Tensor G3 vs Google Tensor
We compared the 9-core Google Tensor G3 (Mali-G715 MP7) with the older 8-core Google Tensor (Mali-G78 MP20) SoC. Here you will find the pros and cons of each chip, technical specs, and comprehensive tests in benchmarks, like AnTuTu and Geekbench.
Review
CPU Performance
Single and multi-core processor tests
Gaming Performance
GPU performance in games and OpenCL/Vulkan
Battery life
Efficiency of battery consumption
NanoReview Score
Overall chip score
Key Differences
Pros of Google Tensor G3
- Performs 47% better in floating-point computations
- Shows better (up to 40%) AnTuTu 11 score β 1409K vs 1010K
- Has a smaller size transistor (4 versus 5 nm)
- Supports 33% higher memory bandwidth (68.2 against 51.2 GB/s)
- Has 1 more cores
- Better instruction set architecture
- Higher GPU frequency (~5%)
Benchmarks
Performance tests in popular benchmarksAnTuTu 11
| CPU | 482403 | 365546 |
| GPU | 384611 | 169803 |
| Memory | 218603 | 178763 |
| UX | 323399 | 295913 |
| Total score | 1409016 | 1010025 |
| Samples | [33 tests] | [51 tests] |
GeekBench 6
Single-Core Score
Multi-Core Score
| Asset compression | 197.6 MB/sec | 159.8 MB/sec |
| HTML 5 Browser | 134.7 pages/sec | 97.5 pages/sec |
| PDF Renderer | 148.4 Mpixels/sec | 138.9 Mpixels/sec |
| Image detection | 82.1 images/sec | 83.4 images/sec |
| HDR | 142.1 Mpixels/sec | 119.7 Mpixels/sec |
| Background blur | 13.2 images/sec | 14.1 images/sec |
| Photo processing | 48.4 images/sec | 30.2 images/sec |
| Ray tracing | 5.34 Mpixels/sec | 3.44 Mpixels/sec |
| Samples | 11 tests | 26 tests |
Compute Score (GPU)
| Background blur | 10.6 images/sec | 13.2 images/sec |
| Face detection | 6.85 images/sec | 7.31 images/sec |
| Horizon detection | 0.16 Gpixels/sec | 0.34 Gpixels/sec |
| Edge detection | 0.35 Gpixels/sec | 0.5 Gpixels/sec |
| Gaussian blur | 0.32 Gpixels/sec | 0.55 Gpixels/sec |
| Feature matching | 0.08 Gpixels/sec | 0.09 Gpixels/sec |
| Stereo matching | 11.9 Gpixels/sec | 39.4 Gpixels/sec |
| Particle physics | 1088 FPS | 1368.9 FPS |
| API | Vulkan | Vulkan |
| Samples | 9 tests | 11 tests |
3DMark
3DMark Steel Nomad Light
| Score | 982 | 704 |
| Stability | 74% | 63% |
| Graphics test | 7 FPS | 5 FPS |
3DMark Wild Life Extreme
| Score | - | 1899 |
| Stability | - | 59% |
| Graphics test | - | 11 FPS |
PCMark 3.0
PCMark 3.0
| Web score | - | 7105 |
| Video editing score | - | 5884 |
| Data manipulation score | - | 10310 |
| Photo editing score | - | 19072 |
| Writing score | - | 15550 |
| Phones tested | - | 3 models |
Recent User Tests
Google Tensor G3
| Date | Benchmark | Result |
|---|---|---|
| π 2026-08-28 (asfiii-malik) | AnTuTu 11 | 1515235 |
| π 2026-08-28 (asfiii-baloch) | AnTuTu 11 | 1515235 |
| π 2026-05-05 (s1mpick) | AnTuTu 11 | 1338293 |
| π 2026-05-04 (reazz) | AnTuTu 11 | 1356451 |
| π 2026-05-04 (Bilal Ahmed) | AnTuTu 11 | 1367231 |
Google Tensor
| Date | Benchmark | Result |
|---|---|---|
| π 2026-09-13 (VladiβοΈSlaveβοΈ) | AnTuTu 12 | 1281085 |
| π 2026-08-19 (duwi) | AnTuTu 11 | 774697 |
| π 2026-08-17 -> Paranoid _Antost | AnTuTu 11 | 930801 |
| π 2026-08-04 (nguyen van nguyen) | AnTuTu 11 | 990021 |
| π 2026-07-29 (Mimo) | AnTuTu 11 | 1117985 |
Specifications
Full list of technical specifications of Tensor G3 and Google TensorCPU
| Architecture | 1x 2.91 GHz β Cortex-X3 4x 2.37 GHz β Cortex-A715 4x 1.7 GHz β Cortex-A510 |
2x 2.8 GHz β Cortex-X1 2x 2.25 GHz β Cortex A76 4x 1.8 GHz β Cortex A55 |
| Cores | 9 | 8 |
| Frequency | 2910 MHz | 2800 MHz |
| Instruction set | ARMv9-A | ARMv8.2-A |
| L2 cache | - | 3 MB |
| L3 cache | - | 4 MB |
| Process | 4 nanometers | 5 nanometers |
| TDP (Sustained Power Limit) | 6 W | 5.6 W |
| Manufacturing | Samsung | Samsung |
Graphics
| GPU name | Mali-G715 MP7 | Mali-G78 MP20 |
| Architecture | Valhall 4th gen | Valhall 2nd gen |
| GPU frequency | 890 MHz | 850 MHz |
| Pipelines | 7 | 20 |
| Shading units | 128 | 32 |
| Total shaders | 896 | 640 |
| FLOPS | 1594.8 Gigaflops | 1085.4 Gigaflops |
| Vulkan version | 1.3 | 1.3 |
| OpenCL version | 2.0 | 2.0 |
AI Accelerator
| Neural processor (NPU) | Yes | Tensor Processing Unit |
Memory
| Memory type | LPDDR5X | LPDDR5 |
| Memory frequency | 4266 MHz | 3200 MHz |
| Bus | 4x 16 Bit | 4x 16 Bit |
| Max bandwidth | 68.2 Gb/s | 51.2 Gb/s |
| Max size | 16 GB | 12 GB |
Multimedia (ISP)
| Storage type | UFS 3.1, UFS 4.0 | UFS 3.1 |
| Max display resolution | 3840 x 2160 | 3840 x 2160 |
| Max camera resolution | 1x 200MP, 2x 32MP | 1x 200MP, 2x 32MP |
| Video capture | 8K at 30FPS, 4K at 120FPS | 4K at 60FPS |
| Video playback | 8K at 30FPS, 4K at 120FPS | 4K at 60FPS |
| Video codecs | - H.264 - H.265 - AV1 - VP9 |
- H.264 - H.265 - AV1 - VP9 |
| Audio codecs | - AAC - AIFF - CAF - MP3 - MP4 - WAV |
- AAC - AIFF - CAF - MP3 - MP4 - WAV |
Connectivity
| Modem | Exynos 5300i | Exynos 5300b |
| 4G support | LTE Cat. 24 | LTE Cat. 24 |
| 5G support | Yes | Yes |
| Download speed | Up to 7300 Mbps | Up to 7350 Mbps |
| Upload speed | Up to 3700 Mbps | Up to 3670 Mbps |
| Wi-Fi | 7 | 6 |
| Bluetooth | 5.3 | 5.2 |
| Navigation | GPS, GLONASS, Beidou, Galileo, QZSS | GPS, GLONASS, Beidou, Galileo, QZSS |
Info
| Announced | October 2023 | October 2021 |
| Class | Flagship | Flagship |
| Model number | GS301 (S5P9865) | GS101 (S5P9845) |
| Official page | Google Tensor G3 official site | Google Tensor official site |
| Successor | - Google Tensor G5 | - Google Tensor G5 |
Cast your vote
Total votes: 249