How much does the performance of your processor change?

Today we are going to experiment with the processor. How much do high frequencies and the large core count influence the performance of this processor in games and professional applications? Let’s solve this question by testing every type of configuration possible!

Let’s take the powerful AMD Ryzen 9 7950X, with 16 cores and 32 threads and frequencies surpassing the 5GHz barrier, and play with its operation to see how much this impacts performance, energy consumption and heating!

Our test bench includes the following components:

• AMD Ryzen 9 7950X
• 2x16GB DDR5-5200 Kingston
• MSI MEG X670e motherboard
• Liquid Cooling Cooler Master ML360 Flux
• Nvidia GeForce RTX 4090 Founders Edition
• Cooler Master v850 source

We use the AMD Ryzen Master software to turn off cores, change frequencies and also disable Simultaneous Multi Thread (SMT). The processor voltage was left at the automatic setting.

In the heating and consumption tests we used the CineBENCH benchmark to stress the CPU as much as possible. To maintain regularity in the tests, the liquid cooling fans were set at 50% rotation and the pump at 100%. Our overclock to 5.5GHz resulted in a system crash in CineBENCH, hence the values ​​zeroed at that level.

We started by tinkering with the processor frequency, starting from the minimum available, which is 600MHz, progressively increasing to an overclock by trying all cores at 5.5GHz, a very high and unstable frequency for a 7950X.

We can separate the behavior of software into two groups. CineBENCH, Counter Strike 2 and 3DMark Fire Strike had a very linear behavior, progressively gaining performance. Or, in the case of Fire Strike, from 2000MHz onwards when it starts running, since below that the test doesn’t even run.

The second group includes Call of Duty Modern Warfare III and Cyberpunk 2077. These two games saw performance jumps after a certain value (1500MHz in Cyberpunk, and 3500MHz in COD). After this critical value, the performance gain becomes more linear again.

Do more cores, more threads make a difference?

We ran our core count tests in two different modes. With Simultaneons Multithread turned on and also turned off. SMT is the technology that creates two threads per physical processing core, using idle parts of the CPU core to perform a second job, that is, creating the second thread.

Here we have a very big difference in evolution depending on the type of application. In games, scaling happened up to a certain number of cores, and then the gains are minimal.

In Cyberpunk 2077, Call of Duty and in the 3DMark test, we have a big increase up to six cores, and gains up to the eighth. Above that, performance is identical or even slightly lower than in octa-core mode. In Counter Strike, the curve trended even earlier, stopping at six cores.

But there are applications that scale continuously, and CineBENCH is a good example. It scales so linearly that it seems like any semiconductor engineer’s deepest dream. Performance practically multiplies in the same ballad as the number of cores are increased. This is a pattern that occurs in specific software, often for rendering, virtualization or other professional applications.

What about the SMT ON and OFF dispute? We don’t have very relevant benefits in games using these techniques, with low variations in the best results with one configuration or the other. What has changed is that now games scale up to 16 enabled cores, showing that in reality the evolution curve did not end with 8 enabled cores, but rather with 16 threads.

Another interesting fact is to see that 6 cores and 12 threads and 8 cores and 8 threads are in an even dispute, as well as 4 cores and 8 threads vs 6 cores and 6 threads. This ends up being more of a curiosity, since this aberration of selling processors without multithreading per performance core was limited to just a few older generation Intel Core processors.

Consumption and temperature versus clocks

Looking at the performance trend versus temperature as we change the processor clock, we see that performance increases at a faster rate than temperature. But this graph has a catch: the temperature of the Ryzen 9 7950X is 95ºC.

In other words, this is the thermal operating ceiling, and it is not that the heating increases more slowly as we increase the frequency. What happens here is that we are extrapolating the maximum possible operating value, and it is no wonder that our 5.5GHz overclock resulted in a system crash. Overheating may have been one of the factors that led to the instability.

Now comparing the increase in performance and consumption as we increase the frequency, it is interesting to see how performance was increasing at a higher rate than consumption until around 4GHz, when the two trends start to become more similar. We were unable to see the trend above 5GHz because of the system crash.

Our OC is much less efficient than letting the 7950X’s frequencies vary, something that is evident by the new balance between consumption and performance in the processor’s factory configuration (stock).

Consumption and temperature versus cores

When we compare the consumption and performance gain curves as we increase the number of cores, we have a good example of why servers set out to configure so many cores on just one CPU.

The trend of performance gains scales faster than the progression of increased energy consumption. This shows how functions that scale well in multithreading have the potential to extract more and more efficiency as we concentrate more cores on a single processor.

In terms of temperature evolution, we again have the effect of the Ryzen operating heating limit, and Turbo Boost is also in action here. This technology accelerates frequencies as we have a greater thermal margin.

The result is that when we have few cores, and less heating, the frequencies are increased to achieve more performance seeking this additional margin. As a result, the temperature increases.

With more cores, the margin for boost is smaller, generating less heating per core. The result is this greater stability in the heating line, with few cores giving a lot of boost and heating almost the same as many cores giving much less boost.

This also explains why in games like Counter Strike we have less, not more, performance with more cores. If the game does not use more than six cores efficiently, adding more cores only reduces boost margins.

Here we close our tests and conclusions with this experiment. Did you get any useful information? Is there another test you would like to take? Don’t forget to use the comment box for your feedback and suggestions!