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

[BonzaiDuck]

Following my "first principle of wisdom" to admit total ignorance: I simply wouldn't know anything about modern-day AMD processors.

In the context of what AMD users think are "good" and "not good" temperatures for those cores, I'd have to look at the spec data-sheet for the processor.

But what AMD users think is good would have a basis of data for following the same sensors, regardless of where they're located.

Maybe what soccerballtux reports as CPU temperatures picked up by the same monitoring program aren't entirely comparable to Intel, but I wouldn't know. Not at all.

The data-sheet from CPU_World on the last processor soccerball listed shows a default thermal design power of 95W, a stock turbo-clock of 4,300 Mhz @ ~1.41V:

http://www.cpu-world.com/CPUs/Bulldozer/AMD-FX-Series%20FX-8310.html

So certainly, any voltage worries for (what'd-he-say?) ~1.44V may be unnecessary. But overclocking with a slightly higher recorded voltage would certainly show a thermal power higher than 95W under most stress tests, which is also the TDP spec on my Sandy Bridge processors.

Maybe HWMonitor doesn't report the same things for the AMD processors as for the Intel. Again -- total ignorance -- I wouldn't know. But the program has been extant for a long, long time, and I don't think it was ever "made for" an Intel chip exclusively.

One thing for sure, besides the matching TDPs: Soccerball's AMD chip and my 2700K are both 32nm processors, in terms of the lithography.

 

[soccerballtux]

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".

it's not, they both measure the I-V curve of an on-die diode to read temp

either AMDs can go higher than 70C safely, or the Intel process change to HK and finfet brought other changes that enable higher silicon temperature as well.

anyone want to fund 80C AMD testing? to find out if we can safely go higher than 70C? another 8310 would run $100, or I can get an 8370e for $120 at microcenter (which would be an upgrade and foot the difference)

 

[soccerballtux]

Following my "first principle of wisdom" to admit total ignorance: I simply wouldn't know anything about modern-day AMD processors.

In the context of what AMD users think are "good" and "not good" temperatures for those cores, I'd have to look at the spec data-sheet for the processor.

But what AMD users think is good would have a basis of data for following the same sensors, regardless of where they're located.

Maybe what soccerballtux reports as CPU temperatures picked up by the same monitoring program aren't entirely comparable to Intel, but I wouldn't know. Not at all.

The data-sheet from CPU_World on the last processor soccerball listed shows a default thermal design power of 95W, a stock turbo-clock of 4,300 Mhz @ ~1.41V:

http://www.cpu-world.com/CPUs/Bulldozer/AMD-FX-Series%20FX-8310.html

So certainly, any voltage worries for (what'd-he-say?) ~1.44V may be unnecessary. But overclocking with a slightly higher recorded voltage would certainly show a thermal power higher than 95W under most stress tests, which is also the TDP spec on my Sandy Bridge processors.

Maybe HWMonitor doesn't report the same things for the AMD processors as for the Intel. Again -- total ignorance -- I wouldn't know. But the program has been extant for a long, long time, and I don't think it was ever "made for" an Intel chip exclusively.

One thing for sure, besides the matching TDPs: Soccerball's AMD chip and my 2700K are both 32nm processors, in terms of the lithography.

HWMonitor wattage report is not accurage.

Also, at that TDP you don't get to turbo all 8 cores at 4.3ghz all day long.

at idle my system uses 150W, CPU loaded at 4.6ghz 430W. at 4.3ghz 300w.

so, the thing about the old tube of AS5 is that it was acceptable until it really started having to perform.

 

[soccerballtux]

might have even more reason to re-test the ACX. Got off the phone with Arctic Silver and to even hope of completely lifting the thermal compound you have to let a rag soaked with alcohol sit for a good 10 minutes-- he gave an example that you can clean and keep cleaning with a new rag and still see residue being pulled off even after multiple cleans. He did compliment my 90% IPA, though

I can confirm that this happened with up to at least 5 cotton swabs on...I don't remember which thermal compound...think it was AS5. I cleaned...and cleaned again...and kept going, to completely lift all the TIM-- you couldn't tell by looking at the HS block surface, -that- looked clean: it was only with the cotton swab that you could tell.

I think this would be a good reason to lap the ACX surface-- I didn't have as much difficulty completely removing TIM from the lapped TR120 as I did the un-lapped ACX. And, if it ensures clean surface/complete thermal compound removal...

they couldn't comment on whether lapping would aid TIM complete removal, though. This has me wondering if my initial tests (where NTH1 was 1-2C better than AS5) were accurate, and how that evened out on subsequent testing due to TIM soaking into the HS block. The NTH1 was consistently 1-2C better on the lapped TR120 surface hm...looks like I'll be lapping the ACX.

 

[BonzaiDuck]

HWMonitor wattage report is not accurage.

Also, at that TDP you don't get to turbo all 8 cores at 4.3ghz all day long.

at idle my system uses 150W, CPU loaded at 4.6ghz 430W. at 4.3ghz 300w.

so, the thing about the old tube of AS5 is that it was acceptable until it really started having to perform.

It seems accurage to me! I get the same approximate readings proportionate to temperature with Intel XTU. Or -- that's my recent take on it. I only installed XTU a few months ago.

So you're saying that your CPU's default turbo setting is [NOT] "for all cores." ? That seems similar to the Intel chips, but it can be changed as a matter of minor over-clocking.

Try the XTU for monitoring wattage. I think you could use XTU exclusively to monitor, and choose a stress test known thermally more robust than XTU.

Oh. Wait a minute. I'm not trying to be funny or sardonic. Being ignorant again, I'd suspect that XTU wouldn't help you with an AMD processor. Speed Fan perhaps? AIDA-64? Latter is probably most reliable, but it costs a Jackson and a Lincoln.

But HWMonitor seems spot-on with those I've used in the last four years. I don't use Speed Fan.

Maybe instead you're referring to crashtech's speculation about sensor-location for the AMD CPUs?

[addendum:] So that's an octo-core processor? And I can see that it's spec'd at 95W TDP?

So I'm wondering how power-consumption scales with clock and voltage on that chip, from the 4.3 spec to the 4.6 OC.

Only thing I can compare it with is my take on the Haswell-E. But that's a 22nm processor -- not 32. Yours may be more comparable to the 3960X on that score. Those may clock better; the Haswell-E's seem stuck around 4.5 to 4.6 Ghz, and there's a significant lottery, with a noticeable S-O-L component. . .

And . . . you posted twice before this. Second one:

I'd say if the TIM degrades over time, the processor had its share of it, and you're testing a second TIM -- yes -- clean the bleeping processor cap.

I still haven't completely decided to apply one of the IX-XS metal-pads to the ACX. I am working out in my mind how I will control the temperature of the stock, loaded processor to re-flow the IX Indium-based material. And cleaning be-damned -- if I go that route over the next several weeks, I'll lap the IHS again.

I've got an idea to use a plastic box shipped as container to a ThermalRight accordion duct, if I can fit it over the ACX. I'm thinking that whatever I use, I'll make some sort of box like that, with two telescoping parts, so I can vary the amount of air that will pass through the cooler from a single intake fan and the exhaust turned off. Then -- run some tests with the existing ICD installation.

Whichever way I go with the IX, the skinny of wisdom I've picked up so far suggests turning a tower computer on its side to make the motherboard, IHS and HSF-base horizontal. I've seen a few folks whining that flecks of old IX would get loose on their motherboard and cause the dreaded shorts. And I think that after the re-flow occurs, it would never flow again unless something went terribly wrong with your cooling system and its redundancy.

 

[soccerballtux]

...

 

[soccerballtux]

being larger than the TR-120 seems to be better for being on the edge of the cooler's TDP capacity: it gives you another 2-3 seconds of room to work with while waiting for the fans to spin up. At 4.6ghz and 1.44v, this is nothing to dismiss, as on the TR-120 I required a rather aggressive fan profile. On this ACX, however, I can have the fan set to 25% until the CPU hits 65C, before it spins up.

For this reason alone I think this may have been a worthy investment. Considering CPU coolers are so much more expensive than they used to be ($45 for TR-120 8 years ago vs $85 for the good Noctua offerings), the lower pricepoint offered by the EVGA ACX was welcome.

I wouldn't feel so passionately about this improvement if Speedfan simply used a more gradual fan spin up and spin down routine-- I am reminded of two experiences:

• our Mercedes bus rental when our study abroad university class toured Europe had a 'weather' setting on the HVAC for the bus. I don't know what the name was, but it had a stormy looking cloud and rain. The vents were controlled by servo motor, and would randomly change direction-- like wind blowing-- and the blower speed would, too-- also just like wind blowing. You knew you were still inside a bus, but the gradual variation in both vent direction and air volume was wonderfully unique, and the variations in white noise created added an extra something to the experience.
• a friend's PC had a CPU fan that gradually changed speeds. Instead of the immediate ramp up on load (distracting), it would change about 20%/second.

Why were these such a big deal? Because they sounded like wind blowing through trees, not a fan on the CPU heatsink or a car's HVAC unit!

I think I'm actually going to look into adding this feature in to Speedfan. I submitted this feature request but he gets so many of those I doubt anything will happen...

I should have saved this little writeup tidbit for the end.

I think I want to move back to the TR-120 for a bit to get a better feel for its performance and perform more thorough testing.

Now that I've killed the ACX base with the ICD7, I think I will lap the ACX in the downtime as the TR-120 is reinstalled and retested.

 

[DrMrLordX]

I'm not sure what is your testing plan from this point forward, but I will note that if you isolate one particular variation between testing (example: same TIM, different HSF), you may get more accurate results more easily.

Also, what did the ICD7 do to your ACX base? Did it scratch it up?

 

[soccerballtux]

I'm not sure what is your testing plan from this point forward, but I will note that if you isolate one particular variation between testing (example: same TIM, different HSF), you may get more accurate results more easily.

Also, what did the ICD7 do to your ACX base? Did it scratch it up?

haven't removed it yet. Going to cut the fan grills out first, do more testing, then switch to the TR-120 again.

I have been isolating variables, but not enough. Temperature around the case seems to vary a lot more than I thought.

 

[ClockHound]

haven't removed it yet. Going to cut the fan grills out first, do more testing, then switch to the TR-120 again.

I have been isolating variables, but not enough. Temperature around the case seems to vary a lot more than I thought.

What about the temps in the case, specifically, the cooler intake temperature?
That's the vital one to track with the ambient.

 

[soccerballtux]

What about the temps in the case, specifically, the cooler intake temperature?
That's the vital one to track with the ambient.

I don't have that many thermocouple readers!!! and for the one I do have, I have to transcribe it by hand!

also, which intake? Side? biggest/140mm fan.
front? 120mm, sits in front of harddrive.
bottom? 140mm, but gets nothing, for now at least, until I cut off the grill covers.

 

[ClockHound]

I don't have that many thermocouple readers!!! and for the one I do have, I have to transcribe it by hand!

also, which intake? Side? biggest/140mm fan.
front? 120mm, sits in front of harddrive.
bottom? 140mm, but gets nothing, for now at least, until I cut off the grill covers.

CPU Cooler intake - as close to the fan without ingestion - an inch or two.

A cheap indoor/outdoor thermometer with a remote probe is all that's needed. Here's one from Amazon for $8.12

Just run the probe wire thru an open pci slot and twist tie the probe in front of the cpu intake fan. Place the 'indoor' display unit outside the case at a convenient location not in the intake/exhaust fan paths.

Then you can test the gradient between ambient and cooler intake temp. If your case airflow is awesome, expect to see a 2-3 degree higher temp than ambient at the cooler intake, at idle and load. If it's above 6-8 degrees, then the case intake airflow still has room for optimizing. If it's over 10 degrees hotter at the cooler intake than ambient, then some serious case modding tools might be required - sawsall level. ;-)

If the temp gradient increases from idle to full load, then the case is recycling hot air and you get to play with the intake/exhaust balance. And front-to-back flow. A case with a top vent (intake or exhaust) can be tricky to optimize - often the cpu cooler fan will pull in heated air from a top exhaust. Assuming the cpu cooler is mounted front-to-back and not bottom-to-top.

 

[BonzaiDuck]

Well, Sah! I came back to this thread after turning up some rather stunning Tech-Power-Up and Overclock.net threads on IC Diamond. Here are the stale links -- the first of them drawing hasty conclusions from the second:

http://www.overclock.net/t/1411528/avoid-ic-diamond-thermal-paste

http://www.techpowerup.com/forums/threads/techpowerup-official-ic-diamond-test.170121/

I read through the entirety of the Tech-Power-Up thread -- all 34 pages through the years 2012 and 2013. Personally, I tend to side with the Innovation Cooling rep (really, its CEO), but he handled his customer-relations challenges very badly.

I found the two examples showing visible deep pit in either the IHS or the a GPU die very suspicious, since the shape and depth of the pits seems almost identical. I can say personally I've been using ICD for 8 years now with as many as 12 PCs and more than one application for any one of them. Nothing such as described by the complaining Tech-Power-Up members ever happened. If ICD destroys the embossed print on an unlapped processor, it is more likely the result of twisting the cooler to remove it.

In my case and that of others, for "scratching" or "polishing" the IHS, we don't much care since we lap our processor caps to bare copper anyway.

Getting down to brass tacks, we're talking about thermal improvements across the entire range of TIMs (including IX on one end and something worse than AS5 on the other) of maybe 5C degrees across that range corresponding to a thermal wattage range between ~110W and ~140W.

If you insist on heatpipe-cooling over water-cooling for an ambitiously overclocked system, and I have said this before, you only may want to focus attention on every last grain of rice in additive improvements.

On the other hand, if you want to worry about your Intel warranty or resale value, you can wait three years when the warranty expires, or abjure lapping the processor. If you think that ICD is going to wipe the embossed print from your IHS and that matters to you, don't use ICD.

But the IC rep in the forum thread is correct about the diamond particles: they are only marginally harder abrasive material than the Aluminum Oxide particles ingredient to many TIMs.

I almost think reading through 34 pages was a waste of my time. The dynamics of human interaction in that thread seemed reminiscent of group political behavior in a volatile environment, providing a bandwagon for people to jump on.

Seldom ever saw anything like that of the same degree in tech-forum threads.

Astounding.

In a few instances, I was impressed with some reports for applying the ICD in direct dGPU die situations. Some folks noted improvements closer to 8C than to 2C. But I cannot find closure on the possibility that application of the material might damage a GPU die. Nevertheless, I've used it once for precisely that purpose in a direct-die application for a GTS 8800 NV card, and nothing disastrous ever occurred.

I just thought I'd share this for soccerballtux's thread participants. Like the President said the other day, "it would seem ridiculous if it weren't so sad.":biggrin: