Allgemein Basics Gaming GPUs Practice Reviews

Nvidia GeForce RTX 2080 Ti – Internal details about power supply, different components and where the spikes have remained!

What is smart here please?

Let's return briefly to the PWM controller. The uP9512 used on my sample has programmable output voltage and active voltage positioning functions to adjust the output voltage depending on the load current so that it is optimal for a good load current transition is positioned. The uP9512 also supports NVIDIA Open Voltage Regulator Type 4i+ with PWMVID function. The PWMVID input is buffered and filtered to create a very accurate reference voltage. The output voltage is then precisely controlled on the reference input.

And what makes such chips so special? It is the flexible hardware requirement to adjust the operating phase number in different load current states. This is not new in principle, but it has been significantly improved and should do exactly what Nvidia showed during the presentation on the slide. So in order to get such a smart control, it does not help to push the number of phases into infinity, but only to use and coordinate the 8 phases sensibly enough or to coordinate them. also switch off when not in use.

In the low-load range, it is therefore very easy to make the three non-double-occupied phases work in a very sensible and finely tuned manner. The fact that the power consumption in the idle for the entire board was still quite high must have other causes, such as: a still active NVLink controller, which pulls significantly too much juice despite the unconnected devices. Here, you should use driver or firmware update if necessary. (can) improve. But it probably has nothing to do with the provision of the VDDC.

In addition, soft start to avoid peaks, channel current limitation, undervoltage protection, overvoltage protection and power good output. The integrated SMBus interface offers enough flexibility to optimize performance and efficiency as well as to connect the right software. The controller also supports, and this is very important, the new Smart Power Stage (PLC) chips, as I found them on the sample.

The FDMF 3170 from ON Semiconductor is actually a PowerTrench® MOSFET and equivalent to the original fairchild and combines the high and low side, the Schottky diode and the gate drivers. But why Smart Power Stage (SPS) and not just Power Stage?  These PLCs then also provide very precise information about e.g. currents (IMON) and temperatures (TMON) high resolution and in real time! Now we also know why Nvidia does not rely on the conventional doublers, because the communication between PLC and PWM controller is extremely important!

 

Power supply killer spikes? But stop! Which spikes?

Most people know that I have been riding around on this topic for years and the basic article linked at the beginning is also one of the reasons why the power supply manufacturers are now much more sensitive to their secondary page and these almost already cuddling. I have now taken At my word to Nvidia, who spoke of significantly lower load changes during the presentation and measured again in the millisecond range!

First, I checked the frequency at which the individual phases of the VDCC are clocking. The GeForce GTX 1080 Ti and the RTX 2080 Ti are 300 KHz each. Apart from the fact that both values are identical, we can (a) exclude the influence of this value on the respective result in the following comparison and (b) we also see that the specs of both PWM controllers are total compared to the real circuit over-the-top. Gallery feeling, nothing more.

In order to remain really comparable in the test, I first overclocked a GeForce GTX 1080 Ti so that with the same 3D load (special Witcher 3 savegame) almost the same power consumption of almost 280 watts was expected. A power peak of 442 watts was not uncommon and you can also see on the picture how the whole construct is in "panic mode", only to be able to keep to the mean with frantic down-regulation.

We also see very nicely that the intervals are very granular and can be the plague for every power supply. Any further overclocking, by the way, would only make things worse. But it's really nice when someone listens and finally counteracts it. It doesn't matter to me why Nvidia implemented this, because it was long overdue! The measured almost 348 watts between the highest and lowest load state in gaming are simply unsustainable with the Pascal card.

Together with Boost 4 and the smart PWM controllers and PLC, the entire load can now be controlled much more beautifully. On the next picture you can see very well that the change intervals are now much smaller and the frequency of the rules increases with it. But you can also see that everything is now much more relaxed! The delivered power is also about 280 watts on average, but the average does not tell everything.

The power peaks are now only up to 377 watts, which is a whopping 65 watts less! And also the regulation down falls by approx. 50 watts lower. This means that the measured maximum difference between the two extremes is an impressive 111 watts lower than the previous one, at just 236 watts. This is already a small universe and the wellness program for every installed electricity supplier.

In addition to the very positive aspects of the current solution, there are currently only two problems. One is the slightly too high power consumption in the idle (I had measured a good 17 watts) and a very rare special case under extreme laboratory conditions, in which extremely high power peaks are said to have occurred. The former should be easy to solve, the latter is probably being investigated by Nvidia. But it is, in order to turn off false sensational messages at once, nothing that the Kevin normal gambler could ever trigger with his surroundings.

The OC guild is already hitting it much harder, that with the exception of special OC and LN2 BIOSe, everything that normally goes over the counter is throttled very restrictively. Nvidia limits the board power very hard and it is already a kind of built-in fun brake, if you get hardly significantly over 2 GHz despite Chiller, even though all values actually look good. There is simply no power, but it is likely to be due to a real voltage deficit. The current is only one factor, because for the OC you need the voltage. And that's where Nvidia set the bar. You will know why.

This is the end of the little excursion, but I am sure that there will certainly be one or the other. Promised!

Danke für die Spende



Du fandest, der Beitrag war interessant und möchtest uns unterstützen? Klasse!

Hier erfährst Du, wie: Hier spenden.

Hier kannst Du per PayPal spenden.

About the author

Igor Wallossek

Editor-in-chief and name-giver of igor'sLAB as the content successor of Tom's Hardware Germany, whose license was returned in June 2019 in order to better meet the qualitative demands of web content and challenges of new media such as YouTube with its own channel.

Computer nerd since 1983, audio freak since 1979 and pretty much open to anything with a plug or battery for over 50 years.

Follow Igor:
YouTube Facebook Instagram Twitter

Werbung

Werbung