Why put a fan, (let alone two!) on a CPU heat sink?
You already have airflow enough to cool disks, it’s enough for a cpu. On the contrary, these fans will only disrupt the airflow, reducing the quality of the cooling and adding noise.
That’s simply not true in non-server cases. CPUs often dissipate 10x the watts from 1/10th the surface area of a HDD: and would throttle with cooling a HDD is OK with. And it’s because heatsinks do such a great job of slowing air down that you put a fan on them to keep it moving.
In servers, where you don’t care about noise and can tightly control where the air flows, you can run high-enough CFM that CPUs only have the heatsink and no separate fan (especially in 1u cases). But desktop cases don’t have the CFM, nor the ducting, and nobody wants fans that sound like a hair dryer.
5 Likes
I understand the reasoning, but in my experience with a number of gaming desktop PCs I’ve built over years I have stopped adding the cpu fans specifically because they made 0 difference. And GPUs dissipate way more heat than a dozen of disks.
Large case fan will produce more airflow at lower noise than a cpu fan. CPU fan on the other hand, if I’d does not exactly match the airflow speed (and if it does — why is it there?) will disrupt the laminar airflow inside lowering the cooling efficiency.
I recommend removing them and running the expected maximum workload. Might save yourself a few dozen bucks.
And lastly, surface air of a cpu heating is dramatically larger than that of a HDD, not smaller.
That’s ok. They can run much hotter than HDDs. And heat transfer is proportional to difference in temperature.
Look at almost any CPU cooler review, and most will include multiple temperature readings by fan RPM (same for GPU thermal testing) - usually for max-rpm and some normalize sound level. If you’ve noticed zero difference: then you simply haven’t measured.
Heat transfer through heatsinks… depends on turbulent flow: laminar reduces it.
I understand you may be able to get away with no CPU fan (or even GPU fans), but you’re still doing it wrong 
1 Like
That could be an interesting experiment but I would need to underclock the CPU (eco mode or something). At stock all core workloads reach TJ max (95C) with just the fairly high-end heatsink.
At low load you’re likely right. Rather than remove the fans, I could just set the fan curve to off and look the effect in HDD temps.
1 Like
Yeah, or I’ve seen people run just the center fan on heatsinks with two banks of fins. Whatever is cool and quiet and works without throttling. Big slow fans FTW!
I have measured. Just not what you are probably measuring.
I won’t even argue with you on this demonstrably false statement.
What is “right”? If I accomplish the same objective (running the device with temperatures within specs with lowest number of components — it’s not wrong)
For reference, this is how I do it: run the sustained workload to get power dissipation at 20,40,60,80, and 100% of TDP, and adjust the fan speed at that point so that the temperature does not go above high end of specified range for each component. Repeat for each load level.
You end up with quiet system that operates within specs.
This is quite an unexpected outcome. Something might be blocking airflow, or the heatsink leaked the refrigerant, etc. I have never seen this, (with 120W CPUs).
No need to do that. Modern (at least last decade) CPUs power gate themselves quite effectively. I see no dependence of power consumption at light loads vs clock speed.
Surface area of a CPU heatsink is vastly larger than that of a HDD. Not smaller.
I don’t know why are you making these demonstrably false assertions.
So on most consumer cases, and I know on my fractal from my pic, the airflow in the case itself is not that strong, and it’s not that directional. I mean, the front 2 and rear 1 ran are 140mm fans are running at low rpm to keep noise down. the air kinda just blow… around inside the case. Most goes out through the rear top exhaust fan, but some blows out through rear I/O vents or even vents in the bottom or top. But unlike a rackmount server case we definitely don’t have a blast of air going directly over the heatsink from front to back.
This relatively aimless airflow is good enough for the hard drives (they don’t get over 50c)
Plus the fans on the heatsink are 120mm or 140mm anyway, so fairly large and slow noise.
I’ve run CPU’s with no fan on the heatsink (usually while hardware swapping or testing) and… they get too hot. They hit 100c and then (hopefully) throttle down on speed.
1 Like
It’s surprising how fast CPUs can ramp up in heat. My last two desktops (3950x and 5950x) both had decent coolers (Noctua 120mm)… and idled and ran browsing/youtube tasks fine with no fan. But under load they’d go to over 90 degrees and throttle in seconds. But with a fan they were fine.
Eventually I got fancy and switched to a 360 AIO on the 5950x. It ran fine… until I had a pump failure. You’d think the pump head with it’s copper plate (and still full of stagnant coolant) would work OK… but I could no longer even boot! If the system was cool I could make it to Windows… but it would inevitably shut down.
So I’ve learned my lesson: back to aircooling for me! I got one of the older NH-15’s and have been fine ever since.
That’s curious. I just tried the same on my main rig (7950X3D capped at 85C in BIOS and a high-end heatsink – NH-D15).
Idles at 56C with GPU a light load below. Starting a workload with only half the cores reaches 85C with fans off in short order, but stays at 75C with 800 RPM. I’d need to ramp up the case fans a ton to compensate if at all. The CPU fans definitely do something!
1 Like
That’s the problem right here. Fans help with this disaster as a secondary effect. I hope you have ran the experiment with the case closed, correct?
The actual problem is it shall not climb to such as high temperature with half cores idle. 55C idle is also way, way too high.
So let’s diagnose and solve this problem first.
Run the workload that has your CPU running at the 85C sustained for half an hour. At this point the heatsink shall be over 65C, meaning, impossible to touch.
If this is not the case — you have a heat transfer issue. grab a thermocouple (most multimeters comes with one) and measure the temperature of the CPU lid, heat exchanger, and pipe on the top of the heatsink.
Depending on the outcome either heatsink is dead, or thermal compound dried out, or cpu package stopped making a contact with the lid.
My tower machines have CPUs idle at under 35C when ambient temperature is 25C and case fans are at the lowest setting, barely moving.
Another thing to consider is whether the higher C and P states are allowed (bios/uefi). 55C idle is wrong.
Now I don’t have a thermocouple, but I do have an oven thermometer! “Use what you have”.
I can’t reach the heat pipes so I laid it flat on top of the heatsink. Ambient=24C
Idle: CPU=50C, Heatsink top=29C (21C delta)
Load with fans: CPU=75C, Heatsink top=32C (42C delta)
Load no fans: CPU=85C, Heatsink top=60C (25C delta)
Remember it’s not 100% idle, GPU running at 45C (heatsink is very close and soaks it up) / AMD idle temps are not great / 8 cores will clock higher than 16 within budget.
Seems to be doing its job overall. The higher delta is curious but probably because I’m measuring the outer areas that’s already cooled by the fans. If the heat transfer was poor then fans wouldn’t have much heat too move off the heatsink and be ineffective, no?
2 Likes
If there was no heat transfer, then fans would not be effective. But there is some transfer obviously and fans help somewhat.
My point is still that if there was sufficient amount of heat transfer airflow from the case fan would have been sufficient. The heatsink would have been hotter, and therefore small airflow from case fans would have been enough.
Looks like heat transfer between cpu and a heatsink OR between heat exchanger and fins may not be great — such a drastic temperature reduction of a heatsink temperature results in only modest reduction of CPU temperature.
But again, we don’t know if this is a heat exchanger problem or paste, (and how much can we trust the over thermometer — albeit numbers don’t look unreasonable)
Have you checked just in case if your processor is allowed to enter c-states in the bios? It shall help with light workloads.
Aren’t they a bit overpriced and more of a fashion statement by now than an actual tech product?
Any low RPM fan will produce low noise and low airflow. There is no magic. Whatever alleged minor improvement they provide does not justify 4x markup they hope to charge.
I bought into a hype, got a few, got thoroughly disappointed, and returned them all. Now my fans are all “cheap crap” from eBay. If I want low noise — I reduce RPM and increase number and/or size of fans. I refuse to pay for spiffy boxes, snazzy cables, and marketing.
/rant.
Noctua fans are indeed very nice and also very overpriced.
Thermalright outta Taiwan has been eating people’s lunch of making cheap but good heatsinks and low priced fans.
2 Likes
I’m not quite satisfied with them to be honest! Their 120mm and 140mm offerings have been great for me, but I am on the lookout for better replacement fans.
The old 32mm deep fans moved a lot of air for a reason. Drivetemp is in the low 40s right now, which is well within spec, but I would love to source a silent 32mm deep fan and spin it at the same RPM for slightly better temperatures.
That’s very much in character for you to say 
Noctua makes nice fans. I agree, that they are as much of a fashion statement than everything else, but they are worth their money. So are NoiseBlockers, Yate Loons, Gentle Typhoons, Phanteks and many other competitors.
Running larger and/or additional fans at low RPM is not always a possibility, and in that case having quality fans is neat. Like I told Roxor, I’m on the lookout for replacements to the noctuas. I’ll see what I can source.
3 Likes
Put ASIC fans on that nasty NAS… they will cool the entire house. 
Jokes aside, it’s nice that you can change fans on some Synologys like this one. On the lower-end models, like 220+, you can’t. They have some switched pins I believe.
Noctua makes great fans as reliability. Those babies will work for years. For the air flow, you need models with higher airflow pressure. And from what I observed, fans with wider blades can do that, not more and thin or longer. Just wider blades. You can see those in ASIC fans. Bitmain used to have very good fans from a japanese company, Nidec. Maybe you can find a proper model for you.
OK I CANT TAKE IT ANYMORE.
I already found the lovely fans below, and have ordered them. I wanted it to be a surprise, but I can’t risk anyone finding them before me and taking me lovely internet points.
at 38 (!) mm deep they should push plenty of air, and with the ability to spin all the way down to 500RPM, they can be made quiet as well. Will report back when I have data on them!
Server Fan - S8038-7K | 80 mm 7K rpm Server Fan | 4 Pieces | ACFAN00292A | ARCTIC
Oh and they came in a 4pack, which is just marvelous for my need.
3 Likes