Is liquid cooling necessary for high CPU performance over extended periods?

Question

I'm planning on building a system for high CPU and memory performance. Graphics performance is not a major concern, as I do not plan to use it for gaming. I noticed that liquid cooling seems to have become common since my last build, and I am wondering if it is needed in my case. I do not plan to overclock any of the components.

The intended use for the system is primarily development and occasional compute-intensive tasks that will run on all cores continuously for several days to a week. The system will not normally be in continuous usage like this, but it needs to be capable of occasional extended periods of continuous use at full capacity.

I have included the main components below.

CPU: i7-5960X

Memory: 64GB DDR4

Graphics card: Nvidia GeForce GT 730 2GB

Drive 1: 1TB SSD

Drive 2: 4TB 7200RPM HDD

Power supply: 700W

Is liquid cooling needed for the above build? If not, is there any advantage in using it? I have read reviews that mentioned liquid cooling unit failures that resulted in leaks, so I would rather not use it if it is not necessary.

Thanks!

Solution 1

In short, no

Liquid cooling is still mainly for bragging rights. Getting those extra MHz from hardware and pushing for extra benchmark points.

Will you need something aftermarket to cool your system? Absolutely.

Of course keeping temperatures as low as possible is always a thing we strive for, but you have to weigh up the costs and risks to the performance benefits.
Pumps, radiators, reservoirs, heat sinks, fans, tubing and coolant add large amounts to a build's cost.
Even All-in-one's are quite pricey (and can cause problems, boil-off, pump fail's etc).

Another downside with water cooling is that it is difficult to tell that anything has failed. They can operate near silently and leaks can go undetected until something blows up.
This is not to say they are less reliable than fan-cooled systems but at least you know if your loud fans have stopped working and there's very little chance of them taking anything else out with them if you are around monitoring.

Basically I would get a beefy heat-sink with at least a 120mm fan, something like a Noctua or Zalman (I am not affiliated) and then an intake and an exhaust to keep cool air moving. RAM cooling is rarely necessary but solutions are available from companies such as Corsair.

Get quality assured names with plenty of reviews. This PC sounds fairly mission critical so invest wisely to keep it running.

Solution 2

Honestly, as an owner of the same CPU in a similar system, I would say most definitely. Do not be fooled, the 5960X is a very fast and efficient CPU, but it's also deceptively hot because it's actually so much smaller than the 3930k I upgraded from and thus the heat is much more concentrated. And even the 3930k was very heat sensitive.

I RMA'd my 3930k due to heat damage while it was running on air and talked to Intel directly about it, so I will talk more about that than the 5930k I have now.

If you actually look at the specification of the CPU, which nobody ever seems to do these days, it's only rated for 66 degrees max (virtually the same as the 3930k). According to Intel, that's because the CPU is so far from the heat spreader and (CPU) case temp sensors that when the sensor reads 66, the cores are significantly higher in places. I RMA'd my 3930k due to heat-induced failure and it never did over 75C but it must have been a lot more because that managed to burn out TSC circuits in the chip and it's self reporting features refused to boot any OS.

That was on a Arctic Freezer i30 at stock clocks. It's actually damn near impossible to keep the 3930k under 62C (it's max is a little lower than the 5960X) under load without a very noisy system, and I'd actually alt-tab from gaming sometimes to let the CPU cool, so add two cores, make all of them smaller (so the heat is more densely packed) and your problems only increase, even with lower power draw.

I bought a closed loop water-cooler. I suggest you do the same. (H80 FYI) Never tried to overclock beforehand, but after I can run the 3930k @ 4.3 @ 63C under load, and low 50s stock under load. It's my only Closed loop, so I expect the same from others, but it's built very solid and I highly doubt it will ever leak. I would recommend not building your own for that very reason.

I used the same cooler on my 5930k, and it maxes 56C. I'm don't really see a point in oc atm so I haven't tried. My baby is still new lol. I will comment that you might wanna consider ECC RAM and a Xeon if you're doing something like Programming. I've ran into guys that had single bit errors while compiling that caused their software to be released with a bug that took them an incredibly long time to figure out and fix.

Solution 3

For reliability, you may want to look into passively cooling the CPU with an oversized heatsink and install extra large fans in the chassis.

The reason is having the CPU fan fail tends to cause a much longer downtime than replacing a chassis fan. If you run the chassis fan directly off the PSU instead of the mobo, you can even change them out without turning off your computer.

Larger fans also last longer because it doesn't need to spin as fast to push the same volume of air. It will also naturally run quieter for the same reason.

Solution 4

Yes.

I just put together a new Xeon E3-1276 v3 (Haswell) board (like the i7-4790 with more cache and ECC memory capability). With the supplied cooler, even with no case, it overheated (100 degrees Celsius! ) after one minute running Prime95 64-bit. (NOT overclocking, BTW)

Even a simple closed-loop liquid cooler essentially moves the bulky radiator outside and removes the size and weight restrictions for motherboard mointing and fitting inside a case with airflow.

I've never had a leak once a system is tested and comissioned, since I turned to water cooling in the Pentium Prescott era.

For mundane systems I use something like the Coolermaster big square radiator that sits on top of the CPU with a 140mm fan on the side. But for sustained performance cooling, you would also need (at least) case fans or ducting to go with that, and I don't know just how much it could handle without being very noisy and take a huge volume of air.

From the discussions in the comments, it appears that performant code can push it far beyond the thermal limits, so water cooling is necessary to make use of that performance capability.

To make it clear: this is not overclocking and is running the MB, memory, and CPU at stock specifications. Number crunching code can significanly exceed what the supplied stock cooler can handle.

Likewise, the turbo core feature is not applicable because (with the standard settings) it will not be engaged unless some of the cores are idle, and simply not boosting will not be considered an error if boosting would push it over the limit.

An after-market heat sink that maxes out the weight that can safely be supported by the motherboard could handle more heat if you give it enough air flow, but I can't say for certain what that is capable of with in-case use. My gut feeling is that if it doesn't make as much noise as a vacuum cleaner, it won't keep up.

---

Considerations: some similar processors have fewer cores or lower clock speed, and that would make a difference.

I also have a E3-1245v3 (3.4GHz instead of 3.6) running a file server and it has no problems whatsoever with a 5" square heat sink. The emphatic yes to liquid cooling is for High Performance Computing (CPU number crunching) usage.

Also, if testing the effectiveness with your actual use case, be careful not to under-do it. The software might get updated to use new instructions or be optimized for that architecture, and you would not be able to take advantage of the performance improvements, if that now pushed you 20W beyond your thermal limit.

Solution 5

The major reason liquid cooling is so much more common is that it has gotten much simpler, cheaper, and more reliable. It is not because new builds require the liquid cooling. That said, there are some times when it's worth the additional set-up complexity.

It depends in part on the motherboard. If it is one advertised for overclocking/gaming et cetera, it may have been designed with the assumption that there would be a liquid cooler in play. I ran into this where there was a capacitor bank right by the CPU, while the CPU never got above 70C, the capacitor bank impeded air flow from the chassis fan and eventually cooked.

Also, liquid cooling is quieter, more efficient, and lets you pull the heat all the way out of the chassis instead of just out of the CPU. If the computer is going to be someplace where people regularly are (office, home, et cetera), then the noise reduction is nice; if it's under a desk, having the chassis stay cool is nice. Also, if you start with an all-in-one kit, and replace the hoses with longer ones, you can potentially mount the radiator outside, which can make the liquid cooler pay for itself in the summer. The cost on sale for an all-in-one kit is often similar to the cost of a chassis fan + better heat sink, so it mostly comes down to a tradeoff between extra maintenance and quieter cooling operation.

View more solutions

Comments

[ps2goat]

I like the closed loop liquid cooling units that are for the cpu only. They don't fit in all cases, but they do keep the case quite cool. Compared to other aftermarket fans, it only costs 1.5 - 2x as much. (start around $99 US or so)

[Perkins]

They don't even run that much, newegg has decent ones around $60, they'll come on sale for $45 about once a year.

[Admin]

Liquid cooling can also help reduce noise. In general, to increase the cooling capacity in air cooling you need to throw more fans at it (more noise), where as in liquid cooling you can throw more radiators at it without necessarily needing fans (you can easily get away with completely turning off the fans under low load).

[Richie Frame]

liquid cooling systems tend to result in increased board temperatures due to the remote fan location, making the system less reliable in the long term in the case of HEDT platforms. If using liquid cooling, make sure a large fan is cooling the board as well

[Kik]

The answer to this is 100% dependent on how much you are planning to overclock your processor.

[Qudit]

@RichieFrame By that you mean large case fans?

[Richie Frame]

@Qudit no, large case fans not blowing air in the right direction will not help, they need to be blowing air directly against the motherboard, I have a HAF-X case and it required a 2nd directed fan because the 200mm side fan did not provide enough cooling above the top PCIe slot, and the front fans provided almost no cooling

[Slizzered]

Just so it is not overlooked, I want confirm the opinion of user1901982 and @MonkeyZeus regarding the use of ECC memory. The more RAM you have (and the more you USE it, so your 1-week pattern is pathological), the more likely is the occurence of bitflips. If you do compute-heavy things, this might or might not skew your results or crash the machine. It's not going to be often, but still... Of course, that would require a different processor (Xeon & the likes). See www.cs.toronto.edu/~bianca/papers/sigmetrics09.pdf & zdnet.com/article/dram-error-rates-nightmare-on-dimm-street

[ChrisInEdmonton]

While I strongly advise and always buy an after-market cooler, I dispute the claim that it's required; the built-in cooler should be sufficient. Nevertheless, this answer hits the high points and gets my +1.

[Ctrl-alt-dlt]

In my experience, the high end 6+ core Intel CPU's (even more so AMD 'equivalent's') come packaged with coolers that yes, keep them cool but not to standards that will avoid affecting lifetime. I believe anything above 60 degrees under max load is too hot for a CPU to be running at 24/7.

[Qudit]

@Jamie Willetts Thanks for the detailed and informative answer! I think I will go with powerful fans and heatsinks as you suggest.

[Qudit]

Thanks for the tip about motherboards designed for liquid coolers. I don't care about the cost of liquid cooling systems as it is negligible compared to the rest of the system anyway. However, I don't like the idea of water near expensive electronics.

[ashes999]

@JamieWilletts most CPU manufacturers suggest keeping your CPU temp under 100C to prevent damage. Why do you suggest a much lower temperature of 60C?

[Robert Harvey]

@JamieWilletts: But not the quad cores. My Intel® Core™ i7-4790K with a stock cooler immediately starts thermal-cycling if I try to overclock it. An after-market cooler seems like a necessity.

[Mixxiphoid]

@ashes999 the border of 100C is to prevent direct burn through damage. The border os 60C is a nice standard (and by no means a law) to prevent wear out damage.

[GalacticCowboy]

Also, make sure that your case will accommodate the cooler. I had to modify mine by trimming away part of the motherboard tray for the under-board part of the mount, and there is less than 1/8" clearance between the top of the cooler and the side of the case - and the case has a "bump out" at that point, so a normal case wouldn't have taken it.

[Michael Stimson]

@ChrisInEdmonton, the standard heatsink is adequate if you want to do 'standard' tasks - browse the web, Excel and Word. If you want to thrash all the cores for many hours the standard heatsink is woefully inadequate. My workstation ([email protected]/32GB RAM) with the supplied heatsink would overheat and crash if I used more than half the cores for 5 minutes; installing a decent heatsink stabilized the system... this is a work PC, definitely NOT overclocked. Now I can process on all 8 cores.

[Michael Stimson]

And don't forget that the fan is responsible for dragging in dust and fibres that get caught in the grills and reduce the effectiveness over time. Most good cases have enough fans to circulate enough (air conditioned) air through to make passive heat sinks viable... on the down side the extra weight can bow the mainboard and stress fractures can occur - passive heat sinks are much heavier than the active ones. Either support it with a strap or lay the case down so it's not dragging perpendicular to the mainboard.

[Luaan]

@MichaelMiles-Stimson This depends hugely on both the CPU and the cooler, though. I've got a quad-core Xeon with a cooler that's incredibly silent (seriously, you can't tell the difference between the computer running or not, even at night) and actually works perfectly - under 100% load, I can run for hours with no issues whatsoever. And both the heatsink and the fan are actually quite tiny. Of course, I also have a great case, and the ambient temperature in the room seldom gets over 20° C. I'm expecting some decrease over summer and over time.

[JamesRyan]

The standard cooler is appropriate for running at standard speeds. A bigger cooler is appropriate for overclocking. This situation is made slightly cloudy by the fact that turbo mode overclocks by default in the latest cpus. This uses conservative speeds though so that should also be fine on the stock heatsink in a cool ambient temp. But if overclocking, high temp environment or continued heavy use better to go safe with a bigger cooler. Ideally you want to be at 60C or less but any temp upto 80C should be fine even for extended use.

[Richie Frame]

With a 140W TDP, a passive heatsink would need to be comically oversized, and the airflow in the case directed using baffles for proper cooling

[Qudit]

I get the impression that your experiences with leaks are in custom cooling systems rather than the sealed all-in-ones that are available these days. Is that right?

[Nelson]

The TDP alone can't really account for the viability of the setup. It really depends on the available chassis fan positions.

[Perkins]

That is correct. The case I use has a slot for running hoses outside the case. Anyway, the custom unit I'm using used to be an all-in-one, but I wanted longer hoses, and a settling bulb. The part inside the case is all original parts and where it sprung a leak during testing. I had to disconnect the hoses from the pump, and didn't quite get them tight when I reconnected them. One word of caution, some of the (old) all-in-ones are filled with coolant that can gel if it overheats, so if the radiator's air flow gets blocked, you could have problems (hence the settling bulb now).

[Dan Is Fiddling By Firelight]

FYI newer versions of Prime95 use FMA instructions in stress testing (and presumably normal prime hunting too); making it function more like a power virus/short term burn test tool than the long term stress tester that it historically was. For my 4790k, IIRC, they pushed it ~20W above TDP at stock speeds/clocks; and 15C above the thermal load of an older non-FMA version. (I was running watercooled so my thermals were reasonable; I wouldn't be so confident about the stock Intel cooler doing well.)

[Rag]

+1 for Noctua. Unbelievably high-quality fans. (Also not affiliated)

[user-2147482637]

@JamesRyan From my understanding, turbo boost on intel does not clock under 100% usage of cores, it only over-clocks when not all cores are used. Also, I have dual 6-core xeon's with the stock heatsink and fan, they run under 100% load for weeks with no issues, just one more datapoint

[JamesRyan]

@user1938107 turboboost will still overclock multiple cores but not as much. It measures power consumption and manages the overclock to stay below the rated TDP of the chip.

[JDługosz]

Isn't that fair game for any performance computation code? That's what the instructions are there for. The stress testing suite is just a normal run with already-known answers, not code designed for benchmarking or burning. I suppose it means that only newer code will push the CPU so hard, and the load can vary considerably depending on the code. I would not want to under-cool a HPC machine because it seems OK, in case I update the software and it's been recompiled to use the new instructions or optimized for that generation.

[JDługosz]

FMA is fused multiply add. Under different names, that's been important for vector (simd or pipelined) processing on floating point DSP chips and other solutions, for > 20 years that I can recall. Your observation can be interpreted as meaning that the stock cooler (even with the packaged Xeon?) Is not suitable for HPC (high performance computing) usage. Most servers are database and web servers that mostly loaf.

[JDługosz]

@jamie, although accepted as the correct answer, no, it's mostly about bragging rights is incorrect for this family of CPU and the stated intent of performance computing work. I strongly suggest that the OP first test the board/ram/cpu with the stock cooler for acceptance testing and burning in, and just try it with a load indicative of his intended use. Then if a 5-inch square heatsink is anticipated, try that too (with Prime95) and post results back here.

[JDługosz]

@ChrisInEdmonton the supplied heat-sink and fan, in my recent test, was laughibly inadequate. Earlier experiences with Intex boxed processors indicated that it was silently cutting the CPU speed to remain in spec. So don't assume the supplied processor is adequate just because it is supplied: it's just not so.

[JDługosz]

Fwiw, I built a radiator into one side of the case so the same fans that push air through that also draws air out of the case. That's some thing you don't get with hose-connected external units. More cases are being sold with that in mind, or a small radiator can screw to the outside of where a case fan would go.

[JDługosz]

@Qudit note that two fans in the same flow (push/pull, opposite sides of the radiator) don't increase air flow very much, and can be much louder. It reduces the resistance to flow, but does not double the flow. For a heat sink radiator that is made to work with one fan OK, its only point is for fail-safe redundancy.

[user-2147482637]

@JamesRyan right, in the scope of this question, i thought it was important to note that the idea of overclocking in relation to heat and intel turbo boost is controlled even with standard coolers. Just as some extra info for readers, As for TDP, intel states in v2 " allows the processor to operate at a power level that is higher than its TDP configuratio" intel.com/content/www/us/en/architecture-and-technology/…

[JamesRyan]

@JDługosz I'm not sure you read the question correctly. Liquid cooling is still not really required because very near the maximum overclock on this CPU can be achieved simply by using a bigger heatsink and fan.

[JDługosz]

I think @user1901982's answer is definitive since he has the exact same chip. Read correctly: I reread the first sentence-high CPU performance...*i7*. He's not overclocking. The 5xxx series has already developed a reputation for heat problems (explained by another Answer here as under-reporting the core temp). Assuming the largest heat sink that the MB can support w/good heat pipes, more air means more cooling, but just how much is needed? Has anyone managed to run a current build of prime95 x64 without it slowing down due to heat, with that? If so, that would be definitive.

[Perkins]

That was my initial setup, but I didn't get as good of performance that way. The air going to the radiator is pre-warmed, so isn't as effective at cooling the radiator. It will very a lot from case to case, but my normal operating temp is ~15C with it outside the case, 35C under load. With it inside the case, or afixed right outside the case drawing air through the case, it was from 25-45C. I do use a separate chassis fan, which keeps the case around 30C regardless of load. If you're going to the expense of a liquid cooler, including a single chassis fan is negligible.

[user1901982]

I could never say that it's not possible for it to leak, but honestly, without a knife I don't think I could pull the pipes from the CPU block the H80 if I tried as hard as I could, they are so incredibly tight. I wouldn't worry about it, but then a H80 is significantly more upmarket than a H55 so that may not be the case for all of them.

[user1901982]

@JDługosz I can verify that, it's mainly because I use an old school manual fan controller with the fan quiet and then have the BIOS beep if it reaches 60c, incase I forget to turn it up under load. If I do, I can blast the two fans at 100% for a moment. Other wise I find the whole thing much quieter with only the fan between the rad and the rear fan grille on the case running at all.