Undervolt and Underclock in Notebook Gamer: more performance and less noise

Less voltage and lower clock can make your notebook much better!

One of the most striking features in a gaming notebook is its sound. When they’re being used to play games, many leave the environment with a “is it raining?” feel, a result of their fans with very narrow blades at high rotation. This is a result of high-performance chips in tight spaces and with little room for air circulation.

Another villain is the adjustment of the manufacturers themselves. In our tests, Intel has been showing a very aggressive behavior at frequencies, seeking to push clock increases to the limit, seeking maximum performance. This is good for delivering more performance, but on the other hand it makes many gaming notebooks operate close to the temperature limit, in this case 100°C on most modern Intel Core processors. Below is a good example of this happening across multiple models:

The big problem here is: Aggressive boosts are great for making your PC fast, especially for opening and loading things, but they make the chip reach temperature and power limits quickly. This means that in a high load duty cycle for long periods, the processor will reach its maximum temperature and will need to slow down, which can create situations like the screenshot below:

The Intel Core i7-10875H has a base clock of 2.3GHz, but it has a boost of up to 5.1GHz. Most of the test with Rainbow Six Siege, the benchmark runs at around 4.3GHz, but at this specific moment we see that the temperature peak forced the CPU to practically go back to base frequency. This is a classic example of Thermal Throttling, the moment when heating restrictions make the component need to reduce operating frequencies to return to acceptable temperature levels.

Did Intel go overboard on your boost? Can you do better? Today we’ll show you that the answer is yes, especially for “sustained” performance, that is, those high-performance scenarios for long periods, typical of those who need to render complex jobs or long gameplay sessions.

For the tests we will use Avell’s model, the LIV A72. We already have a full review of it on this link. Settings include:

– Nvidia GeForce RTX 3060 6GB GDDR6
– Intel Core i7-10870H
– 15.6″ IPS 165Hz QuadHD Screen (2560×1440)
– 2x16GB DDR4 @2666MHz
– 512GB SSD M.2
– 355 x 236 x 19 mm
– 1.7kg


– Continues after advertising –

adjusting

To change the settings we will use the ThrottleStop software, available for free download through this link. Always good to remember: changing your processor’s operating settings outside the factory specifications can cause instabilities and are also not covered by the warranty, so proceed with caution. In general, the procedures we will do are not critical: in fact we will even reduce power and frequencies.

ThrottleStop Download Link

One of the procedures we will do is undervolt, which is when you reduce the operating voltage on the processor. To achieve more performance and higher frequencies, a chip needs higher voltages, but this impacts stability and heat. Here comes an important element: different chips operate in different ways, even though in theory the same processor model. We cover more of that subject in this other article here.

Intel works with safe thresholds, which includes putting a voltage with a margin to ensure the chip is capable of achieving performance. Virtually every chip has the potential to run stable at a lower voltage, but each will behave differently. Then there’s no other way: just testing yours to find out where the stable adjustment is. After a few rounds of testing, we were able to stabilize the 10875H with an offset of up to -102.5mV. in processing cores (CPU Core) and 102.5mV in cache memories (CPU cache).

The other adjustment was for frequencies. The performance doesn’t evolve in a linear way: this means that if we reduce the frequencies, the performance doesn’t follow at the same pace. Below is an example of this in CineBench:


– Continues after advertising –

It is visible that the evolution curve gets worse after 3 to 3.5GHz. The reason here is that the processor can’t even keep all cores constantly at frequencies above 3.5GHz. Despite still managing to extract a little more performance and give more margin, the gains keep getting worse and worse until we reach the point where performance drops, in the case of the 4.5GHz boost on all cores – the default setting was 4.3GHz.

It may seem counter-intuitive, after all among enthusiasts we usually want to increase the operating frequency (overclock) to get more performance. However, due to the instability created by thermal throttling, lowering frequencies (underclock or downclock) can, paradoxically, be better for performance over long periods of time!

With that in mind, we’ve reduced the boost’s aggressiveness. The maximum has been reduced from 5.1GHz to 4.8GHz, and the turbo across all cores has gone from 4.3GHz to 4.2GHz. You can check each of the adjustments in the screenshot below.

So what is better? The factory setting or our change? Time to see the graphics!

Tests

CineBench uses the Cinema4D graphics engine, an important 3D rendering tool. This test is extremely CPU-heavy and loves lots of cores and threads. In it it is evident that our setup has inferior results in single-thread scenarios, something obvious considering that we reduced the boost frequencies, but in contrast all cores in action see the opposite. Lowering frequencies improves performance. Here we have the first clear evidence of throttling the system in its factory configuration.

The undervolt was ahead, but the margin of only 2% is far from being relevant and could be a margin of error. Interestingly, reducing the frequency and voltage did not impact performance.

With 3DMark we have two draw results, with irrelevant variations. In this case, victory again for our adjustment, after all it consumes and heats up less, as we will comment shortly.

Games show us an amazing scenario. While Red Dead Redemption 2 has a tie, Rainbow Six Siege and even Counter Strike showed performance gains above the 5% margin of error. This is unexpected considering that, especially CSGO, is a game that benefits from high frequencies rather than large amounts of threads. While we were waiting for a repeat of what happened at CineBench in single-thread, again our more moderate clock turned out better for the final average than higher voltages and more aggressive boosts.

And finally, our last test checks consumption and heating. The decrease in heat on the processor is visible, something that contributed to the improvement in performance in several of the tests, as it enabled more efficient sustained performance. But another factor also comes into play: the heating of the GPU. With parts of the shared heatpipes and fans structures, reducing heat on the processor gives more “cooling budget” for the graphics chip. You can see the structure of this Avell model in the video below:

In practice, in the Rainbow Six Siege test, the RTX 3060 averaged at 1850MHz at factory setting, rising to 1950MHz after our undervolt. Here it is good to remember that we did not touch any configuration of the GeForce chip. It is automatically taking advantage of the larger margin available in cooling on its own through GPU Boost 3.0.

I must do?

Based on the results displayed in the article, plus a battery of tests playing with other settings, it is apparent how undervolt and underclock can bring great benefits to an Intel Core-based gaming notebook. Even if performance was slightly degraded, we would still recommend tuning just for the reduction in heat and noise production. But this is not the case, on the contrary! In several the performance was the same, in some they surprisingly improved, and only a few had an inferior performance.

This shows how Intel’s aggressive boosts strategy can even have positive results on ultra-thin devices or more localized actions, such as loading an app. However, in more prolonged stress situations, such as rendering applications or games, this strategy of using any and all energy and heating margins is more harmful than beneficial to the consumer.

Here it is also good to make a distinction between the results we use throughout the article. So far, the Ryzen gaming notebook models we’ve tested do much better in warming up, but the latest one uses the 3000H series. AMD has greatly increased operating frequencies on the 5000 series, and only in this generation has it “equaled the game” with rival Intel. We will need to test this new generation to see if the company will do better or end up in the same problem as “the blue side of the force”.