Home / Technology / Undervoltage of NVIDIA Pascal and Turing mobile graphics cards

Undervoltage of NVIDIA Pascal and Turing mobile graphics cards



  Undervolting of NVIDIA Pascal and Turing mobile graphics cards Undervolting is one of the most interesting things you can do, for example on a laptop or graphics card, and thanks to which we can quickly and easily improve the fan's working culture (at stationary GPUs) or reduce the component's temperature (this applies mainly to laptops). Undervolting of me personally is used very much, because for more than two years it is part of the game table testing. During this time I checked a few dozen models, including laptops with NVIDIA GeForce GTX 10×0 graphics card and newer GeForce RTX 20

5;0. On this basis, I have already managed to form a certain opinion on how the undervoltage of graphics cards on laptops really works and which generation is more susceptible to it.

In PurePC, I tested dozens of different laptops with NVIDIA GeForce GTX 10×0 and GeForce RTX 20×0 cards for intrusive vulnerabilities. From the tests performed, it can be concluded that Pascal systems are more susceptible to voltage reduction.

  Undervoltage of Mobile Video Cards by NVIDIA Pascal and Turing [3]

What exactly is the undervoltage process for a graphics card? In the case of graphics cards belonging to the Pascal or Turing architecture from NVIDIA, the third and fourth generation of Boost GPUs are crucial, allowing you to easily change the voltage and timetables. The program that enables such changes is, for example, MSI Afterburner and I always use it myself, mainly because of easy access to the voltage and timetable, and the quick opportunity to save selected settings. After activating the program, just press the "Left Ctrl + F" key combination. While keeping the factory settings and times in the tested laptops, the values ​​stored in the voltage table are variable and when the core temperature is too high, they are lowered dynamically so that the clock in the GPU Boost 3.0 / GPU Boost 4.0 often falls to the lowest specified level. In the vast majority of cases, small changes in the voltage table showed that clocking in the GPU Boost 3.0 or GPU Boost 4.0 was much more stable, and getting a similar value for timing at reduced voltage resulted in lowering the temperature of the mobile graphics card.

  Undervoltage of Mobile Video Cards NVIDIA Pascal and Turing [7]

  Undervoltage of NVIDIA Pascal and Turing Mobile Video Cards [2]
Sample voltage settings in the voltage and clock table for NVIDIA GeForce GTX 1080 (top) and GeForce RTX 2080 (lower) respectively.

Test ASUS G703GX – an impressive laptop with GeForce RTX 2080

By continuing with examples of the prime mobile graphics cards belonging to the architecture of Pascal and Turing, we will explain how the cards functioned before and after the undervoltage process. For the GeForce GTX 1080 in the SMART7 H773F notebook (which, of course, was tested in PurePC for undervoltage), the pre-UV clock rate was about 1830 MHz at a voltage of 1000 to 1050 mV. At that time, the goal was to lower the voltage of the graphical core so that the same time (or as close as possible) could be achieved at a lower voltage. It turned out that the mobile GeForce GTX 1080 is very susceptible to this type of operation and already at a voltage of 900 mV worked with a very similar clock frequency of 1820 to 1830 MHz. Achieving the same performance at lower voltage naturally resulted in a lower temperature. For this graphics card, the difference was 6 degrees, but there are cases of major differences. It is obvious that different graphics cards can react differently, but over the past two years I have noticed that cards with the same specification, but placed on different laptops, have similar results after voltage reduction.

  Undervoltage of NVIDIA Pascal mobile graphics cards In Turing [4]

Test Dream Machines RX2070 – RTX 2070 more efficient than GTX 1080?

The second example is NVIDIA GeForce RTX 2080, in its full version which is very important in terms of undervolting. Here I use the ASUS ROG G703GX notebook. During the tests, the core point was about 1890-1900 MHz at a voltage of 1000 to 1025 mV. During the voltage reduction (which is performed for the Turing family cards in the same way as for Pascali), we managed to achieve clock speeds of 1880 MHz at a voltage of up to 925 mV. Such a large difference also significantly affects the final result, since the temperature dropped almost ten degrees, since almost identical core time was maintained. However, it cannot be concealed that, although the Pascal mobile cards are concerned, all three of the highest GPU models always had very clear differences before and after the undervoltage process. So we're talking about NVIDIA GeForce GTX 1060, GeForce GTX 1070 and GeForce GTX 1080 card. This was largely due to the fact that their watches were slightly lower compared to desk equivalents. All three systems rise into the Boost 3.0 GPU mode. Even though the TDP coefficients were reduced, the differences in the relationship between desktop and mobile were not as great as in the Turing architecture.

  Undervoltage of NVIDIA Pascal and Turing mobile graphics cards [5]

MSI test GE75 Raider 8SE – RTX 2060 more efficient than GTX 1070?

Generally, GeForce GTX 1060, GeForce GTX 1070 and GeForce GTX 1080 were equipped with fewer CUDA cores, fewer TMU and ROP, and the graphics card was much smaller than Turing. By adding a new technological process (as in the launch) (16 nm instead of the worn-out 28 nm), it was not necessary to significantly reduce clock clocking and TDP. This, in turn, led to the fact that the cards during the game achieved quite high real clock speeds, often over 1700-1800 MHz, which for laptops was previously impossible to accomplish. The cards, in spite of all, had relatively high voltages, so that the undervoltage had a greater influence especially on the temperature. In turn, the NVIDIA GeForce RTX 20×0 graphics card is a completely different creation and this undervoltage in laptops is definitely less likely to provide real benefits. I have a very similar problem with processors, because the increased number of physical cores has meant that reduction of the voltage, e.g. for the Coffee Lake-H family, does not provide such measurable benefits as those at Kaby Lake-H. Although they are 4-core and 8-wire undervoltage systems, not only does the timing stabilize, but also often significantly reduce the temperature. In the case of Coffee Lake-H, it is rarely possible to lower the temperature after undervoltage. The biggest advantage here is just stabilizing the timing. So I suspect that when it comes to 8-core and 16-wire systems, lowering the voltage will give the user even less than before.

  Undervoltage of NVIDIA Pascal and Turing graphics cards [6]

Test Dream Machines RG2060 – RTX 2060 at a good price

But back to NVIDIA GeForce RTX 20×0 mobile card. You cannot conceal that all three systems, GeForce RTX 2060, GeForce RTX 2070 and GeForce RTX 2080 have very large graphics chips. Turing TU106 graphics core has an area of ​​445 mm², while Turing TU104 has a surface area of ​​545 mm². Compared to Pascal GP106 (200 mm²) and Pascal GP104 (314 mm²) cores, these are very large differences which, if they do not want to, must affect the larger cuts in the core and TDP ratio. In fact, only the GeForce RTX 2080 in its full version is closest to the desktop layout, but is therefore only mounted on the largest DTR notebooks with the most advanced cooling systems. Only with this graphics card undervoltage gives tangible benefits, which I also presented on the example of the ASUS ROG G703GX. The GeForce RTX 2070 is already operating at much lower voltages, resulting in a relatively low clock speed in the Boost 4.0 GPU, which reaches the 1600-1650 MHz level. After all, this card can still be reduced by about 5 degrees graphics card temperature, if we correctly perform the undervoltage process. For all other cards, ie the GeForce RTX 2060 and all Max-Q Design systems, the voltage does not reduce any benefits. I have tested three laptops with GeForce RTX 2060 and two with GeForce RTX 2080 Max-Q Design. In any case, it was not possible to lower the temperature or significantly affect the power consumption. All these cards work at low clock speeds, especially for GeForce RTX 2060.

  Undervoltage of NVIDIA Pascal and Turing Mobile Graphics Cards [8]
NVIDIA GeForce RTX 2060 Mobile, not Max-Q Design, has been significantly reduced

Test MSI GL63 8SE – the cheapest laptop with GeForce RTX 2060

Temperature GPU

Sleep / Charge

Degrees Celsius (less = better)

ASUS ROG G703GX
Core I7-8750H , RTX 2080

MSI GE75 Raider 8SE
Core I7-8750H, RTX 2060

Clevo P775DM3-G
Core I5-6600K, GTX 1070

MSI GE75 Raider 8SE
Core i7- 8750H, RTX 2060

ASUS ROG G703GX
Core I7-8750H, RTX 2080

Hyperbook Pulsar Z15
Core i7-8750H, GTX 1060

Clevo P775DM3 -G
Core i5-6600K , GTX 1070

Dream machines RX2070
Core i7-9700K, RTX 2070

Hyperbook Pulsar Z15 [19659027] Core i7-8750H, GTX 1060

Dream machines RX2070
Core i7-9700K, RTX 2070

SMART7 H773F
Core i7-7700, GTX 1080 [19659028] SMART7 H773F

Core i7-7700 , GTX 1080

Although the GeForce RTX 2060 is not called Max-Q Design, for some reason this layout is among the three most common. Very low clock speed, which in games cannot reach a ceiling higher than 1500 MHz (if the system is not overclocked) is also the low voltage that these systems work on. Undervolting in this case gives nothing, and the same goes for Max-Q Design cards, which are energy-saving versions of full, mobile Turing systems. An important issue that arises, however, is whether the lack of real impact of undervoltage, such as lower temperatures, is actually a problem with the GeForce RTX 20×0 system. In the case of Pascali undervoltage, it was more affected by temperatures, but on the other hand, they were often higher on the default settings compared to the new GeForce RTX cards. In the case of Turing, the situation is reversed – I have not yet encountered a model that would have any problems with too high temperatures. The values ​​in the order of 70-75 degrees are somewhat normal here. The exception is basically the aforementioned ASUS ROG G703GX, whose card has already reached low temperatures in the default settings of 72 degrees, but after UV they were very low (for a laptop) and suddenly reached 63 degrees. Full, mobile GeForce RTX 2080 is the only one of the new high-voltage cards (even though it is still lower than in stationary versions of the GPU) and only here are the differences visible to the naked eye. The lower we go down with the cards, the differences before and after the UV will be blurred until we finally get to the place where these differences will simply not be.

At present, there is no point in implementing NVIDIA GeForce RTX mobile graphics card 2070, RTX 2060 and energy-saving versions of Max-Q Design undervoltage procedures. Due to the low voltage of these short and low core times, further voltage reduction will not affect in any way lower temperatures or better working culture.

  Undervoltage of Mobile Video Cards NVIDIA Pascal and Turing [9]

Undervoltage is a very useful thing, especially in relation to mobile graphics cards from the Pascal family. Here I had virtually no situation where lowering the voltage would not bring visible benefits such as lower temperatures, a somewhat quieter cooling system or extended battery life. For this reason, I have highly and openly recommended using this procedure on laptops with GeForce GTX 10×0 cards. As far as Turing is concerned, the situation is a bit different, as it is basically just for the best notebooks DTR with the highest GeForce RTX 2080 chip I can fully recommend doing so. Only here are the differences so large and visible to the naked eye that it is actually worth performing an undervoltage graphics card. For all other notebooks with lower cards, this process will be virtually useless. You do not only have to lose time, and you will not feel any positive effects. The temperatures do not decrease (unless it will be symbolic), the working culture will not change at all, and a slightly longer battery life will be more related to the CPU voltage reduction than to the graphics card. Although I am a total pursuer of undervolting, it would currently sell laptops with the GeForce RTX 20×0 card recommended to lower the voltage only for processors, where it still has a real effect on stabilizing the timing during a long-lasting load. Well, if we don't have really expensive equipment with the RTX 2080, I also recommend the UV graphics card. I'm very curious, how possible undervolting will affect the mobile cards that recently debuted – of course the GeForce GTX 1650 and GTX 1660 Ti. Soon the test will give me the answer, but I suspect that the differences will be more noticeable here than with the RTX 2060 or even the RTX 2070.


Source link