What hasn't been the noise in the leaf and video forest in recent weeks. There was even talk of a VRM disaster or a TIM gate (heat paste instead of lot), but if you look at everything carefully and carefully, the whole thing is nothing more than a long causal chain, in which an extremely hot-headed CPU is the actual Starting point. But what do we write one, in principle it is even the same as the three. But always nice in turn, because we try to keep it as easy as possible.
A brief stocktaking at the beginning can't hurt, so we'll just briefly summarize the two key points that keep spilling over in the current discussions and after which we've structured our test:
(1) Skylake-X is hardly coolable even in normal operation out-of-the-box, as the power consumption in individual situations is already extremely high and the thermal paste additionally prevents optimal and appropriate heat dissipation.
(2) There is hardly any overclocking room for the normal user and many motherboards are due to design defects, such as insufficient cooling of the external voltage converters, already unreasonably limiting the CPU on its/her/her part. Extremely overclockers can hardly do anything with the current hardware.
The whole thing is then usually seasoned with far too much polemic, which, however, is perhaps even the real problems of potential buyers.
Test setup and measurement methods
That's why we grab one of the simpler motherboards for the 2066 socket, build a new benchtable for vertical operation and test what's (or maybe not). On the one hand, we will discuss the sensor values of the respective areas, as well as their origin, and on the other hand, by means of our non-contact measurements with the infrared thermal imaging camera, the heating of the circuit board in the area around socket and voltage converter to plausibility Check.
In addition, we can even document the warm-up and warm-up, as well as display the process in special time-lapse videos. After all, we are also interested in whether other components may also be caused by the resulting hotspots or heat transfer in the motherboard could be negatively affected.
For the safe reading of the sensors and the smoothest possible operation of the test setup, we use the latest BIOS for our motherboard, as well as HWinfo in the current latest beta version from v5.53-3190 (click on beta version when downloading!).
[Update vom 11.07.2017]After closer inspection, the consultation with the manufacturer and also hints from the forum, we have corrected some details below or Deepened.
The board has a total of 5 + 1 phases for CPU power supply, which are provided and controlled by an IR35201 from International Rectifier. This multi-phase Buck controller supports Intel's VR12 and obviously VR13. Anyone who suspects more voltage converter circles on the basis of the coils for Vccin is absolutely right, because by means of the so-called doubling, two circles can be realized at five phases per phase and thus also relieve the individual VRM and equalize the hotspots in terms of area. We will come back to this chip, as well as its tangible data such as voltages and currents, later.
An IR3555 from International Rectifier serves as voltage converter sandals per control circuit. These highly integrated power-stage chips combine the necessary gate drivers, synchronous MOSFETs for the high and low sides, and the Schottky diode in one package. In addition, unlike most of the usual MOSFETs, they also have integrated analog temperature sensors. But how else can you determine the temperatures of these voltage converters exactly if you can't use a suitable IR camera?
On the tested motherboard, MSI uses the Nuvoton NC6795D as a so-called Super-IO chip that can capture and provide a variety of sensor values. This also includes the temperature value of the voltage converters, which is determined by means of a thermistor placed in the middle between the power stage chips (picture below). We have therefore chosen the measuring point for our video recording at the back just below this thermistor.
In addition, we also check the temperatures of the coils and condesors of these voltage transformer circuits, as well as the board temperatures up to the CPU.
Forced shutdown and emergency shutdown
In order to better understand the further tests, as well as the problems that have occurred and often discussed far too polemically in forums, we need to know that the motherboard manufacturers use some security mechanisms. This includes, for example, that our testboard clocks down the Skylake-X at exactly 105°C measured temperature at the thermistor (HWinfo under line MOS, Nuvoton NCT6795D) to 1.2 GHz and maintains this state until the temperature has dropped to below 90°C. Then again, there is full throttle.
This makes perfect sense when one knows that although the flash point for the board material used (FR4) is significantly higher, the recommendations for a maximum temperature in continuous operation are only at values between 95 and 105°C, because the recommendations for a maximum temperature in continuous operation are only between 95 and 105°C, because the recommendations for a maximum temperature in continuous operation are only between 95 and 105°C, because the recommendations for a maximum temperature in continuous operation are only between 95 and 105°C, because the recommendations for a maximum temperature in continuous operation are only between 95 and 105°C, because the recommendations for a maximum temperature in continuous operation are only between 95 and 105°C, because the recommendation Multilayer boards could otherwise be affected by dry-out, bending and possible hair cracks of the traces. This is to be welcomed, because graphics card manufacturers usually have the (unnecessarily) better nerves in this problem zone.
If you use Intel's Extreme Tuning Utility (XTU), you'll get this downclocking displayed as Thermal Throttling: Yes in yellow. But what about status indicators such as Motherboard VR throttling?
[Update vom 11.07.2017]Here, too, we need to add a small addition that concerns the values read out by HWInfo. What is far less well known is that the IR35201 also returns temperature values. These values for VR T1 and VR T2 are significantly higher and at first glance quite contradictory to those of the external sensor.
It was obvious at first that here, as so often, only a kind of chip temperature of the controller is output. This would then be roughly equivalent to what you would e.g. for graphics cards with these PWM controllers in various tools then supposedly displayed as voltage converter temperatures VRM1 and VRM2 (mostly AMD cards put on these controllers). In fact, the chip almost always measured itself.
However, the combination of IR35201 and ir3555 occurring here assumes that the voltage values provided by the IR3555 and in relation to the temperature inside are also used.
For these values, an upper limit of 125°C is set, before in the XTU yellow before motherboard VR throttling: yes is warned and the CPU is also clocked down to 1.2 GHz. From 135°C, the motherboard is simply switched off without warning, because otherwise the generated voltages drift dangerously outside the specifications and could damage the hardware.
Weree exemplary for a CPU core
But the CPU also protects itself. Based on various integrated digital temperature sensors (DTS), the temperatures of the computing cores and the package are determined. These are computational values whose accuracy increases with increasing temperature. Everything below 40°C can actually be forgotten, starting at approx. 80°C, the area where it matters, it becomes quite accurate. However, we also see that both the temperatures of the cores and those of the package can lead to a thermally-induced throttling of the clock.
Especially the package temperatures also include the power dissipation of the IVR, i.e. the voltage converter integrated in the CPU for the provision of the individual partial voltages of a CPU. Here, especially with high overclocking and manual voltage increase, unexpected limit overruns can occur quickly, which not every tool can detect equally securely. Then the CPU throttens without the user being able to see the cause. But there's more to the IVR. First of all, however, let us sum up once again:
Fact sheet #1
A clock reduction of the CPU can be reduced by both too high core and package temperatures within the CPU (which is best known), as well as from the Super-IO chip due to the excessively high VRM temperatures or the PWM controller due to too high, own chip temperature and the dangers of an unstable power supply. But the fact that the PWM controller can deliver VRM temperatures is an urban legend.
The test system in detail
We have summarized all the hardware components, measuring instruments and tools used once again in tabular terms:
|Test system and measuring instruments|
Intel Core i9-7900X
MSI X299 Gaming Pro Carbon AC
4x 4 GB G.Skill RipJaws IV DDR4-2600
Nvidia Quadro P6000 (Workstation)
1x 1 TByte Toshiba OCZ RD400 (M.2, System SSD)
||Alphacool Ice Age 2000 Chiller + Alphacool Ice Block XPX
Alphacool Polar Bear 240 (AiO)
Noctua NH-D15 (Air)
Thermal Grizzly Kryonaut (for cooler change)
||Direct DC measurement via shunts (clamping case)
Direct DC voltage measurement at the measuring points
Non-contact DC measurement at the EPS supply port
2x Rohde & Schwarz HMO 3054, 500 MHz multi-channel oscillograph with memory function
||Optris PI640, infrared camera
PI Connect evaluation software with profiles
Still images and radiometric videos