Temperature curve and clock rate
The fan control is quite conservative, so that the maximum temperature of 85°C is reached relatively quickly. But then the card already has approx. 6% of their performance was lost from the cold state, which was almost exclusively due to an automatic reduction of the clock rate by approx. 9%. We calculated an average clock frequency in 5-degree increments, which resulted from all the ups and downs of the respective run. This ranged from 1533 MHz in cool conditions down to 1401 MHz in the worst and hottest case.
Temperature curve and power consumption
Now it's getting really interesting. At temperature-related 1401 MHz, we measure on average approx. 285 watts, at 1533 MHz cold around 310 watts. For 9% more clock, we also need a good 9% more supplied power and receive approx. 6% more gaming performance. So the deal is acceptable at first, but it also shows that the efficiency curve is already falling sharply. So there is not much more room for manoeuvre.
But what can also be said is that leakage losses obviously no longer play a major role. Times when you could easily save around 40 watts or more at the same clock, if the temperature remains low enough, should be a thing of the past. At least something.
Temperature history GPU vs. HBM2 Memory Modules
As far as the read-out values are correct, we can assume a maximum of 84 (85°C peak) for the GPU and a maximum of 90°C (94°C Watt Peak in the Torture test) for the HBM2 modules. The latter appears relatively high, but is also considered the upper limit for the current GDDRX5 memory. Of course, we will always keep an eye on these values during further tests, because we cannot currently confirm the accuracy of the sensor interpretation 100 percent.
During the stress test, the heating occurs so fast that the curves for the open and closed structure are closer together.
Temperature gradients on the board ("heat flux")
What we can immediately determine: the board below the base is approx. 5°C cooler than the values we were able to determine within the GPU! But what is the reason for this? The explanation was already on page two, because we were able to find a very thick package board, which is still located between the interposer and the PCB. In addition, the interposer is obviously not fully on the package ("underfill issue"), so that the air in between almost acts like an insulation layer.
In the stress test, the temperature is slightly lower, which is probably also due to the increased fan activity and the regulation by the lowered clock (see power consumption).
Noise emission ("volume")
The measured 48.2 dB(A) is the maximum value in balanced mode, but we have already made it comfortably over the 50 dB(A) mark. What was still cause for praise at the Vega Frontier Edition is now instead of a quiet evening wind already a decent hurricane, with which one can let one's ears blow nonchalantly and which is simply due to quite high power consumption. However, a GeForce GTX 1080 Ti FE with a maximum power target of 295 watts is not significantly quieter.
The version of the cooling presented here differs significantly from what we liked with the Frontier Edition. Too aggressive, too hot and, logically, too loud in the end. Just on Edge has never been a truly harmonious marriage.
- 1 - Einführung und Übersicht
- 2 - Details zu Architektur und HBM2-Speicher
- 3 - Demontage, Kühler und Interposer-Details
- 4 - Platinendesign und Detailinformationen
- 5 - Ashes of the Singularity: Escalation
- 6 - Battlefield 1
- 7 - Warhammer 40,000: Dawn of War III
- 8 - Doom (2016)
- 9 - Tom Clancy's Ghost Recon Wildlands
- 10 - Hitman (2016)
- 11 - Metro Last Light (Redux)
- 12 - Rise of the Tomb Raider
- 13 - Tom Clancy's The Division
- 14 - The Witcher 3
- 15 - Und kann sie Mining?
- 16 - Leistungsaufnahme im Detail
- 17 - Takt, Temperaturen und Geräuschemission
- 18 - Zusammenfassung und Fazit
Kommentieren