It's a 45W TDP quad core CPU in a tiny case. What did you expect?
Also BIOS isn't the best place to monitor idle temps as the BIOS probably doesn't have all power management features enabled (deep C states etc.). Better check on Linux or Windows. But anyways, the CPU gets really hot under load. I've had it hit thermal throttling (i.e. reach 100°C) on hot days under full load (and performance set to max in BIOS, which I think means 50 W sustained, 60W peak power allowed).
Also, I don't think 60°C is crazy. It's a pretty normal operating temperature. My current laptop (35W Ivy Bridge) idles at 60°C under Linux, i.e. with full power management enabled. If you want lower idle temps you can increase the fan's minimum duty cycle. I wouldn't recommend that, though, because it doesn't help with load temperature and 60°C is fine.
I EXPECT it to run at a low temperature when it is not under load. Like my laptop. It packs an i7-6500 and runs cool. I don't know the temp, but it stays cool. The fan barely spins up at all.
My desktop is in an ATX chassis of course. But it has a 1000W power supply. 8 DIMMS (128GB). XS1715 enterprise U.2 nVme SSD. TWO 10k RPM Enterprise SAS Hard Drives. GTX 680 card. At a room temperature of 80F, it idles at about 49C. In 1440p gaming (World of Warplanes), it peaks about 56C.
When the room is cold in the winter, it idles at about 39C.
Maybe in Windows, the CPU will be handled differently and it might be a little cooler.
3 of 3 people found this helpful
While in Visual BIOS, the processor is never in an idle (no load) state. First of all, all power management capabilities in the processor are disabled. Secondly, all interrupts are disabled. This means that the processor cannot halt itself (as it does in an O/S) when it has nothing to do and wait for an interrupt to awaken it. Instead, it must continuously loop around polling all of the possible input sources to see if anything has changed. Bottom line, monitoring temperatures while in BIOS is simply not the thing to do and a waste of time as a comparison method.
Secondly, let's put these temperatures in perspective. For the processor, this level is considered nice and cool. This processor has a TJmax of ~100°C and a projected Tcontrol of ~80°C. This means that the processor can run at or below 80°C all the time - and in fact can do so for its entire warranted lifetime - without suffering any thermal-related degradation.
Thirdly, let's talk about the chassis. We are talking about a very tiny case with a very small air volume inside. Your tower system is very highly dampened by the (comparatively) massive air volume within that chassis. In this small chassis, on the other hand, with so little air mass, it takes a lot of air movement to dissipate the heat generated by the processor and other components in this system. This effort is hampered by the small size of the chassis. It is so tight inside that we cannot fit a (relatively speaking, efficient) fan into the chassis; instead, we have a (relatively inefficient) blower unit that must spin at far higher speeds to move the same amount of air. On top of this, folks demand a quiet system and do not want this blower spinning at speeds that generate noticeable sound - a requirement completely at odds with that for cooling...
Finally, let's look at the cooling strategy. In an effort to keep the system as quiet as possible, the default algorithm used does not begin speeding up the fan until the temperature reaches a significant level (specified by the Minimum Temperature parameter). If you care more about keeping the temperatures down, lower this setting and start the blower running faster at a lower temperature. You can also raise the Minimum Duty Cycle parameter and have the fan run at a faster speed all the time. Note that, if you adjust either of these parameters, you must also adjust the Duty Cycle Increment parameter so that the fan reaches 100% duty cycle at or before the processor temperature reaches ~80°C. The algorithm you should follow for doing so is,
Duty-Inc = (100 - Min-Duty) / (80 - Min-Temp)
For example, If you set the Minimum Duty Cycle (Min-Duty) to 25% (note: this is the lowest you should ever use) and the Minimum Temperature (Min-Temp) to 40°C, you need to set the Duty Cycle Increment (Duty-Inc) to,
(100 - 25) / (80 - 40) = 75 / 40 = 1.875
Rounding up to the nearest integer, this means that you must set to the Duty Cycle Increment to 2. Overall, this means that, for temperatures below 40°C, the fan will spin at ~25% of its maximum speed. Then, for ever degree the temperature gois above 40°C, it will increase the speed of the fan by 2%. Because of the rounding we had to do, for this particular example, the fan will actually reach full speed (100% duty cycle) at 78°C, but that's the best we can do (the hardware only takes integer values).
I hope this helps you understand the situation. You really need to do your comparison while in Windows, where all of the power management features (and interrupts!) are enabled and active...
Thank you for the good information. While in Windows, the NUC does seem quiet and cool. Although I don't have a handy app to monitor it. To Lokutos, you experienced THROTTLING at 100C? I had a cooler failure on my desktop. It completely shuts down above 95C.
The full on state in BIOS leaves a bit to be desired as it does not seem to detect plugging in and disconnecting USB sticks without rebooting. I suppose no one thinks you will be sitting in BIOS for exended periods of time.
Its nice to know it can run at 80C all day long with no problems. I just prefer mid 50s MAX. For now, it will be general purpose. Watch videos / streaming. MAYBE light browsing via wireless keyboard.
To Lokutos, you experienced THROTTLING at 100C?
Yes. The processor's thermal protection kicks in at 100°C (Tjunction) and causes it to throttle. I've had that happen with the KY under heavy load (x264 encode, linpack) on a hot day. With ambient room temp at ~21°C it does not happen but still gets up to about 95°C under such loads.
I had a cooler failure on my desktop. It completely shuts down above 95C.
That was probably a feature of your BIOS, not the thermal protection of the CPU. The CPU can also shutdown ("halt", probably means complete freeze, power gating of all units) as a final thermal protection measure if throttling doesn't help and another trip point is reached (on Skylake somewhere above 100°C, so I don't know the exact point from the top of my head).
1 of 1 people found this helpful
It doesn't necessarily need to be the result of some BIOS feature.
In Desktop, there can be individual variation in the Tjmax temperature set-point of a processor (my understanding is 100±6°C) and thus it is certainly possible that his processor's Tjmax could be 96°C. Secondly, the Digital Thermal Sensors (DTS) inside the processor only have the ability to measure the temperature offset from their Tjmax temperature. That is, they cannot measure or expose temperature readings that are above their Tjmax temperature. As a result, if temperatures continue to rise above the Tjmax set-point, the Tjmax temperature value will continue to be displayed. If the temperature then reaches the Tcrit temperature, the processor will be shut down to protect it from any further damage...
Hope this helps,
P.S. Just so you know, the Mobile/Embedded SOC packages used in the NUCs have a fixed Tjmax temperature of 100°C.
Setting "Minimum Duty Cycle" in Skull Canyon NUC BIOS to 25% results in the fan being loud enough to be a nuisance with the fan spinning close to 2800 RPM. Setting "Minimum Duty Cycle" to 15% makes the fan spin at around 1800 RPM, which makes it basically silent. Why are you saying that "Minimum Duty Cycle" should not be set below 25%?
I almost returned the Skull Canyon NUC before I found these settings in BIOS today. I was able to make it completely silent while running VMware ESXi (without any VMs started yet) by setting "Minimum Duty Cycle" to 15%, setting "Minimum Temperature" to 75 C, and setting "Duty Cycle Increment" to 1%. I have not yet stressed the CPU, so I would like to understand your take on the settings that would ensure the lowest fan speeds possible (without frying the CPU) to be able to run CPU at a constant (more or less) 30-40% CPU utilization with the lowest fan noise possible. In my use case, the GPU will not be utilized at all.
The way that these settings were set by default in BIOS made it impossible for me to keep the Skull Canyon NUC running at 30-40% CPU utilization in the same room where my desk is. The fan speed was way too loud for my taste.
75°C with 1% increment is way too low. Together with your starting duty cycle this gives a mere 40% duty cycle for the fan at the already critical 100°C, i.e. the fan will never reach full speed. Given that this NUC model gets really hot already with the fan at 100%, I doubt this is a good idea. You will definitely run into thermal throttling. Thermal throttling is not designed as a normal means to operate the CPU; it is meant as a last resort to protect the CPU from damage.
You should set the fan increment much sooner and/or steeper so you reach 100% latest at 90°C (better already at 85). If heat is a problem (because of fan speed) you can set PL1 / PL2 in the BIOS to something low, which would cause the processor to throttle when hitting those power limits. Note that you cannot beat physics by simply setting the fan speed how you desire. The thermal power needs to be dissipated / removed somehow.
1 of 1 people found this helpful
The blowers that are used in the NUCs have a spec'ed minimum reliable operating duty cycle. When asked to run at duty cycles lower than this level, operation is not guaranteed. This can manifest as the fan simply stalling or it may struggle to keep spinning, speeding up and slowing down as it fights its own inertia. Every fan is different and how it operates over time will vary - and degrade as bearings settle, dust accumulates, etc. In the latter case, this struggling can sometimes manifest as annoying sounds/patterns but, more importantly, the struggling can affect the reliability of the fan over time. I thus do not recommend using settings below this level. Fans are sourced from multiple vendors and this setting will vary from one brand/model to another. The average was in the vicinity of 25% and thus that is the level I recommend. This is only a recommendation; you are free to do whatever you want. I certainly cannot hear my fan at all at 25%. Of course, the blower's noise level will vary from one system to another - and I am in my late 50's and my hearing has certainly degraded some...
I've done some experimentation with the "Minimum Duty Cycle," the "Minimum Temperature," and the "Duty Cycle Increment" settings in the Intel Visual BIOS in Skull Canyon NUC. Certainly, my preference is for the NUC to run as cool as possible, but my even bigger preference is for the NUC not to annoy me with its fan noise 8 hours or more per day while I'm working in my office.
The task that I acquired the Skull Canyon NUC for is to install VMware ESXi to run multiple VMs. I'm currently using 2012 i7 quad-core Mac Minis for this purpose, but they have a limitation of 16 GB of RAM per Mac Mini. At this point, I need to run twice as many VMs in my lab as I'm able to run on two Mac Minis (6 VMs per Mac Mini). My hope was that I would be able to run twice as many VMs on one Skull Canyon NUC as I can on one Mac Mini, and this proved to be possible with 32 GB of RAM installed in the Skull Canyon NUC. However, the fan noise of the Skull Canyon NUC with the default BIOS Cooling settings was unacceptably loud.
So, after some experimentation, I've arrived at the following settings:
Minimum Duty Cycle: 15%
Minimum Temperature: 76C
Duty Cycle Increment: 5%
These settings would result in the fan speed reaching 100% RPMs when the CPU temperature reaches 93C
I've also lowered PL1/PL2 levels by setting the "Processor Power Efficiency Policy" in the Power screen of the Intel Visual BIOS to "Low Power". I believe, this lowers the PL1/PL2 levels to 35Watt/45Watt.
With the above settings, I'm able to run 12 Linux-based VMs concurrently on one Skull Canyon NUC with the fan noise being barely detectable within 1 foot from the NUC, and CPU utilization oscillating around 31%. As I fire up the VMs, and as the VMs start loading services (these are Linux-based Cisco Unified Communications server appliances), the CPU utilization may reach into mid-to-upper 70% range, and the fan spins up and gets rather loud. However, once the VMs load all their services, the Skull Canyon CPU utilization settles down and the fan speed drops to very low noise levels. In fact, with about 8-9 VMs, I cannot hear the CPU fan even if I get my ear within a few inches of the Skull Canyon NUC. I can only hear it if I put my ear on the top cover of the NUC. But, as I add 3-4 more VMs for a total of 11-12 VMs, the CPU utilization gets closer to 30%, and the fan becomes a little more audible. At 30% CPU utilization, the fan is probably at the 25% Duty Cycle, which generates noise that I can detect from within 1 foot. However, since I sit about 2.5 feet away from the equipment rack, this amount of fan noise is acceptable to me. I'm not sure if setting the "Processor Power Efficiency Policy" to "Low Power" was necessary, and I may try to set it back to "Balanced" in the future to see how this setting affects the fan noise. However, for my purposes, it appears that having this setting in "Low Power" provides enough compute power for my purposes.
Before I changed the Cooling settings in BIOS to the above values, the fan noise when running 10 VMs was so loud that I was about to pack the Skull Canyon NUC and return it to the store I bought it from.
In the KY units, it is a special heatpipe unit with factory-applied TIM. There is also the possibility of thermal pad(s) for additional IC(s) that are at different profile; this was the case on many previous NUCs -- but I am not pulling mine off to find out as I don't have any replacements and they are easily damaged!