- Joined: Sun Jan 04, 2004 11:35 am
- Location: Chicago, Ill., USA
- Contact:
Another vote for this... based on the previous review of this heatsink, my conclusion was that it was a definite stand-out, given its price relative to other heatsinks of similar performance. It seems unmatched in the "cooling performance per dollar" criteria.Monkeh16 wrote:Any chance of getting the Xigmatek HDT-S1283 retested?
I'm in the process of building an Athlon II X2 250 system, and using this heatsink. The HDT-S1283 is definitely overkill for this CPU, but, it can be had for $30 (after rebate) with free shipping from newegg at the moment. When you include the cost of shipping (and tax), it's hard to get anything decent for less than $20, and in the $20 to $30 range (as far as I can tell), nothing comes close to the performance of the HDT-S1283.
First,
thanks for the very insightful article!
I generally recommend/use Minja for non-extreme machines and I would like to know how much performance I lose by going for not-the-most-massive class of heatsinks.
Naturally if someone can point me to worthy competitors for Minja, that would be lovely. Something that's "good enough" to cool <100W CPU inaudibly, has less mass and torque towards MB and does not break your bank. This far Minja has fitted this bill very well.
thanks for the very insightful article!
This would be brilliant. Also, I would love to see some known good smaller coolers (like Minja) be measured against these behemoths on a modern machine with <100W CPU.MikeC wrote:A few points that may have been missed:
3) As it stands, the new test platform is most useful for the OC / gaming silencer, who represent a sizable but still minority portion of the SPCR audience. As a result...
4) A lower power test platform looks unavoidable. We may keep the P-D950 platform alive and add a lower power test to simulate typical 65W processors. This could be used for smaller, less ambitious heatsinks; the i7 platform would be used for the biggies. Just thoughts for now...
I generally recommend/use Minja for non-extreme machines and I would like to know how much performance I lose by going for not-the-most-massive class of heatsinks.
Naturally if someone can point me to worthy competitors for Minja, that would be lovely. Something that's "good enough" to cool <100W CPU inaudibly, has less mass and torque towards MB and does not break your bank. This far Minja has fitted this bill very well.
I would not artificially limit heatsinks per the airflow direction (sideways ws. towards MB), especially as results of this article seem to imply the airflow towards the MB does not affect VRM performance.javitxi wrote: - low profile /"blow the air to the motherboard-profile" types, because all of them do the same task (Noctua NH-C12P, Kabuto, Nexus low-7000, Big Shuriken, etc)
Myself I found the low-profile HS roundups posted quite recently very helpful. The problem with low-profiles is that what's low enough depends on your particular case.. so the best you can do is what SPCR already did, say that "is you have less height clearance than X cm, this is the best of the bunch".
zds
About Ninja. I'm wondering the same thing and if I'm correctly deciphering data from the two test (links below), the original Ninja is still one of the very best cooler out there, easily in the top 5, if not 3-4.
newest platform results and Copper Ninja
I wrote about it in the 1st page of discussion
About Ninja. I'm wondering the same thing and if I'm correctly deciphering data from the two test (links below), the original Ninja is still one of the very best cooler out there, easily in the top 5, if not 3-4.
newest platform results and Copper Ninja
I wrote about it in the 1st page of discussion
-
- Site Admin
- Posts: 12285
- Joined: Sun Aug 11, 2002 3:26 pm
- Location: Vancouver, BC, Canada
- Contact:
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.zds wrote:First,
thanks for the very insightful article!
This would be brilliant. Also, I would love to see some known good smaller coolers (like Minja) be measured against these behemoths on a modern machine with <100W CPU.
When I wrote about needing a lower power test, I was thinking of new as yet untested HS. A 2-power 2nd test setup -- for maybe 50W and 85W at the AUX12V.
-
- *Lifetime Patron*
- Posts: 541
- Joined: Sun Jan 04, 2004 11:35 am
- Location: Chicago, Ill., USA
- Contact:
Another random thought... what about an "artificial CPU"? For example, take a large piece of steel or wood, and in the center, cut a hole exactly big enough for a real CPU headspreader. Mount (somehow) that headspreader, then drill holes in the surrounding structure where you can mount a heatsink/fan. Then put a burning candle underneath the heatspreader. Obviously, you'd want something more controlled and measurable, but hopefully you get the point.
In other words, a pure, controllable heat source, where the heat is focused to an actual CPU heatspreader.
Then you could simulate everything from a low-powered CPU at idle, up to the hottest CPUs at full load.
I have vague memories of one review site doing that in the past. But maybe I dreamed that.
In other words, a pure, controllable heat source, where the heat is focused to an actual CPU heatspreader.
Then you could simulate everything from a low-powered CPU at idle, up to the hottest CPUs at full load.
I have vague memories of one review site doing that in the past. But maybe I dreamed that.
You have a point there. It might just be that I had one discussion too many about "that review is from 2008 and on old test platform, it must be outdated now" today :-D.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.
Simulated testing platforms are the worst idea ever. Or its not that bad if your trying to promote one technology very aggressively.
Because even cpus from the same batch differ in overclocking characteristics and heat output what good will you achieve if you consider all of them equal to a piece of heated metal. Its like talking about how a car drives without owning a drivers licence and a car.
Because even cpus from the same batch differ in overclocking characteristics and heat output what good will you achieve if you consider all of them equal to a piece of heated metal. Its like talking about how a car drives without owning a drivers licence and a car.
Having gone through this latest article and low-profile heatsink roundups once again, I concluded something like this:
-If you want the very best performance, go for Megahalems.
-If you have very hot CPU and want bang-for-buck, get Scythe Mugen2
-If you have less hot CPU, consider also Ninja Mini and Big Shuriken.
What is missing from this is Ninja Mini numbers with 92mm fan, one and two.
The fan that Minja ships with is 80mm, yes, but 92mm Nexus or similar is what most noise-conscious users will want to use on it because the width of the heatsink greatly exceeds even 92mm and it ships with mounting hardware for both.
So, if you go with stock fans, Big Shuriken wins over Minja. However, Minja is easy to equip with larger (92mm) fan and almost as easy to equip with two, so my hunch says Minja would win over Big Shuriken on low- to medium power CPUs (<100W), especially as you can add another fan if the results are not good enough for you.
-If you want the very best performance, go for Megahalems.
-If you have very hot CPU and want bang-for-buck, get Scythe Mugen2
-If you have less hot CPU, consider also Ninja Mini and Big Shuriken.
What is missing from this is Ninja Mini numbers with 92mm fan, one and two.
The fan that Minja ships with is 80mm, yes, but 92mm Nexus or similar is what most noise-conscious users will want to use on it because the width of the heatsink greatly exceeds even 92mm and it ships with mounting hardware for both.
So, if you go with stock fans, Big Shuriken wins over Minja. However, Minja is easy to equip with larger (92mm) fan and almost as easy to equip with two, so my hunch says Minja would win over Big Shuriken on low- to medium power CPUs (<100W), especially as you can add another fan if the results are not good enough for you.
You should also update your burn-test application, LinX ( http://www.xtremesystems.org/forums/sho ... p?t=201670 ), based on Intel LINPACK, is far better than Prime95.
True, it's a heavy one. But I'd personally rather want to have comparison to the old platform and previously tested coolers. Even tho Prime95 won't heat up the CPU as heavily as Linx, it still represents unrealistic load. Games, F@H, picture editing - none of these is as heavy on the CPU as Prime95, Orthos, Linx, OCCT etc.ekerazha wrote:You should also update your burn-test application, LinX...
Slightly outside this range, the CoGage True Spirit is only $35 + shipping from SVC, and includes a 12cm PWM fan.matt_garman wrote:in the $20 to $30 range (as far as I can tell), nothing comes close to the performance of the HDT-S1283.
CoGage is Thermalright's 'value' brand, so as far as I can tell, this is is a four-pipe version of the TRUE (TRU? ) and should perform extremely well. It also is compatible with the various Thermalright mounting kits.
http://www.svc.com/truespirit.html
HDT 1283: $23 after MIR / 22 Euroszds wrote:-If you want the very best performance, go for Megahalems.
-If you have very hot CPU and want bang-for-buck, get Scythe Mugen2
-If you have less hot CPU, consider also Ninja Mini and Big Shuriken.
Scythe Mugen: $35 / 33 Euros
HDT is significantly lighter (less stress on MB), and has better cooling performance (tested on the Pentium D).
Would you like to reconsider?
But... LINPACK has long since been superseded by LAPACK, which has largely been superseded by FLAME in the HPC community.ekerazha wrote:You should also update your burn-test application, LinX ( http://www.xtremesystems.org/forums/sho ... p?t=201670 ), based on Intel LINPACK, is far better than Prime95.
And why on earth would you use a vendor provided library for benchmarking???
Additionally, all those Libraries rely heavily on the BLAS, so performance and load will vary a lot with whatever (and please, NOT vendor-specific) implementation of BLAS is chosen...
-
- *Lifetime Patron*
- Posts: 541
- Joined: Sun Jan 04, 2004 11:35 am
- Location: Chicago, Ill., USA
- Contact:
Why?dev wrote:Simulated testing platforms are the worst idea ever.
By that logic, then using just one CPU is the same as driving just one car for a car review.dev wrote:Because even cpus from the same batch differ in overclocking characteristics and heat output what good will you achieve if you consider all of them equal to a piece of heated metal. Its like talking about how a car drives without owning a drivers licence and a car.
What good will you achieve if you consider all CPUs equal to the one (and only one) used in the test rig? When you just said that CPUs, even within the same batch, can have different characteristics and tweaking features?
What you seem to be suggesting is ideal---and I certainly agree---is to have a huge collection of actual CPUs for use in heatsink testing. That includes several CPUs of the same model, to make up for manufacturing variation. But that's just not practical. But I don't see why it couldn't be approximated with a simulator.
What we're concerned about with heatsink testing really doesn't have anything to do with the CPU anyway. What it ultimately boils down to is a product's ability to transfer heat from the die (generally via the heatspreader). Whatever can extract the most heat is best.
To me the car analogy doesn't hold. We're not really testing the car---it's more akin to testing a by product of using the car. For example, a muffler or catalytic converter. And with these types of products, I'd argue that you don't need an actual car and driver's license---just something that outputs chemical toxins or noise.
A thought I have is how much can the i7 platform be under-volted and clocked? If it can be brought down to 65-85w level through heavy under clocking and under volting this will save work / time as just reboot and change BIOS settings rather than having to use a separate platform. It would also allow heatsinks that are middling to be compared on the high and low ends. If you could / did go this route personally for the heavy weight heatsinks I think you'd only need to test the lowest / quietest fan configs. It's obvious if you use a large heatsink on a cool CPU with a high performance fan that the cooling is going to be great! What's less obvious is how it might perform for someone who buys a high-end heatsink to put on a middle or low heat CPU and then uses a 500rpm fan in the hope of still getting good cooling with super low noise.
You would need a bit in the reviews stating the power level of the CPU and what current CPU's it's equivalent to so people don't get confused, hopefully!
Regards, Seb
You would need a bit in the reviews stating the power level of the CPU and what current CPU's it's equivalent to so people don't get confused, hopefully!
Regards, Seb
-
- Site Admin
- Posts: 12285
- Joined: Sun Aug 11, 2002 3:26 pm
- Location: Vancouver, BC, Canada
- Contact:
Tried this a bit. Got the system power down to about 135W w/CPU at 1.6 GHz. This is fairly close to the D950 heat profile -- probably still 5W or more higher. Temp readings are quite a bit higher than with D950 -- about 10C higher. Haven't tried undervolting.SebRad wrote:A thought I have is how much can the i7 platform be under-volted and clocked? If it can be brought down to 65-85w level through heavy under clocking and under volting this will save work / time as just reboot and change BIOS settings rather than having to use a separate platform. It would also allow heatsinks that are middling to be compared on the high and low ends. If you could / did go this route personally for the heavy weight heatsinks I think you'd only need to test the lowest / quietest fan configs. It's obvious if you use a large heatsink on a cool CPU with a high performance fan that the cooling is going to be great! What's less obvious is how it might perform for someone who buys a high-end heatsink to put on a middle or low heat CPU and then uses a 500rpm fan in the hope of still getting good cooling with super low noise.
You would need a bit in the reviews stating the power level of the CPU and what current CPU's it's equivalent to so people don't get confused, hopefully!
Regards, Seb
good idea... maybe you should reconsider that most people don't run a pentium d anymore... or reconsider why spcr went to an i7 920 testing platform, before quoting test results that don't jive with any other website... perhaps you could reconsider that newegg is having a fire sale on the HDT 1283 because it apparently doesn't ship with either 1156 or 1366 socket adapters, they aren't even listed in the specs... reconsider that you are claiming euros for a rebate that may only be available in the u.s.... etc.K.Murx wrote:HDT 1283: $23 after MIR / 22 Euroszds wrote:-If you want the very best performance, go for Megahalems.
-If you have very hot CPU and want bang-for-buck, get Scythe Mugen2
-If you have less hot CPU, consider also Ninja Mini and Big Shuriken.
Scythe Mugen: $35 / 33 Euros
HDT is significantly lighter (less stress on MB), and has better cooling performance (tested on the Pentium D).
Would you like to reconsider?...
I do not trust any other website. 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.danimal wrote:good idea... maybe you should reconsider that most people don't run a pentium d anymore... or reconsider why spcr went to an i7 920 testing platform, before quoting test results that don't jive with any other website...
And until the HDT will be retested (which I would like to see, too), you can speculate all you want what effect a newer CPU would have on those results - I stick with the ones published.
Err, no. As you can very easily see, I did not calculate the Euro price from the $ price - it has been a very long time since the US $ was worth as much as the Euro.danimal wrote:perhaps you could reconsider that newegg is having a fire sale on the HDT 1283 because it apparently doesn't ship with either 1156 or 1366 socket adapters, they aren't even listed in the specs... reconsider that you are claiming euros for a rebate that may only be available in the u.s.... etc.
The Euro prices are from
http://geizhals.at/deutschland/ ( a price comparison site for German speaking countries)
I think some people still haven't grasped the concept that the test platform is to compare heatsinks, not to compare platforms. The more is kept constant, the easier the comparison will be. Whether one can achieve the same temperatures with a tested heatsink in a personal built is not the question Mike & colleagues are trying to answer here.
I believe the Pentium-D results were perfectly fine -- for my needs --, but I appreciate that the bigger LGA1366 heatspreader makes a bit of a difference. And boys want new toys, don't we?
I believe the Pentium-D results were perfectly fine -- for my needs --, but I appreciate that the bigger LGA1366 heatspreader makes a bit of a difference. And boys want new toys, don't we?
This benchmark should not be confused with LINPACK, the library, which has been expanded upon by the LAPACK library.K.Murx wrote:But... LINPACK has long since been superseded by LAPACK, which has largely been superseded by FLAME in the HPC community.ekerazha wrote:You should also update your burn-test application, LinX ( http://www.xtremesystems.org/forums/sho ... p?t=201670 ), based on Intel LINPACK, is far better than Prime95.
And why on earth would you use a vendor provided library for benchmarking???
Additionally, all those Libraries rely heavily on the BLAS, so performance and load will vary a lot with whatever (and please, NOT vendor-specific) implementation of BLAS is chosen...
Source: Intel
Moreover, I'm not saying to use it as benchmark but as burn-test, this is what LinX does and it's probably the heavier burn-test you can run today... Prime95 is outdated and you could think a system is stable because it does run Prime95, while using LinX it could miserably fail.
MikeC , I am really greatful for the work you and others involved in SPCR do. Your thoroughness in your reviews and testing is why I keep coming back to SPCR for buying advice.
Let me wish for three new coolers to be tested on the new platform. They all have the potential to best the Prolima Megahalems.
Noctua NH-D14
This is Noctua's top tower cooler and should best the NH-U12P by a fair margin. It is however significantly bulkier and heaver (900g vs 600g without a fan) which might sway people away from it.
Titan Fenrir
An unknown brand too me, but a quick searh shows that it is readily available in retail. This is the current champion at bit-tech.net beating both the Thermalright Ultra Extreme and the Noctua NH-U12P in a limited test. The review can be found here.
Corsair H50 "water cooling"
By far the most interesting of the three. Like the CoolIT Domino ALC this is a closed loop watercooling system with a bump, cpu-block, reservoir and radiator in one nice package for about $80. It is developed by the danish company Asetek who already has a presence in the watercooling market, but probably is most famous for their vapochill products. From the current crop of reviews it seems to be good quality and fairly quite, here, here (swedish) and here. None of theese have the eyes and ears of the SPCR crew though![/url]
Let me wish for three new coolers to be tested on the new platform. They all have the potential to best the Prolima Megahalems.
Noctua NH-D14
This is Noctua's top tower cooler and should best the NH-U12P by a fair margin. It is however significantly bulkier and heaver (900g vs 600g without a fan) which might sway people away from it.
Titan Fenrir
An unknown brand too me, but a quick searh shows that it is readily available in retail. This is the current champion at bit-tech.net beating both the Thermalright Ultra Extreme and the Noctua NH-U12P in a limited test. The review can be found here.
Corsair H50 "water cooling"
By far the most interesting of the three. Like the CoolIT Domino ALC this is a closed loop watercooling system with a bump, cpu-block, reservoir and radiator in one nice package for about $80. It is developed by the danish company Asetek who already has a presence in the watercooling market, but probably is most famous for their vapochill products. From the current crop of reviews it seems to be good quality and fairly quite, here, here (swedish) and here. None of theese have the eyes and ears of the SPCR crew though![/url]
I hate Intels naming schemes... But you're right. And as I just realized we are not talking about using that tool with AMD processors too, but just to heat up the i7 that is used... Yeah, it kinda makes sense to use Intel's benchmark/library...ekerazha wrote: This benchmark should not be confused with LINPACK, the library, which has been expanded upon by the LAPACK library.
Source: Intel
Moreover, I'm not saying to use it as benchmark but as burn-test, this is what LinX does and it's probably the heavier burn-test you can run today... Prime95 is outdated and you could think a system is stable because it does run Prime95, while using LinX it could miserably fail.
Sorry!
Although, maybe it would be preferable to use a tool that AMD users can use, too? The benchmark is Intel-only (and, at a cursory glance, source code is not provided), and even though some of the binaries would probably run on AMD too, they would not utilize all possibilities/instructions on AMD hardware.
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.K.Murx wrote: I hate Intels naming schemes... But you're right. And as I just realized we are not talking about using that tool with AMD processors too, but just to heat up the i7 that is used... Yeah, it kinda makes sense to use Intel's benchmark/library...
Sorry!
Although, maybe it would be preferable to use a tool that AMD users can use, too? The benchmark is Intel-only (and, at a cursory glance, source code is not provided), and even though some of the binaries would probably run on AMD too, they would not utilize all possibilities/instructions on AMD hardware.
...err... "patched"? How?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.
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).