Cooling with the Chiller crowbar
In order to be able to achieve usable (overclocking) results, we had to switch from the normal water cooling to the Alphacool Ice Age Chiller 2000, as already mentioned in the previous chapter. Trials with AiO compact water cooling systems such as a Corsair H100i and the Enermax LiqTech 240 were already running into the thermal limit at the standard clock and Prime95, while the custom loop cooling later failed at 4.6 GHz.
But stop! Normally, should such cooling solutions be able to cool such a CPU sufficiently? After all, we were able to operate the Core i7-5960X at 4.8 GHz and a normal AiO compact water cooling system, although even then up to 250 watts of power consumption were recorded. In the end, however, we were simply forced to force a constant 20°C in the cooling circuit in order to be able to do any experiments at all!
High delta values as a cooling trap
The real reason is Intel's incomprehensible fun brake in the form of inappropriate (but arguably much cheaper) thermal paste instead of a sensible solder. However, since the DTS (Digital Temperature Sensors) at Intel only started at approx. 35 to 40°C provide reasonably resilient values, we have changed our view in this way and only recorded values above in the first evaluation. In order to be able to correctly classify the misery with the thermal paste, we now show the temperature differences between the constant 20°C cool water block and what we have determined as CPU temperature from the sensor values.
The following curve shows extremely clearly that the waste heat can only be dissipated poorly and insufficiently. What just worked like this with a thermal 100-watt bread roll like the Core i7-7700K, however, now leads the paste strategy into the absurd. We also measured the temperature of the heat spreader using our own, very thin copper plate, analogous to the Ryzen launch article, whereby the delta determined here later flows into our individual curve.
At the end, the following curve thus represents the temperature differences between the heatspreader top and the calculation cores, whereby the results have shocked us:
Although with our Chiller, the Alphacool XPX water block and the Thermal Grizzly Kryonaut thermal paste we have used pretty much the best and most expensive that the market has to offer, at the end there are a whopping 71 Kelvin as the temperature difference between the temperature of the cores and top of the heatspreader on the bill! This is especially annoying in that it makes normal cooling solutions look quite silly at full load and factory cycle. AMD has impressively demonstrated how to better disprove such power consumption values with Ryzen 7, even if we were upset with the artificially offset output values of the X-models.
In order to illustrate our incomprehension, we now have the temperature gradients in detail, as they are out-of-the-box with Prime95 or Prime95. Luxrender, packaged in a diagram. It is reasonably logical that the Chiller or a proper conventional custom loop cooling system still work well here, but any other cooling solution is already being pushed to the limits with these loads. Even the motherboard manufacturers have told us about the onset of the aiO compact water coolers used when Prime95 is released without an AVX choke.
Up to 65°C Tcore at a heatspreader temperature of approx. 24°C already results in a delta of more than 40°C. We reached this value with a power dissipation of just under 230 watts. However, as soon as you move above the 300-watt limit, which you can even use with simpler rendering programs from approx. 4.6 or 4.7 GHz and the necessary voltages (depending on the chip quality), hardly anything goes with the Chiller. With the maximum values of just over 300 watts we have achieved, the CPU is already running permanently thermal limit of 100°C, shortly afterwards there is a logical shutdown.
Leakage
We measured the power consumption values with identical load and different cooling solutions, whereby we have of course set physical limits here. The increase in leakage currents at higher temperatures is kept within a very manageable framework, as shown by the next curve. The power consumption increases by a total of 5% when the core temperatures increase by approx. 40 Kelvin. This is more than just acceptable, but really good. The measured values above approx. 95°C, however, are already somewhat inaccurate due to the onset of throttling. For once, we worked with a low-pass filter that smooths the short break-ins to some extent.
Intermediate conclusion
Yes, it could have been so beautiful if it wasn't for the trouble with the thermal paste between heat spreader and Die. Well, the normal user in the semi-professional area will hardly overclock the CPU, which is likely to significantly reduce the mourning community. But even reasonable users of the Core i9-7900X will not be able to avoid investing at least in a clever cooling. Whether you then use a really good AiO compact water cooling or better use a proper water cooling, let's face it. Air cooling is eliminated in full-load operation, especially in summer.
- 1 - Einführung und Übersicht
- 2 - Intels Fabric - Mesh statt Ringbus
- 3 - Cache und Latenzen, IPC, AVX und Kryptographie
- 4 - Chipsatz, Testsystem und -methoden
- 5 - Problemanalyse mit Civilization VI und VRMark
- 6 - AotS Escalation, Battlefield 1, Deus Ex: Mankind
- 7 - GTA V, Hitman, Shadow of Mordor
- 8 - Project Cars, Rise of the Tomb Raider, The Division
- 9 - Workstation und HPC
- 10 - Leistungsaufnahme und Übertaktung
- 11 - Temperaturverläufe und thermische Probleme
- 12 - Zusammenfassung und Fazit
Kommentieren