- The Google Tensor G4 on the Pixel 9 phones performs poorly on the Geekbench CPU test, despite packing the latest ARM cores.
- The older ARM Mali-G715 GPU on the Tensor G4 is also pretty weak, demonstrating a stability of just 59% in the 3DMark test.
- The chipset is severely restrained from amping up the clock speed to preserve battery life and keep the device cool. That results in a dismal performance.
Google has launched its latest flagship Pixel 9 series, featuring the Tensor G4 chipset. The Google Tensor G4 is said to be a small upgrade over last year’s Tensor G3. So to evaluate its performance gains, we have benchmarked the Tensor G4 chipset on Geekbench, 3DMark, AnTuTu, and many other tests in this guide.
Google Tensor G4 Specifications
Before we look at the numbers, take a quick look at the Tensor G4 specifications below:
Specs Tensor G4 CPU Eight-core CPU
(1+3+4)CPU Cores 1x 3.10GHz (Cortex-X4)
3x 2.60GHz (Cortex-A720)
4x 1.95GHz (Cortex-A520)Process Technology Samsung’s 4nm GPU Mali-G715 MC7 GPU
Up to 940MHzMachine Learning and AI Google Custom TPU (Codename Rio) ISP Google Custom DSP (Codename Callisto) Connectivity Wi-Fi 7
Bluetooth 5.3Modem Samsung Exynos 5400 5G modem
Up to 14.79Gbps Peak Download
Satellite SOSOthers AV1 Encoder and Decoder
Titan M2 security chip
Ultra-wideband chip
NavIC
Tensor G4: Geekbench 6 CPU
In the Geekbench 6 CPU test, the Google Tensor G4 scored 1,897 in single-core tasks and 3,721 in multi-core tasks. The CPU score is abysmally low, even lower than the Snapdragon 7+ Gen 3, a mid-range Qualcomm processor.
We ran the Geekbench benchmark multiple times to ensure there were no lacunae in our findings. However, each time the score came even lower than what is displayed below.
Despite packing the newest ARM cores, the multi-core CPU score couldn’t even cross the 4,000 mark. And mind you, this is on the larger Pixel 9 Pro XL that features a Vapor Cooling chamber for proper cooling. The numbers on the standard Pixel 9 will be even lower. In all likelihood, the culprit appears to be Samsung’s older 4nm process node.
It seems like Google didn’t go for Samsung’s improved 4LPP+ process node and FOWLP packaging for the Tensor G4. Also, Google is heavily underclocking the CPU to keep the thermals in check and maintain a healthy battery life. I say this because in no circumstance, the chipset could achieve such a low score despite clocking the prime core up to 3.10 GHz.
During the test, the device temperature remained under 37 degrees C, which means performance is being compromised for battery life. In conclusion, Tensor G4’s CPU can’t deliver peak performance or handle sustained workloads beyond the initial few minutes.
Geekbench 6 CPU Tensor G4 Single-core 1,897 Multi-core 3,721
![tensor g4 geekbench score](https://beebom.com/wp-content/uploads/2024/08/tensor-g4-geekbench-score.jpg?w=336&quality=75)
Tensor G4: AnTuTu Benchmark
Next, the Google Tensor G4 scored 1,253,407 in the AnTuTu benchmark test. Again, the score is unusually low for a processor that is supposed to power a flagship smartphone. In the AnTuTu CPU test, it scores just 337,131 points, and the 7-core Mali-G715 GPU manages to get 444,488 points.
What is more disappointing is that Google chose to include older-gen UFS 3.1 storage on the Pixel 9 series, which is much slower than UFS 4.0 storage. In the AnTuTu storage test, it only gets 82,120 points.
AnTuTu Benchmark Tensor G4 AnTuTu Score 1,253,407 CPU 337,131 GPU 444,488 Memory 243,643 UX 228,145 Storage 82,120
Tensor G4: CPU Throttling Test
In the CPU Throttling test, the numbers we see above become much more evident. We ran the test multiple times to ascertain the CPU behavior. In the first test, the CPU throttled to 71% (higher is better) of its maximum performance, and it only managed to get 42% in the second test. Both tests were done for 15 minutes.
The Pixel 9 series starts to throttle the Google Tensor G4 CPU from the very beginning, and only if the thermals are in check, it manages to perform at a slightly higher clock speed. Simply put, the Tensor G4 CPU is not designed for sustained peak performance.
Tensor G4: 3DMark Wild Life Extreme Stress Test
Coming to the GPU, we ran 3DMark’s Wild Life Extreme Stress test to evaluate the 7-core ARM Mali-G715 GPU on the Tensor G4. Google again chose to keep the same GPU from last year, although the GPU frequency has been elevated from 890MHz to 940MHz.
In our testing, the GPU got the best loop score of 2,590 and the lowest loop score of 1,525. The stability stood at a dismal 58.9%. And the temperature during the GPU stress test hovered around 42 degrees C. Hence, the GPU on the Tensor G4 is decidedly weak, and that’s on Google.
3DMark Wild Life Extreme Stress Test Tensor G4 Best loop score 2,590 Lowest loop score 1,525 Stability 58.9%
Tensor G4: Geekbench AI
To test the NPU, or what Google calls TPU, we ran the newly released Geekbench AI benchmark on the Tensor G4. By the way, the TPU on the Tensor G4 is also the same as last year’s Tensor G3. The TPU, running on the TensorFlow Lite framework, scored only 289 points in the Single Precision test.
Geekbench AI Tensor G4 CPU Tensor G4 GPU Tensor G4 TPU/NPU Single Precision Score 1,883 720 289 Half Precision Score 1,837 825 4,145 Quantized Score 2,899 772 6,103
To give you an idea of how bad the score is, Apple’s A17 Pro achieved 3,930 points in the same test using its Neural Engine. You can check out the GPU score below on AI workloads, which is again pitifully low. AI inference on the CPU, however, puts respectable numbers, but again, the competition is almost 2-3x faster.
Tensor G4: 5G Speed Test
Finally, in the 5G speed test, the new Samsung Exynos 5400 modem, managed to achieve 174.8 Mbps peak download speeds and 35.1 Mbps peak upload speeds. We ran multiple 5G speed tests, and the modem was very inconsistent. At times, the modem performed well, and often, the speeds would drop to half.
Tensor G4 Benchmarked: The Verdict
On paper, the latest Google Tensor G4 brings the latest ARM cores, and the CPU cores are clocked moderately high. However, Google’s orchestration between the OS and the processor is very (very very very) conservative. The phone severely underclocks the CPU to keep the Pixel 9 phones cool and preserve battery life, but that compromises performance to a large extent.
The GPU performance and stability are also poor. The inclusion of older UFS 3.1 storage is another disappointment. Not to mention, it packs the TPU from the last generation too. The only redeeming factor is the upgraded Exynos 5400 modem, which again delivers inconsistent speeds.
I understand that benchmark numbers don’t tell the complete story. However, when you are paying more than a thousand dollars ($1,490 in India, converted from INR to USD), you expect the phone to have at least a competitive processor.
In my opinion, the Google Tensor G4 is by no means a flagship processor. It seems we will have to wait another year for Google to fix its silicon (or maybe not). Are you already using the Pixel 9 or 9 Pro phone? How has your experience with the phone been so far? Tell us in the comments below.