Tensor G4 vs Tensor G3

The Tensor G4 was released with the Pixel 9 series in August. The tech community, including us at Gizmochina, was expecting some significant improvements over last year’s Tensor G3, but the benchmark results surprised us.

The Google Tensor G4 lags behind the Snapdragon 8 Gen 3 by a good margin on benchmarks (see the Tensor G4 vs. Snapdragon 8 Gen 3 comparison).

According to Google, the new Tensor chip offers 20% faster web browsing, 17% faster app launching, and up to 20% improved power efficiency. But how does it perform on benchmarks against the Tensor G3? And is it worth spending extra cash for Tensor G4? Let’s compare the two chips to find the answer.

Let’s begin our comparison with the spec sheet:

Tensor G4Tensor G3
Release DateAugust 2024October 2023
Process Node4nm4nm
ManufacturerSamsungSamsung
CPU Cores89
CPU Architecture1 x Cortex-X4 @ 3.10 GHz
3 x Cortex-A720 @ 2.60 GHz
4 x Cortex-A520 @ 1.92 GHz
1 x Cortex-X3 @ 2.91 GHz
4 x Cortex-A715 @ 2.37 GHz
4 x Cortex-A510 @ 1.70 GHz
GPUArm Mali-G715 MC7
(940 MHz)
Arm Mali-G715 MC7
(890 MHz)
Machine Learning3rd-gen Tensor Processing Unit3rd-gen Tensor Processing Unit
MemoryLPDDR5XLPDDR5X
StorageUFS 3.1, UFS 4.0UFS 3.1, UFS 4.0
Media DecodeH.264, H.265, VP9, AV1H.264, H.265, VP9, AV1
ModemExynos 5400 5G modem
Downlink: 14.79 Gbps
Exynos 5300i modem
Downlink: 10 Gbps
ConnectivityWi-Fi 7
Bluetooth 5.3
Satellite connectivity
Wi-Fi 7
Bluetooth 5.3

Tensor G4 vs Tensor G3: Benchmarks

[Note: For testing on different benchmarks, we picked the Pixel 9 Pro XL (powered by the Tensor G4) and the Pixel 8 Pro (powered by the Tensor G3).] 

[Geekbench 6]

It’s a CPU-centric test that runs a device through a series of computation workloads, including compression (text & images), encryption, physics simulations, rendering, ray tracing, and speech recognition.

The benchmarking platform outputs single-core and multi-core results. The single-core score shows the device/chip capability when a single core is used, and the multi-core score tells the CPU performance when multiple cores are at work.

Tensor G4 vs Tensor G3 - Geekbench 6

The latest Tensor G4 manages to pull off higher points in both single-core and multi-core tests than last year’s Tensor G3. However, the difference is not significant. Let’s look at some more benchmarks for better understanding. 

Tensor G4Tensor G3
Single Core1,9671,772
Multi Core4,6974,485

[AnTuTu 10]

AnTuTu is a popular benchmarking platform that measures the CPU, GPU, memory, and UX performance of devices.

Tensor G4 vs Tensor G3 - AnTuTu 10

It’s quite surprising to see the Tensor G3 leading the CPU chart on AuTuTu 10, that too, by a significant margin. While GPU, memory, and UX scores went in favor of Tensor G4.

Tensor G4Tensor G3
CPU1,86,6082,70,092
GPU4,48,6754,01,268
Memory1,69,1761,42,784
UX1,90,7871,84,212
Total Score9,95,2469,98,356

Before you start thinking a lot about these benchmarking results, let me quickly remind you that these scores do not reflect real-time performance, as it depends on many other factors, such as software optimizations, devices, and usage patterns.  

Tensor G4 vs Tensor G3: Key Differences

Samsung has manufactured the Tensor G4 and Tensor G3 using its 4 nm process node. However, the Tensor G4 has been manufactured using a newer technology, which translates to better efficiency.

The Tensor G3 has nine CPU cores, while the Tensor G4 has one less CPU core. It doesn’t mean lower performance or efficiency on the latest Tensor chip as it has high-performance and more efficient cores.

The latest Tensor chip has one Cortex-X4 prime core clocked at 3.1 GHz (vs. Cortex-X3 @ 2.91 GHz). The four Cortex-A720 performance cores also run at a higher clock speed than the Cortex-A715 in the Tensor G3. The remaining Cortex-A520 cores are more power-efficient than Cortex-A510 in Tensor G3.

Both chips have the same 7-core Mali-G715 GPU, although the newer chip runs at a higher clock frequency for better peak performance.

Tensor G4 doesn’t bring a big performance leap over last year’s Tensor G3, and Google is okay with it. Tensor chips have never been about blazing-fast performance; they’re more about unique features. “When we are designing the chip, we’re not designing it for speeds and feats. We’re not designing it to beat some specific benchmark that’s out there. We’re designing it to meet our use cases,” said Soniya Jobanputra, a key member of Google’s Pixel product management team.

The focus with Tensor G4 was not to beat benchmarks but to improve the speed and efficiency of opening apps. “We knew that we had a pain point with opening apps. And so as we built G4, we really focused on, okay, what do we need to do to make sure that experience is better for users,” Jobanputra added.

Besides CPU and GPU performance, some key changes have been made to improve connectivity. Tensor G4 has a new Exynos 5400 modem, which offers a 50% higher peak download speed. The new chip also supports satellite connectivity, which will be enabled when upgrading to Android 15, which is lacking on Tensor G3.

Tensor G4 vs Tensor G3: Verdict

Even though Tensor G4 has the latest ARM CPU cores, performance has not gone up drastically compared to last year’s Tensor G3. The GPU is unchanged, but it now runs at a higher clock frequency that might offer better peak performance, but again, it’s not a big deal.

On benchmarks, the Tensor G4 can’t even match the Snapdragon 8 Gen 2, released in 2022. So, Google seems a generation behind the latest chips from Qualcomm and Apple as far as performance is concerned. However, Tensor chips enable AI features you wouldn’t find on non-Pixel devices.

The bottom line is Tensor G4 is not a major upgrade over Tensor G3. However, we see big improvements coming next year, as Google will reportedly switch to TSMC for the manufacturing of Tensor G5 using its cutting-edge technology.

Popular Chip Comparisons:
1. Tensor G4 vs Snapdragon 8 Gen 3
2. Snapdragon 8 Gen 3 vs 8 Gen 2
3. Snapdragon 8 Gen 3 vs Dimensity 9300

Comments