[POLL] EPIC SHOOTOUT MATCH!!!! AS5 vs NT-H1 on Thermalright Ultra-120 vs. EVGA ACX!!!

[DrMrLordX]

at any rate nickel's thermal conductivity is still pretty high, I don't see a reason to remove it

It's 90.7 W/mK @ 300K, which is better than TIM but worse than copper. May as well just remove it while "normalizing" the surface (lapping). Also, I have found that, at least on budget AMD CPUs that I have lapped (x2-3600+, x4 635, 7700k), that the thickness of the nickel coating is variant based on the underlying copper surface. Usually the copper is uneven, so to produce a flat surface, they laid on nickel in a correspondingly-uneven fashion.

It isn't a huge deal, but it's good for maybe 1-2C alone.

 

[jj109]

How thick is the nickel plating on average? 0.1 mm? 0.05 mm? At 0.1mm, we're looking at a minimum of 300W/K through the plating for a FX-83xx.

Doesn't really feel worth spending time sanding it off unless they sloughed a lot of nickel on top of the copper.

 

[soccerballtux]

It's 90.7 W/mK @ 300K, which is better than TIM but worse than copper. May as well just remove it while "normalizing" the surface (lapping). Also, I have found that, at least on budget AMD CPUs that I have lapped (x2-3600+, x4 635, 7700k), that the thickness of the nickel coating is variant based on the underlying copper surface. Usually the copper is uneven, so to produce a flat surface, they laid on nickel in a correspondingly-uneven fashion.

It isn't a huge deal, but it's good for maybe 1-2C alone.

Yeah we need to run the bath before baking claims like that. We're an order of magnitude off

 

[DrMrLordX]

How thick is the nickel plating on average? 0.1 mm? 0.05 mm? At 0.1mm, we're looking at a minimum of 300W/K through the plating for a FX-83xx.

Doesn't really feel worth spending time sanding it off unless they sloughed a lot of nickel on top of the copper.

Whether or not it is worth the trouble is a question of goals and willingness to spend time and effort on the project. You also have to be willing to take some risks. I got sloppy with my x4 635 years ago and killed it on a re-lap (was trying to bring up fresh copper after using CLU . .. totally unnecessary) by letting metal shavings infest the pin array underneath the chip. At least I didn't kill the board doing that.

The only reason I lapped my 7700k was that I removed the IHS anyway (to replace the underlying TIM). It was safe and easy to lap the IHS once it was free of the CPU.

Today, I would not personally lap a chip with solder underneath unless I knew I was having temperature problems with it and wanted to maximize cooling effectiveness.

Yeah we need to run the bath before baking claims like that. We're an order of magnitude off

I can only speak from personal experience. Using any kind of mathematical analysis of the effect of the nickel coating requires taking into account all variables - the surface characteristics of the nickel coating, whether or not the thickness of the coating is uniform, and so forth. Simply treating it as a uniform surface of area A with thermal conductivity of k and thickness L is not necessarily going to produce results that fall in line with my personal experiences, nor those of others who have lapped chips. Take anecdotal evidence with a grain of salt.

I can say that using CLU on chips that I had no real intention of "cooling to the max", such as my old Sempron 140 and its replacement, an x2 220, produced results that were extremely good . . . about as good as what I got with CLU on my lapped x4 635 (before it died). That even applied when I had the x2 unlocked to a tricore with l3 cache running @ 3.9 GHz. It is a little hard to compare the two since the 635 had one extra core running, but it had no L3 and it was at a lower clockspeed (highest clockspeed I got out of that thing was 3.75 GHz or so).

 

[soccerballtux]

Whether or not it is worth the trouble is a question of goals and willingness to spend time and effort on the project. You also have to be willing to take some risks. I got sloppy with my x4 635 years ago and killed it on a re-lap (was trying to bring up fresh copper after using CLU . .. totally unnecessary) by letting metal shavings infest the pin array underneath the chip. At least I didn't kill the board doing that.

The only reason I lapped my 7700k was that I removed the IHS anyway (to replace the underlying TIM). It was safe and easy to lap the IHS once it was free of the CPU.

Today, I would not personally lap a chip with solder underneath unless I knew I was having temperature problems with it and wanted to maximize cooling effectiveness.



I can only speak from personal experience. Using any kind of mathematical analysis of the effect of the nickel coating requires taking into account all variables - the surface characteristics of the nickel coating, whether or not the thickness of the coating is uniform, and so forth. Simply treating it as a uniform surface of area A with thermal conductivity of k and thickness L is not necessarily going to produce results that fall in line with my personal experiences, nor those of others who have lapped chips. Take anecdotal evidence with a grain of salt.

I can say that using CLU on chips that I had no real intention of "cooling to the max", such as my old Sempron 140 and its replacement, an x2 220, produced results that were extremely good . . . about as good as what I got with CLU on my lapped x4 635 (before it died). That even applied when I had the x2 unlocked to a tricore with l3 cache running @ 3.9 GHz. It is a little hard to compare the two since the 635 had one extra core running, but it had no L3 and it was at a lower clockspeed (highest clockspeed I got out of that thing was 3.75 GHz or so).

I disagree, because orders of magnitude.

 

[BonzaiDuck]

All measurable improvements in C-degrees grow smaller as the thermal wattage increases.

I think I had a C2D 6700 with temperatures under the same load and ambient before and after lapping and removal of the nickel-plate.

How much you could attribute to the perfectly flat IHS surface, and how much was due to removal of the nickel, I can't say. But for the thermal power of that chip, same ambient, same test -- there was a 2C improvement. This -- from lapping the nickel from both HSF and IHS, to perfect flatness.

Do the same with a 5820K, and you won't see much. There, you're starting at 140W, and overclocking would push it higher.

One could have warranty-reticence for lapping the IHS; or second-hand buyers might prefer being able to read the information about the chip stamped in the nickel-plate. But even recently, I've done it for every machine in the house except my C2Q server and my brother's i5-3570K.

A person could argue that it's a lot of trouble for nothing. On the other hand, with the right wet-or-dry sandpapers it might take less than a half hour-- maybe even half that time -- to grind down to copper.

If the goal is to build a system for some mild overclocking, one could use whatever TIM is available, follow all the instruction guides without any such "custom refinements," and avoid the time and trouble. But thermal improvements are additive. If one can have ~4 or 5C for certain thermal power with a ducted cooler and stronger fan, adding 3C for a better TIM and even 1C for lapping-to-copper, that could be a 9C improvement overall.

With water-cooling -- you may or may not want to bother. With heatpipe "air-cooling" -- you may or may not want to bother. But in the latter case, anything approaching a 10C improvement overall is more significant than it would be for a water-cooled system.

 

[MongGrel]

You have to let the AS5 cure for so many hours running before it's at maximum effectiveness.

Also, I'd save the Indigo Xtreme tests for last, in case you screw the pooch and it doesn't flow properly.

With heatpipe mounting pressure, "too much" shouldn't matter: it will ooze out the sides and drool on the lip of the processor cap. [Same with ICD.]

With the ICD, find a rectangular razor blade -- second-best, an old credit card without a fraying laminate. Put a pea-size chunk of ICD in the middle of the IHS and spread outward in four directions. If you have a second razor-blade or a clean screwdriver, you can clean off the razor-blade, drop the excess on the IHS, and keep spreading until you get an overall uniform coat with no tearing. Stuff is like thick pavement cement; it takes a little patience.

On my end, I have the Indigo Xtreme "kit" arriving sometime next week. I am still ruminating about how to proceed so that it heats up properly and flows like it should. This trouble about temperature and flow is enough to make one walk away from a perfectly-good ICD installation and just leave it alone. We shall . . . endeavor . . . to persevere . . .

+1

 

[MongGrel]

NM, Bonzai all over this one all ready.

AS5 is a outdated conductive TIM that needs to cure, IMHO.

 

[BonzaiDuck]

NM, Bonzai all over this one all ready.

AS5 is a outdated conductive TIM that needs to cure, IMHO.

Well, like the brothers who promote their own six-pack wine cooler -- "Thank you for your support." But it all fits into the information about the materials.

Silver ranks enough more in thermal conductivity than copper that I've seen waterblocks made of silver offered when the commodity price was probably lower. Or maybe it never caught on enough. I don't think gold is as good as silver, it could even be worse than copper but I'd have to review the data. Diamond ranks higher by . . . wait-a-minute . . .

http://hyperphysics.phy-astr.gsu.edu/hbase/tables/thrcn.html

There you have it. The only formulations that work better than nano-diamond are liquid-metal products using indium or gallium -- possibly something else similar. Indium and Gallium aren't in the professor's list.

Although there was one exception, perhaps exposed in another thread on "pyrolitic graphite" or something like that. There is a problem with certain design options of PG that they only spread heat in two dimensions. Maybe sooner or later some one will invent a carbon-based or carbon fiber material that can be shaped into heatsink fins.

But -- you see, the best-performing products are either Carbon, or the formulations using metals with low melting points, high conductivity, or an easy tendency to alloy with other metals. IIRC, Gallium will invade the lattice structure of Aluminum, and make it brittle. So we can guess something about the components of either Indigo Xtreme or Coollaboratory Liquid Pro/Ultra: IX is recommended as safe for aluminum, while CLU is verboten. . . . "not advised."

 

[DrMrLordX]

Gallium itself doesn't do much of anything to aluminum. It's aluminum oxide that is affected by gallium. All aluminum has a thin oxide layer (alumina) that protects the underlying metal from exposure to futher sources of oxygen. Gallium compromises that oxide layer, allowing oxygen to reach the metal freely. As a consequence, the entire body of aluminum will turn to alumina, provided continued exposure to gallium.

Back on topic: soccerballtux, what's the current verdict on your HSF testing? Or are you still tied up with testing different TIM applications?

 

[soccerballtux]

Gallium itself doesn't do much of anything to aluminum. It's aluminum oxide that is affected by gallium. All aluminum has a thin oxide layer (alumina) that protects the underlying metal from exposure to futher sources of oxygen. Gallium compromises that oxide layer, allowing oxygen to reach the metal freely. As a consequence, the entire body of aluminum will turn to alumina, provided continued exposure to gallium.

Back on topic: soccerballtux, what's the current verdict on your HSF testing? Or are you still tied up with testing different TIM applications?

I'm actually not sure, and re-installing the TR120 doesn't really interest me. There were three variables I wasn't recording, that have so far thrown off my results:

• 10 year old AS5 potentially being 8C worse than new AS5
• ambient temperature
• [noob mistake] I switched from 'In-Place large FFTs' to 'Custom' when I noticed that 480kb FFT size generated the maximum amount of heat. When I made the switch, the earliest of, and I'm not certain which [though I have an idea], the Custom runs I made I did not manually check the 'Run FFTs in-place' box. This initially made the new vs. old AS5 look 10-11C better, until I realized my mistake.

as good as our 4.5kwh HVAC unit is, leaving it on is not the same as coming home, turning it on, waiting 30 minutes, and testing.

Anyways, the highest confidence cooling performance (excepting ambient temp) I have out of the 'Thermalright Ultra-120 Non-Extreme, Lapped' testing stabilized at 59C. Highest confidence I have out of the ACX, on a warmer day, was 58C-- across all TIMs. Not very exciting, huh?!

For the good of everyone, perhaps I -will- go back and re-install now that I've revamped my case cooling (extra fans, better placement, and more balanced in vs. out CFM) and added a little something that I expect will improve test consistency. I need to think of a good place to install an ambient-temp thermocouple first. Check back in a month I should some ideas by then.

 

[DrMrLordX]

Huh, I figured you'd get more than 1 C out of the ACX. Maybe the TRU120 was better than expected. I'll keep an eye out for future results, thanks!

 

[soccerballtux]

Huh, I figured you'd get more than 1 C out of the ACX. Maybe the TRU120 was better than expected. I'll keep an eye out for future results, thanks!

I've got to have made a mistake. More reason for me to retest

 

[BonzaiDuck]

I was just browsing the recent Anandtech 9-cooler roundup.

I think I also looked at some reviews from Tech-Power-Up and Tweaktown.

One of the reviews seemed to distinguish differences in comparable performance at lower load temperature values and those at higher loads. But in the tests I ran, mentioned again at the end of this post, the expected and verified differences were approximately the same.

Also, there's one thing I can't remember between the original TR Ultra 120 and the TRUE model: At least one of those coolers had a base with a convex bow in the mating surface. SVC or Silicon-Valley Compu-Cycle was actually offering a lapped-to-bare-copper shipment for which they charged extra. But I can't see how these factors would do anything but bias an outcome and masquerade it to look as we would expect -- contrary to your initial results.

Personally, I lapped mine -- both of them -- flat. But I think you're test outcomes don't reflect the performance of the ACX. Given the long exchange of posters about application of TIM, I might imagine that could have something to do with it, but I don't know. I only know what I personally verified about two identical systems loaded to ~135W with the same stress-test, pitting the NH-D14 against the ACX, each using ICD TIM applied in the same amount and in the same way.

Additional thought: This review seems to indicate that different processors, with different TDPs and perhaps IHS' of different sizes, will have an impact on rank-ordering of coolers:

http://www.bit-tech.net/hardware/cooling/2010/01/05/noctua-nh-d14-cpu-cooler-review/3

I always trusted Bit-Tech reviews, and so I'm not sure what to make of this. But like I said, the results would vary over a range of thermal wattage, and one should readily assume they might vary over a range of contact-area in square mm.

 

[soccerballtux]

I lapped mine.

and I'm really certain the AS5 is significantly to blame. I must have reseated that cooler 5 times, lapped, etc trying to get better temps. Maybe AS5 changed the formulation?

 

[BonzaiDuck]

I lapped mine.

and I'm really certain the AS5 is significantly to blame. I must have reseated that cooler 5 times, lapped, etc trying to get better temps. Maybe AS5 changed the formulation?

What is the thermal power of your processor? Or what is the "package power" reported in CPUID HWMonitor in watts? What stress-test did you use?

Also, if it's an AMD chip, what is the lithography spec in nm or nanometers?

It could be the AS5, if that was a basis of your test comparison. I wouldn't know what happens to that stuff stored at room ambient for a long time. I do know that nothing like that would happen with the ICD. Even if the oils dried out in the tube, you could refresh it with another oil or silicone based TIM. . . . If you could get the remaining sludge out of the tube, anyway. The diamond is inert, so I think you could mix it with anything except the liquid metal or "metal-pad" type of TIM.

But I've kept a tube of that stuff for several years, and it still comes out of the tube with no change.

 

[crashtech]

Man, I buy tubes of TIM 3 at a time. If I wanted to stick with the same one for several years, it'd have to be caulking gun size!

 

[soccerballtux]

What is the thermal power of your processor? Or what is the "package power" reported in CPUID HWMonitor in watts? What stress-test did you use?

Also, if it's an AMD chip, what is the lithography spec in nm or nanometers?

It could be the AS5, if that was a basis of your test comparison. I wouldn't know what happens to that stuff stored at room ambient for a long time. I do know that nothing like that would happen with the ICD. Even if the oils dried out in the tube, you could refresh it with another oil or silicone based TIM. . . . If you could get the remaining sludge out of the tube, anyway. The diamond is inert, so I think you could mix it with anything except the liquid metal or "metal-pad" type of TIM.

But I've kept a tube of that stuff for several years, and it still comes out of the tube with no change.

the old AS5 was tested across 5 CPUs--

AMD Athlon 2500+ Barton XP
AMD Sempron64 3100+ at 2.3ghz (inconsequential, wasn't even paying attention to these)
an e2180 clocked at 3.4ghz and 1.48v,
an X3 720BE->X4 unlocked, 3.5ghz/2.6ghz 1.47v
a Phenom 2 X4 965BE at 4ghz/2.6ghz 1.44v
FX8310 at 4.3ghz, and it couldn't handle 4.6ghz

this last one was the first with a problem. At 4.3ghz I had no trouble, but 4.6? instantly 66C at 1.443v on In-Place Large FFTs

 

[BonzaiDuck]

the old AS5 was tested across 5 CPUs--

AMD Athlon 2500+ Barton XP
AMD Sempron64 3100+ at 2.3ghz (inconsequential, wasn't even paying attention to these)
an e2180 clocked at 3.4ghz and 1.48v,
an X3 720BE->X4 unlocked, 3.5ghz/2.6ghz 1.47v
a Phenom 2 X4 965BE at 4ghz/2.6ghz 1.44v
FX8310 at 4.3ghz, and it couldn't handle 4.6ghz

this last one was the first with a problem. At 4.3ghz I had no trouble, but 4.6? instantly 66C at 1.443v on In-Place Large FFTs

For whichever of these processors was your intentioned objective for final installation of the EVGA heatsink, or whichever one is most relevant to the testing and your daily needs and uses -- What is the "package power" reported in HWMonitor during some specified stress-test? Best use the sFFT test to get maximum CPU temperatures and wattage, but either test would suffice at this point.

 

[soccerballtux]

For whichever of these processors was your intentioned objective for final installation of the EVGA heatsink, or whichever one is most relevant to the testing and your daily needs and uses -- What is the "package power" reported in HWMonitor during some specified stress-test? Best use the sFFT test to get maximum CPU temperatures and wattage, but either test would suffice at this point.

95w. not accurate

 

[BonzaiDuck]

95w. not accurate

95W at 4.3 to 4.6 Ghz? If that's the thermal wattage of your overclock in load conditions, or even if it's as high as 140W, you should be getting measureable positive results with the ACX.

I thought I'd read where the more recent AMD processors had a 32nm lithography. If that's also true, there's nothing about your hardware that should pre-empt good results with the ACX and any one of the TIMs, as long as the TIM hasn't degraded in its tube over the time you had it. Only the AS5 fit that description, if I understood what you said before.

 

[soccerballtux]

off topic: it really is better [as Jesus said] to live for serving rather than being served. If you're living for yourself, what do you do when you run out of things that interest you? How much better to live for the betterment of others, then you'll always be motivated.

 

[soccerballtux]

95W at 4.3 to 4.6 Ghz? If that's the thermal wattage of your overclock in load conditions, or even if it's as high as 140W, you should be getting measureable positive results with the ACX.

I thought I'd read where the more recent AMD processors had a 32nm lithography. If that's also true, there's nothing about your hardware that should pre-empt good results with the ACX and any one of the TIMs, as long as the TIM hasn't degraded in its tube over the time you had it. Only the AS5 fit that description, if I understood what you said before.

it's inaccurate. it never goes above, honestly about 70 watts

 

[BonzaiDuck]

the old AS5 was tested across 5 CPUs--

AMD Athlon 2500+ Barton XP
AMD Sempron64 3100+ at 2.3ghz (inconsequential, wasn't even paying attention to these)
an e2180 clocked at 3.4ghz and 1.48v,
an X3 720BE->X4 unlocked, 3.5ghz/2.6ghz 1.47v
a Phenom 2 X4 965BE at 4ghz/2.6ghz 1.44v
FX8310 at 4.3ghz, and it couldn't handle 4.6ghz

this last one was the first with a problem. At 4.3ghz I had no trouble, but 4.6? instantly 66C at 1.443v on In-Place Large FFTs

You know -- I'm looking at this with a little more discernment.

I'm not an AMD builder -- not since I used their replacement for the Intel 8088 processor. Certainly, the power statistic at some particular load test is more important than voltage. My sig-rig and its twin are set to give me a highest-measurable turbo VCORE of ~1.38V.

So I won't concern my thoughts here about what your voltage represents in terms of some "safe range" and an "operable range" above it. I wouldn't know.

But 66C with LFFT test . . . wait-a-minute . . . OK -- with GRID2 in a menu-mode still running, here's a short run of Prime95 LFFT "with AVX" for my cool-running 2700K system:

Room ambient is 78F. This is done with the ACX [aka SuperClock] cooler EVGA 100-FS-C201-KR, and ICD applied improperly, according to the Innovation Cooling web-site and "Application Guide."

CPU pusher fan: Noctua iPPC 3000
Exhaust [CPU-puller] fan: Gentle Typhoon AP-30
Intake fan [side and front]: 2x BitFenix Spectre-Pro 200mm rated 144 CFM

2x 970 GTX gfx cards warm things a bit, so the PCI plates above each card has been removed for an ongoing project and passively directed airflow in the meantime.

The AP-30 is ducted to the rear of the ACX with a ThermalRight blue accordion duct.

I think the peak package temperature showed up at 74C. The Average-of-cores maximum is ~71C, but the prevailing values fluctuate below these extremes over a 3C range. Note the package power maximum pegged at 130W.

So, if you're reporting your peak power accurately, you should be able to do better, but "better" may involve other things besides the cooler specifically and the TIM. In replacing my RAM the other day, I removed the Noctua pusher fan and these thermal statistics increased by about 2C under the dual-fan Fan-Xpert profile. I would have to shift my CPU_FAN "fan-curve" to the left in the graph to regain the 2C.

And after my long sig-twin explanatory babble, the point is simple. Unless your 66C is outside recommended thermal limits for that AMD processor, you're fine, temperature-wise. If the corresponding 1.4+V VCORE is not far above some recommended limit for the processor -- then fine. If it isn't, you choose your own priorities.

But all the temperature reductions in the world with strategies not exotic like phase-change aren't likely to get you extra Mhz unless you raise the voltage. Some other AMD aficionado can pipe in here, but I'm only looking at your thermal wattage, the cooler -- even the paste, and the temperatures compared to what one might like to achieve with my Intel processors.

 

[crashtech]

Seems to me that temp data for AMD CPUs is collected from outside the die somewhere (not sure exactly where), while Intel temps are sampled on-die. This accounts for the discrepancies in what are considered usable temps. I am willing to be corrected on this, this is "afaik".