Overclocking Megathread 3.11 for Workgroups

The Something Awful Forums > Discussion > Serious Hardware/Software Crap > Overclocking Megathread 3.11 for Workgroups

«‹›2 »

Alereon: Feb 6, 2004; Dehumanize yourself and face to Trumpshed; College Slice

Splash Damage posted:

I don't understand a word you've posted. How would I go about overclocking an ATi Radeon 4800HD just a little bit, to get consistent 30fps in BF3?

Use a canned air duster to dust out the videocard heatsink and fan and monitor the temperatures with GPU-Z. As long as temperatures are okay (not going above 100C), bump the GPU clock speed up a bit (5-10Mhz) then test to see if it worked. If you don't get any crashes or weird visual artifacts (white pixels, miscolored objects, corrupted/spikey geometry). Repeat until you find the highest speed that doesn't give you errors or heat the card up too bad. If your card supports fan speed control, you can also turn the fan up a bit to give you better cooling at the cost of noise. You're probably not going to milk a whole lot out of a Radeon 4800-series card at this point though, especially if you only have 512MB of video memory.

# ¿ Feb 11, 2012 09:47

Adbot: ADBOT LOVES YOU

# ¿ Apr 26, 2024 04:46

Alereon: Feb 6, 2004; Dehumanize yourself and face to Trumpshed; College Slice

Josh Lyman posted:

In any case, I came here for this question. Cooler Master Hyper 212 Plus is $20 AR at Newegg. My current CPU temps are 35C idle with 30C ambient and max out at 86C under Prime95, but realistically the most CPU-intensive stuff I do is play Diablo 3 and Starcraft 2 and Civ 5 (assuming H.264 video is being decoded by my 560 Ti). Is it worth the $20? Before you suggest I spend money on case fans, all 6 spots are occupied, albeit by lower dB/CFM 120mm fans.

If you're not planning to overclock or otherwise run in a thermally stressful environment there is no reason to upgrade beyond the stock cooler. If you're using the stock TIM, replacing it with thermal paste (no need to buy Arctic Silver or other stuff) will improve temperatures noticeably and for free. You should also probably remove some of your case fans, six fans is pretty ridiculous and you're generating a lot of extra noise for not much benefit.

If you ARE wanting to overclock, then obviously you should upgrade the cooler, preferably to something even more powerful than a Hyper 212, unless you just can't stomach the extra dollars. A Thermalright HR-02 Macho isn't really that much more expensive, and it's an entirely better class of performance and quiet.

# ¿ Jun 17, 2012 16:26

Alereon: Feb 6, 2004; Dehumanize yourself and face to Trumpshed; College Slice

You can use electrically conductive thermal paste on exposed CPU dies without issue, we did it for years before they started putting the IHS on. You just need to make sure it doesn't touch any of the gold bits, and you'd have to be using vastly too much for that to happen. That said, in general terms it's smarter to just use non-conductive paste since there's no risk at all no matter how badly you mess up.

# ¿ Jun 18, 2012 03:00

Alereon: Feb 6, 2004; Dehumanize yourself and face to Trumpshed; College Slice

Josh Lyman posted:

Although I didn't remove my heatspreader, I was confused about the concern. The last time I owned a desktop was 2004 and that machine had an Athlon XP Thunderbird whose exposed core was notorious was being damaged by fat fingered system builders.

Tbirdchat: I bought one of those copper shims that was the exact height of the CPU core to prevent that from happening. Because I put a Thermalright SLK-900A on top :cool:

# ¿ Jun 18, 2012 03:06

Alereon: Feb 6, 2004; Dehumanize yourself and face to Trumpshed; College Slice

It's the core temperatures from RealTemp that matter. Other programs do read different temperatures, which is why it's important to use RealTemp.

# ¿ Aug 12, 2012 19:44

Alereon: Feb 6, 2004; Dehumanize yourself and face to Trumpshed; College Slice

Usually one core is hotter than the rest, that's due to the physical layout of the electronics that consume power on the die. Definitely do run with the lowest voltage you can, but if the temperatures are safe and it's stable I wouldn't worry TOO much.

# ¿ Aug 12, 2012 20:24

Alereon: Feb 6, 2004; Dehumanize yourself and face to Trumpshed; College Slice

Fuzz1111 posted:

Is it normal that it dials in extra voltage itself as I increase multiplier? I have VCore on normal and DVID on +0.00 and all the LCC stuff on default and here's what I'm getting with multiplier of 42:

Warning: Be absolutely sure LoadLine Calibration (LLC) is disabled before attempting overclocking. With LLC on the motherboard will happily send large enough voltage spikes to the CPU to crash it. This is a problem at stock settings, overclocking and increasing the voltage setting both magnify the issue to the point where the CPU could be permanently damaged.

You also can't perform reliable voltage monitoring on Gigabyte motherboards without special software as the board intercepts the readings and replaces them with healthy, normal values for the voltage you set (to disguise how terrible the power quality is). I'm not sure what programs can see through the Gigabyte spoofing, a bit of Googling should find you a tool that will work.

# ¿ Aug 14, 2012 20:11

Alereon: Feb 6, 2004; Dehumanize yourself and face to Trumpshed; College Slice

It's critical to get decent overclocking without jacking up the voltage, but the problem is that means you're sending huge voltage spikes to the CPU every time it exits a high load condition. I'd rather have a high but known voltage most of the time (or limited overclocking because the board can't supply a safe voltage in a stable manner) than have a more reasonable voltage or better overclocking but with voltage spikes to an unknown, higher level.

# ¿ Aug 14, 2012 20:31

Alereon: Feb 6, 2004; Dehumanize yourself and face to Trumpshed; College Slice

Xbitlabs has posted a round-up of 135-150mm fans (practically, a bunch of 140mm plus the Thermalright TY-150). Predictably, the winners were the Thermalright TY-150 and the Corsair AF140 Quiet Edition. The TY-150 runs about $20 shipped and the AF140 is about $19. It looks like the AF140 is more expensive in most stores, so the TF-150 will probably be the better deal in most situations where it will fit.

# ¿ Oct 13, 2012 23:22

Alereon: Feb 6, 2004; Dehumanize yourself and face to Trumpshed; College Slice

Anandtech's Podcast 13 has some interesting Haswell (Core i5/i7 4000-series) overclocking-related tidbits. First, Intel is going back to soldering the heat spreader to the die as on Sandy Bridge, or at least will do something that improves thermal conductivity over Ivy Bridge's thermal paste arrangement. Even more importantly, some functions of the VRMs have been moved from the motherboard to the CPU. Instead of the motherboard supplying a number of different voltage rails that powered the various planes on the CPU, the board now supplies a single fixed voltage and the VRMs on the CPU handle the details. This is expected to be a boon for overclocking, as we now have better voltage regulation than the best overclocking motherboards, but built-into the CPU. There's still a need for good, responsive VRMs on the motherboard to convert from 12V down to package voltage, but that's much less complex and expensive than the job they currently do.

All in all, it looks like if you have a high-end cooler you'll be able to do some amazing things with Haswell. I'm hoping for over 5Ghz on air with a nice per-clock boost. I think that'll finally justify an upgrade

# ¿ Jan 7, 2013 00:36

Alereon: Feb 6, 2004; Dehumanize yourself and face to Trumpshed; College Slice

One thing to keep in mind is that this is heat that still needed to be dissipated from the CPU socket area, you previously only had tiny passive heatsinks to do it with (though it remains an open question how the VRM components will be bonded to the heatspreader). However, this heat will also be off-die, a major overclocking limiter for Ivy Bridge was that the die shrank in area more than TDP decreased. When combined with the higher thermal resistance of the unsoldered heatspreader, that made Ivy Bridge hard to cool. Haswell presumably has a larger die, is back to a soldered connection, and it seems like most of the added heat will probably be around the outer edge of the heatspreader. Once you've got heat out of the die any overclocker's heatsink can do a great job of dissipating it.

It seems that voltage is fully unlocked with far more tweakability than ever, which is awesome. The IAMA Intel guy at Reddit said that Haswell would be great for overclockers and that the reason was in the Anandtech article, and I think it's the voltage regulation bit here. If I'm understanding it correctly, I think we will get better quality power delivery from this than we did from the most expensive motherboards, meaning we can push CPUs even further.

# ¿ Jan 7, 2013 02:42

Alereon: Feb 6, 2004; Dehumanize yourself and face to Trumpshed; College Slice

I would caution that the voltage spikes caused by LLC are too short to appear in your monitoring logs, so you'll never see how high the voltage is actually getting. Also, I'll just point out that some of you guys are talking about voltage levels similar to what I used on a 45nm Core 2 Quad, I'm not sure how sane it is to put that on a 32nm or much less 22nm processor, though maybe I'm just conservative. Do be careful with auto settings, they aren't always smart, and sometimes use more voltage than they should.

# ¿ Jan 17, 2013 09:31

Alereon: Feb 6, 2004; Dehumanize yourself and face to Trumpshed; College Slice

Regarding temperature monitoring software: traditionally CoreTemp was for AMD CPUs only, RealTemp for Intel. This is because CoreTemp doesn't consistently have correct TjMax values for Intel CPUs hard-coded, resulting in incorrect temperature readings. Note that RealTemp GT is an old version for Gulftown (i7 980X) CPUs only, the current version for Ivy Bridge is RealTemp TI Edition. This should no longer matter on current-gen CPUs as the TjMax value is now stored in a register on the processor and can be read by the tool, but I think it's still a good idea to use the correct program, otherwise there's the potential for forgetting about CoreTemp's limitations and using it on a CPU it doesn't support correctly.

# ¿ Jan 19, 2013 21:04

Alereon: Feb 6, 2004; Dehumanize yourself and face to Trumpshed; College Slice

Guni posted:

Is GPUTool a decent tool to test my GPU overclock? I know when I pushed my GPU hilariously high it made my system BSOD..So I'm sure it must be doing SOMETHING. I know playing games is a good indicator as well and obviously I'll do this too, but is GPUTool a good tool for general testing?

No, it's over 3 years old, unmaintained, and not very useful for testing. It's basically a rip-off of FurMark, which is good for stressing your power supply and cooling, but doesn't stress a GPU overclock much because the card is throttled to keep it below the TDP cap. I found that after a point overclocking the card just didn't change the actual clockspeeds in FurMark.

# ¿ Feb 10, 2013 19:21

Alereon: Feb 6, 2004; Dehumanize yourself and face to Trumpshed; College Slice

It sounds like it's just reading a disconnected sensor, if RealTemp says your temperatures are fine.

# ¿ Feb 14, 2013 03:49

Alereon: Feb 6, 2004; Dehumanize yourself and face to Trumpshed; College Slice

Dogen posted:

Seeing that sexy looking new closed loop thing from Swiftech (H220) has me thinking about looking at water cooling for the first time in years, since it seems like you get the performance of a custom solution, the ability to open it up if one were to say acquire a 780 with waterblock down the road, and how the pump(!) and fans have PWM for noise control.

I've also been looking at the Corsair H110, their newest and highest-performing water cooler. I'm still not sure if any of the dual-fan water coolers can beat air at the same noise level, I have to think that if they could there would be a comparison showing that somewhere. I know the Swiftech H220 can beat high-end air, but it is a giant triple-fan setup.

# ¿ Feb 17, 2013 18:22

Alereon: Feb 6, 2004; Dehumanize yourself and face to Trumpshed; College Slice

Factory Factory posted:

You want a liquid cooler competitive with air on everything but cost, look at the NZXT Kraken X60. It's a 280mm radiator, and that's big enough to let its "silent" fan setting perform at the noise floor, yet match the performance of the Swiftech H220 running at full tilt. Of course, your case has to support a 280mm radiator, which is significantly rarer than 240mm support.

If you're going to run them all-out, though, the H110 performs as well as the X60 but is slightly quieter.

I was looking at the Kraken X60 in the Anandtech review, but I think it would be too loud. HardwareSecrets found that when set to silent mode it was roughly equivalent in noise to a 140mm air cooler with a 1500rpm fan, and I wouldn't be able to live with something faster than 1200rpm or so (and I'm aiming for two fans at 1000rpm or less in my next cooler).

# ¿ Feb 17, 2013 19:04

Alereon: Feb 6, 2004; Dehumanize yourself and face to Trumpshed; College Slice

The fan speed on the Intel stock cooler is controlled by the motherboard, so it does make sense that you'd see different behavior on different systems. Guru3D got similar relative noise results for the Kraken X60, and reviews seem to confirm that it is quite audible even when set to silent mode. I think the problem is that the pump is loud enough that you can't use fans that push enough air through the radiator and still stay below the noise level of an air cooler that doesn't have a pump to deal with.

# ¿ Feb 17, 2013 19:36

Alereon: Feb 6, 2004; Dehumanize yourself and face to Trumpshed; College Slice

Dogen posted:

http://www.anandtech.com/show/6716/closing-the-loop-ii-new-liquid-coolers-from-corsair-and-swiftech/6

The H220 is a new dual fan design, as opposed to the H20-H220.

The pump is louder, but more effective, and also you can turn it down via PWM apparently...

edit: That anandtech review has it showing 30dba under both idle and load, also

Sorry, I got confused about Swiftech models, though on further review I probably wasn't even thinking of the H2O-H220 when I wrote that. I think you're giving the Swiftech H220 a bit too much credit here, it seems to epitomize the problems with water cooling: low performance, high noise, and a high price. The Anandtech review notes that the pump is very noticeable even at the lowest settings, and do keep in mind that because of how they did the measurements ALL products tested that aren't jet engines came in at <30dba so that doesn't tell you very much. The upgradeability of the H220 also seems overblown since the only component you can re-use for a graphics card is the pump, you still need a new radiator to dissipate 3-4X the heat of a CPU.

# ¿ Feb 17, 2013 20:40

Alereon: Feb 6, 2004; Dehumanize yourself and face to Trumpshed; College Slice

Dogen posted:

I flipped the top fan on my 650D to intake and slapped one of those demciflex filters on it a while back and the case interior is near pristine, which I attribute to both positive pressure and having all filtered intakes.

Since heat rises you probably want your top fan to be an exhaust fan, to act like a chimney. Cool air in the front and sides, warm air out the top and back. The most critical thing is that nearby fans work together, you don't want an "airflow short circuit" where air comes in one fan to go out the one next to it.

# ¿ Mar 1, 2013 05:54

Alereon: Feb 6, 2004; Dehumanize yourself and face to Trumpshed; College Slice

Do not loving run your Ivy Bridge at 1.4v or tell people it's safe to run at 1.4v :siren:

This is absolutely not safe and will damage your processor. Note that a processor losing its ability to overclock with time means that you are damaging it from excessive voltage. Also note that using power saving modes results in MORE voltage to the processor, especially if load-line calibration is enabled, because the voltage will spike up above the nominal value when the processor exits load. If you're already feeding the processor a high enough voltage to damage it, this is very bad.

Doing "suicide runs" or other extreme overclocking while being aware of the risks is one thing, convincing yourself that excessive voltage is "safe" so you can milk the last few Mhz out with terrible power efficiency while damaging your processor is another. I'm sure there are plenty of overclocking forums where you can talk about being an idiot with your hardware if you really want to.

# ¿ Apr 2, 2013 18:06

Alereon: Feb 6, 2004; Dehumanize yourself and face to Trumpshed; College Slice

Yeah that's been a standard feature of pretty much all processors since the mid-2000s, the generic term is Dynamic Voltage and Frequency Scaling, or DVFS. This is also used by your graphics card and cellphone, Intel calls their implementation SpeedStep. The more recent innovation is allowing the CPU to overclock itself if needed and power/temperature permit, which is Turbo Boost.

# ¿ Apr 5, 2013 01:47

Alereon: Feb 6, 2004; Dehumanize yourself and face to Trumpshed; College Slice

Death Bot posted:

Umm??? I'm looking at the temperature readings in CPUid and seeing them bounce by 10-20 degrees every second or two. Is something wrong with my heat sensor, or did I apply the thermal paste wrong (I have redone it once at the time of this post for the same reason) or what?

The CPU core temperatures should be visible if you scroll down the page in CPUID Hardware Monitor Pro, what you're looking at are probably random numbers from unconnected/uncalibrated sensors. If in doubt, check RealTemp TI. Wide variations in temperature are normal if load is varying, though.

# ¿ Apr 20, 2013 21:21

Alereon: Feb 6, 2004; Dehumanize yourself and face to Trumpshed; College Slice

Yes the core temperatures are correct, there's also a "CPU Package" temperature which should be the reading from the hottest core (so should be the same as the highest of the CPU core temperatures).

# ¿ Apr 20, 2013 22:24

Alereon: Feb 6, 2004; Dehumanize yourself and face to Trumpshed; College Slice

Anandtech has a new "Megaroundup" of air and liquid coolers using new testing methodology to better show the differences in noise and be more fair to air coolers. Here's page 7, which compares low-noise performance. The Swiftech H220 and NZXT Kraken X60 turn in quite respectable performance at the top of the pack, though the air coolers are generally leading the liquid coolers.

# ¿ Apr 26, 2013 21:18

Alereon: Feb 6, 2004; Dehumanize yourself and face to Trumpshed; College Slice

Daysvala posted:

I put about a pea sized blob of paste on the CPU, I spread it using a flat plastic card and then I mounted the cooler.

That is way too much thermal paste, so you probably have too paste blocking good contact with the CPU. You should be using a VERY small amount of paste, only enough to spread into a paper-thin, even layer BARELY thick enough to not be transparent. That said you may just need to back off on your overclock.

# ¿ Apr 30, 2013 20:43

Alereon: Feb 6, 2004; Dehumanize yourself and face to Trumpshed; College Slice

A quick correction to that: Thermal paste MUST be manually spread for direct-touch coolers, because otherwise it will just settle in the gaps between the heatpipes. On other types of heatsinks you can often get away with using the dot/line methods, but spreading yourself over the surface of the IHS will provide the best, most consistent contact since you're not leaving anything to chance. If you don't use WAY too much thermal paste clamping pressure from the mounting system will squeeze any excess out the side and you'll have a beautiful thermal paste imprint.

Alereon fucked around with this message at 22:36 on Apr 30, 2013

# ¿ Apr 30, 2013 22:23

Alereon: Feb 6, 2004; Dehumanize yourself and face to Trumpshed; College Slice

Daysvala posted:

I'm pretty clueless about the different types of heatsinks; my cooler has a copper faceplate which makes contact with the surface of the CPU, and has ~4 pipes which run out of that copper faceplate and into the radiator unit. Does this fall under the category of direct-touch?

You do not have a direct-touch heatsink. In theory the dot+squish method would work fine, though a pea-sized blob is just so ridiculously much that there isn't enough clamping force to get good contact, a BB is more like what you're looking at. Keep in mind you are only trying to fill in the microscopic surface imperfections in the metal so you don't need very much paste at all.

# ¿ May 1, 2013 01:19

Alereon: Feb 6, 2004; Dehumanize yourself and face to Trumpshed; College Slice

amp281 posted:

You have a really good chip. Delid it and see if you can get 5ghz!

While I understand that you're excited about extreme overclocking, you shouldn't be suggesting delidding to people who aren't already very experienced overclockers and without being very clear about the risks.

# ¿ May 1, 2013 17:22

Alereon: Feb 6, 2004; Dehumanize yourself and face to Trumpshed; College Slice

Initial Haswell overclocking numbers are out and they are very disappointing

It seems to get hotter faster than even Ivy Bridge, and require very high performance cooling.

# ¿ Jun 1, 2013 18:07

Alereon: Feb 6, 2004; Dehumanize yourself and face to Trumpshed; College Slice

I'd be cautious of the long-term reliability of an overclocked system with a 4-phase VRM setup. I mean on a $150 board you're probably going to have an okay experience if you're not goosing the voltage too much, but is the same true of a $120 or $100 board?

# ¿ Jun 2, 2013 22:45

Alereon: Feb 6, 2004; Dehumanize yourself and face to Trumpshed; College Slice

Factory Factory posted:

The motherboard VRM just sends in V_CCIN, a single fixed voltage that the IVR splits off into a half-dozen or so rails (including IA core voltage and DRAM voltage). The only just those four phases have to do is step 12V to a relatively static value between 1.8-2.3V and feed enough current. It's less work than full CPU voltage regulation.

Yes I understand that, my concern is the higher current draw through the VRMs on the motherboard than they're engineered for. Just like how there were some cheap/crappy AMD motherboards back in the day that caught on fire when you attempted to use 140W Athlon 64 X2s, I'm concerned that it may be pretty easy to overclock Haswell past the safety margin on cheaper boards.

# ¿ Jun 3, 2013 01:07

Alereon: Feb 6, 2004; Dehumanize yourself and face to Trumpshed; College Slice

Anandtech posted an article about overclocking using a custom liquid cooling setup. They built a fully custom setup using Swiftech waterblocks, radiators, and fans. The upshot is that while they got lower temperatures when overclocking, they were unable to actually push the CPU or videocards any further, and there was a huge noise penalty. I feel like if they had just turned up the fans on their heatsinks to comparable levels they would have achieved comparable or even better cooling performance.

# ¿ Sep 30, 2013 18:54

Alereon: Feb 6, 2004; Dehumanize yourself and face to Trumpshed; College Slice

Using four fans in push-pull on a radiator seems pretty excessive anyway. Most radiators aren't THAT restrictive, and if your fans truly aren't high enough pressure just doubling up on them seems like a worse alternative to getting more appropriate fans.

# ¿ Oct 2, 2013 21:43

Alereon: Feb 6, 2004; Dehumanize yourself and face to Trumpshed; College Slice

Protokoll posted:

Overclocking a 4770K on a Gigabyte Z87 board.

Yeah I'd be very, very concerned setting LLC to Extreme. On previous Gigabyte boards this pushed dangerously high voltage spikes to the CPU, especially at elevated voltage levels and with Turbo/Enhanced C-States enabled. I'm not sure what the behavior of the Z87 boards is (it may be vastly different due to the FIVR and changes to their VRM setup), but their Z77 and earlier boards manipulated voltage readings to hide these kinds of issues (reducing granularity and averaging over a window ensures monitoring software will never see spikes). Anandtech doesn't set LLC above High in their overclocking tests.

Then again I guess I'm pretty cautious with CPU voltage when overclocking and still hate Gigabyte for how godawful their power delivery was on 5-series boards, so I'm biased.

# ¿ Nov 15, 2013 02:08

Alereon: Feb 6, 2004; Dehumanize yourself and face to Trumpshed; College Slice

Silent PC Review has a their review of the new Thermalright HR-22 heatsink. This is a loving massive heatsink designed for "semi-fanless" operation: it comes with a flexible duct to connect to your case's 120mm rear exhaust fan, giving performance similar to a forced-air heatsink but without an additional fan. It's so big that in some cases you don't need the duct, it just butts right up against the rear fan and any top exhaust fans.

Pros:

Innovative duct exhaust system
Shelf cut-out at front so it doesn't interfere with DIMM heatsinks
Not that expensive given the size and potential cooling capacity
Best fanless performance ever tested in a case (as long as there is a case fan for ventilation)

Cons:

Mounting doesn't work very well because it's so big
Performance isn't better than current products when used as a normal single/dual-fan cooler
May not work in all cases
Open-air fanless performance not as good as the Nofan CR-95C

I'm not sure that SPCR did a good job testing this thing's capabilities, but it also isn't the obvious knock-out that I was expecting when I first saw it.

Alereon fucked around with this message at 03:36 on Nov 16, 2013

# ¿ Nov 16, 2013 01:38

Alereon: Feb 6, 2004; Dehumanize yourself and face to Trumpshed; College Slice

ShaneB posted:

I just snagged an H55 and an Enermax Magma fan to replace the stock fan. Under $70 on Amazon. This is going to be used for my G10 GPU adapter, but if it's sufficiently quiet I might be convinced to get one for my CPU. The H55 stock seems to perform within a few degrees C, even on an overclocked CPU, than the NH-D14. Even with something like a Gentle Typhoon and an H55 it's under $80 total.

I think you're overestimating the performance of the H55. In this HardOCP review it was 6C hotter and 3.7dBa louder than the Thermalright True Spirit 120M, which is a value-priced 120mm performance cooler and definitely not competitive with a high-end cooler like the Noctua NH-D14. You'd need a high-end closed-loop cooler to rival the performance of good heatsinks.

# ¿ Feb 13, 2014 04:09

Alereon: Feb 6, 2004; Dehumanize yourself and face to Trumpshed; College Slice

ShaneB posted:

To be fair I did mention that I would consider the H55 for a CPU cooler if it performed admirably enough on the GPU.

http://www.hitechlegion.com/reviews/cooling/liquid/32191-corsair-h55?showall=1& is a review I found yesterday that shows it closer to the air cooler, but who knows.

The important bit is that you have to consider noise along with performance, as the conclusion points out the Corsair H55 is very loud. In general terms a tower heatsink will beat a closed loop cooler if you run the heatsink's fan at a speed that matches the combined noise level of the CLC's pump+fan.

You can't put a giant tower heatsink on a graphics card though so that's where CLCs rule. I do feel like we're only a next-gen Accelero Xtreme away from air cooling coming back into competition, though. Maybe Thermalright will save us!

# ¿ Feb 14, 2014 03:38

Alereon: Feb 6, 2004; Dehumanize yourself and face to Trumpshed; College Slice

Spreading is the only method that provides consistent coverage and works with all types of heatsinks, air bubbles don't really occur in practice (your heatsink is not a flexible concave piece of plastic). You can check this pretty easily by looking at the contact patch after you dismount the heatsink. Though really, if you're using a heatsink with a correctly shaped base and sufficient clamping pressure it will even out any variation in paste application techniques. The main reason I'm not a fan of line/dot methods is it's an easy way to get awful coverage with direct-touch heatsinks.

# ¿ Mar 12, 2014 04:35

Adbot: ADBOT LOVES YOU

# ¿ Apr 26, 2024 04:46

Alereon: Feb 6, 2004; Dehumanize yourself and face to Trumpshed; College Slice

Ignoarints posted:

I'm probably doing it wrong, but I tend to use just a little more than recommended in the middle and tighten the poo poo out of it until I can see it come out all the sides and q-tip it up (on cpu). I've spread on GPU's because I read somewhere to do so

Honestly I think its more personal preference more than anything else, unless you're doing something obviously wrong like spreading the paste with something that will contaminate it (your finger or a Q-tip). If you're using electrically conductive paste (arctic silver) you need to be very careful, and if you have a direct-touch heatsink you could conceivably only get paste on one or two heatpipes if you apply it in the wrong direction, but otherwise there SHOULD be enough mounting pressure on the CPU to spread the paste where it needs to be and squeeze any excess out. It's harder on graphics cards where the mounting method is weaker, but there is also often an exposed die, increasing pressure by reducing surface area.

It also depends on the mechanical properties of your paste, if you use a thicker paste you can't depend on mounting pressure to spread it for you. For example Apple systems used to overheat because Apple used far too much thermal paste with the consistency of modelling clay, so mounting pressure was insufficient to squeeze it out from between the heatsink and die. If you used Apple official paste for the replacement you basically had to use a small trowel to spread it over the die like mudding drywall, but thicker.

# ¿ Mar 12, 2014 16:05

The Something Awful Forums > Discussion > Serious Hardware/Software Crap > Overclocking Megathread 3.11 for Workgroups

«‹›2 »