Nvidia’s Pascal chips were and are much more reluctant to change loads than their predecessors with Kepler. And still we learn something new with these measurements. What exactly, I’ll show you now.
Measurements under maximum gaming load
Even though the average here is also around 250 watts, the absolute peak values are over 40 watts below those of the Radeon RX Vega64. But this is also due to the fact that the picture of spikes and valleys differs completely. You can see very well that Boost is much finer than Power Tune, whereby the power supply of the Founders Edition is not even really optimal and leaves some potential.
In this loop there is a maximum of 360 watts over 1.1 ms, which causes a sensation. Like also the approx. 375 watts within approx. 0.4 ms. You could also measure peaks of well over 400 watts if you increased the resolution again by a factor of 10. But if this doesn’t interest anyone any more because of the short duration, that’s why I leave it out here. Every power supply has to do that with its left hand, no matter if it’s a big brand or not.
The flows show the same picture, so I don’t want to repeat myself at this point. But if you remember what happened with your finger and the hot stove plate, you can see why Vega is a bit more difficult to cool, because the voltage converters have a lot more to do.
Measurements during the stress test
The intervals for the stress test are now suddenly much longer, because the Power Estimation can no longer play out its potential together with Boost. Where Pascal can still buy a little bit of the edge from Vega through a clever and much faster look ahead during gaming, there is now some sand in the gearbox under continuous load. It’s about 9 ms, which can be a peak load of about 300 watts. Even the 345 watts over 2.2 ms are almost at Vega level.
So in the end Nvidia boils only with water and before Furmark all are suddenly the same again. So the possible construction site for Vegas efficiency is not the clock gradation, because that is ok, but the frequency and speed with which an adjustment is foreseen and then carried out. Driver or firmware? AMD keeps a low profile as usual. But we are at Pascal right now.
Here, too, the power consumption is based on the measured currents times the applied voltage. If you look at both individually, it looks a bit better than with the RX Vega64, but also not better for worlds.
Interim conclusion
In the same power consumption class, Pascal is certainly one visible step ahead of the RX Vega when it comes to the load and its peaks in games. This advantage is lost under constant loads. What you have to consider when choosing and dimensioning a power supply is also a useful secondary side, which should produce at least about 350 watts over 10 ms and 400 watts at one millisecond to be completely safe. This is a bit less than with the RX Vega, but not so much better.
Kommentieren