How efficient is Google’s top chipset?

The maximum peak performance of a CPU is a key metric to consider when comparing different CPU designs. However, that’s not the only number that matters. Other important metrics include sustained performance and performance per watt. In this article, I take a deep dive into the performance per watt of the Google Tensor G3 system-on-chip (SoC) currently used in Pixel 8 phones, and potentially in future Google phones as well.

Google Tensor G3 is an SoC designed for mobile devices, especially Pixel devices. It’s made by Samsung, likely in partnership with Google, and has many similarities to other Samsung chips. The CPU consists of 9 Arm cores: 1 Cortex-X3 with a maximum frequency of 2.91GHz, 4 Cortex-A715 with a maximum frequency of 2.37GHz, and 4 Cortex-A510 with a maximum frequency of 1.7GHz. The SoC also includes Arm Mali-G78 GPU and is built on Samsung’s 4nm process.

Google Tensor G3: Performance results per watt

To measure performance per watt, I used a rooted Pixel 8 device. This allows me to control which CPU core my benchmarks run on. I can control the maximum CPU frequency so I can see the performance at each frequency step. I was also allowed to set up governors. In this case, I set it to Performance. I then measured the smartphone’s energy consumption to calculate performance per watt.

Watch the video above for a full description of my findings. Below is a brief summary of my Tensor G3 performance per watt results.

The first graph plots performance versus frequency. As clock frequency increases, performance increases, but the gradient of the wires is different for each CPU core. The A715 and X3 are very close, but the A510 is obviously different, showing a large performance gap. However, the A510 consumes less power, so it is used as an energy-efficiency core rather than a high-performance core.

The graph below visually displays the relationship between frequency and power consumption and demonstrates the A510’s efficiency at low frequencies. It uses significantly fewer cores than the two faster ones, which is clearly reflected at the 1.0GHz mark. Still, this test shows that increasing clock speeds will result in more power consumption, but at a certain point there are diminishing returns.

The last chart compares power versus performance and shows where each core is most efficient. Up to a point, the A510 uses very little power to perform well. Thereafter, the A715 delivers higher performance while running at lower frequencies at the same power. The A715 starts to plateau at higher frequencies, and this is where the X3 proves its inclusion. It can extract the same performance while running at lower frequencies than the A715, providing better efficiency in higher performance bands.

Overall, the Google Tensor G3 system-on-chip delivers better power performance than previous generations. I expect this positive trend to continue as we make progress across different microarchitecture and process nodes.