Google's Tensor G4 Chip Configuration, Benchmark, and More
TapTechNews August 1st news, the tech media AndroidAuthority released a blog post yesterday (July 31st), sharing relevant information about the Tensor G4 chip搭载 on the Pixel 9 series phones, indicating that the performance upgrade of this chip is not significant.
Tensor G4 Chip Configuration
The Google Tensor G3 chip has a total of 9 CPU cores:
One Cortex-X3 large core
Four Cortex-A715 medium cores
Four low-power Cortex-A510 small cores
The Google Tensor G4 chip adopts a more traditional 1+3+4 cluster configuration, upgrades to use the latest ARMv9.2 core of Arm, and slightly increases the core clock frequency:
1 Cortex-X4 large core
3 Cortex-A720 cores
4 Cortex-A520 cores
Tensor G4 also uses the same Mali-G715 GPU as Tensor G3, but the main frequency is increased from 890 MHz to 940 MHz.
Tensor G3(zuma)Tensor G4(zumapro)CPU small core 4xCortex-A510@1.7GHz 4xCortex-A520@1.95GHzCPU medium core 4xCortex-A715@2.4GHz 3xCortex-A720@2.6GHzCPU large core 1xCortex-X3@2.9GHz 1xCortex-X4@3.1GHzGPU Mali-G715MC7@890MHz Mali-G715(unknown number of cores)@940MHzTensor G4 Chip Benchmark
Based on the GeekBench benchmark data, The single-core performance of Tensor G4 is 11% higher than that of Tensor G3, and the multi-core performance is increased by 3%.
Tensor G4 Chip Modem
The main power consumption source of the Google Tensor SoC has always been the modem, not the CPU.
TapTechNews cited sources to report that the Tensor G4 is paired with the brand-new Exynos Modem 5400, which is important to improve energy efficiency in addition to supporting satellite connections.
Source disclosed that compared to the Exynos Modem 5300搭载 on the Pixel 8 series, the power consumption is reduced by 50%, but this requires more detailed testing later.
Google is currently still developing the Tensor G4+Exynos Modem 5300 combination, and it is estimated that it will be equipped on the Pixel 9a phone in the future.
Tensor G4 Chip Packaging
Google can continue to build more advanced Pixel features, and the most important thing is in the fields of artificial intelligence, camera, and security. In terms of Tensor G3, Google has quite a few custom IP modules:
Edge TPU (ML accelerator)
GXP (digital signal processor, mainly used to accelerate camera tasks)
BigWave (AV1 encoder/decoder)
Titan M2 security chip
There is no change in Tensor G4. The Edge TPU is still the same model, code-named rio, and runs at the same clock speed; the GXP is still callisto and also runs at the same frequency; BigWave has not changed at all.
Previously, there was news that it would adopt Samsung's fan-out wafer-level packaging (FOWLP), so as to perform better in terms of heat dissipation, etc. However, from the shipping list, the packaging of Google Tensor G4 chip is still fan-out panel-level packaging (FOPLP).
