Google Pixel 8’s Tensor G3: Weak Performance, Yet Efficient

Google has once more joined forces with Samsung to create the Tensor G3, the chipset powering the new Pixel 8 and Pixel 8 Pro. This chipset is produced in Samsung’s 4nm foundries and boasts a 9-core CPU configuration (consisting of 1x Cortex-X3 + 4x A715 + 4x A510) along with an ARM Mali-G715 GPU.

All About the Google Pixel 8’s Tensor G3

While Google has remained tight-lipped about the specifics of this GPU, it appears to comprise 7 compute units, mirroring last year’s Tensor G2 (which featured a Mali-G710). To assess the efficiency of this GPU, Golden Reviewer conducted a series of tests.

Starting with 3DMark Wildlife Extreme, the chipsets were evaluated based on their performance-to-power consumption ratio, with higher scores indicating better performance. Unfortunately, the Tensor G3 didn’t fare well in this test, performing at a similar level to an Apple A14 but falling short of the Snapdragon 8 Gen 1.

Interestingly, the chipset worth noting in these tests is MediaTek’s Dimensity 9200. Despite also using the Mali-G715 (with 11 compute units), it outperformed the Tensor by a significant margin while displaying nearly identical efficiency characteristics.

Moving on to GFXBench Aztec (Vulkan/Metal back end, 1440p offscreen), the Dimensity once again outperformed the Tensor, offering nearly double the frame rate but drawing nearly double the power, resulting in similar efficiency.

In another GFXBench test (v3.1) at 1080p, the gap between the Google chip and the MediaTek chip was smaller, but when accounting for power draw, both chipsets exhibited similar overall efficiency.

According to Golden Reviewer, Samsung’s 4nm LPP node lags behind TSMC’s 4nm node by 1.5-2 generations. In more demanding tests, the Tensor G3 proved to be non-competitive with current flagship chips, let alone those on the horizon, such as the Snapdragon 8 Gen 3 and Dimensity 9300, expected to be unveiled later this year.

The disadvantage stemming from the node technology doesn’t bode well for Samsung’s own Exynos 2400. However, it’s worth noting that the Exynos 2400 features a distinct GPU architecture based on RDNA 3 technology from AMD, which could offer an architectural advantage.

Regarding Google’s future plans, the Tensor G4 is likely to continue being produced at Samsung’s foundry, given Google’s reliance on the expertise of the Exynos team. Nevertheless, there are rumors that Google is working on an in-house design that won’t be dependent on Samsung, with an expected release in 2025.