I recently received the Google Pixel 8 Pro to review, which will likely be completed by the end of next week.

In the meantime, I have done my usual benchmarks to give some insight into the performance of the new Google Tensor G3.

Blocked Benchmark Apps

Shortly after the Pixel 8 series was launched, it was reported that Google had blocked many benchmarking apps, including Geekbench and 3D Mark, from running. There was no official statement on why this was, but there were a few theories.

  • It is easy to jump to the conclusion that Google did this deliberately to hide the fact that the Tensor G3 is not as good as other flagship chips, namely the Snapdragon 8 Gen 2.
  • A similar but alternative view was that Google has put a lot of emphasis on the AI features of the chipset, and these benchmarks are not representative of that.
  • The last explanation, and the most likely, was that those apps just weren’t compatible with the Pixel 8 series at launch.

Since the block was reported, all the usual benchmarking apps can be installed.

Edit – It has now been reported that Google did block the benchmarking apps during the review embargo period.

Benchmarking Caveats

Considering the uproar about the blocked benchmark apps, I feel I should highlight that the following benchmarks are deeply flawed and don’t necessarily represent real-world performance.

Different phones using the same chipset can often score wildly different results. This will likely be due to the way manufacturers will vary performance based on specific tasks. In the past, some brands have cheated benchmarking results by making the phone run at a higher-than-normal performance to achieve better results. A less sinister situation is that some brands focus more on performance than battery. You can see this with RedMagic phones, which often outperform other phones but run much hotter and use much more battery during a benchmark.

I still do benchmarks because they are easy to do and work reasonably well as a guide. Sadly, I just don’t have the time or readership to spend hours doing a deep dive analysis or real-world performance.

Google Tensor G3 vs G2 vs Tensor vs Snapdragon 8 Gen 2 Specification

Google Tensor G3Google Tensor G2Google TensorQualcomm Snapdragon 8 Gen 2
ProcessSamsung 4nm Samsung 5nmSamsung 5nmTSMC 4nm (N4)
CPU9-Core
1x Arm Cortex-X3 (2.91GHz)
4x Arm Cortex-A715 (2.37GHz)
4x Arm Cortex-A510 (1.70GHz)
8-Core
2x Arm Cortex-X1 (2.85GHz)
2x Arm Cortex-A78 (2.35GHz)
4x Arm Cortex-A55 (1.80GHz)
8-Core
2x Arm Cortex-X1 (2.80GHz)
2x Arm Cortex-A76 (2.25GHz)
4x Arm Cortex-A55 (1.80GHz)
8-Core
1× Cortex-X3 (3.2 GHz)
4× Cortex-A715 (2.8 GHz)
3× Cortex-A510 (2.0 GHz)
GPUArm Immortalis-G715Arm Mali-G710 MP7Arm Mali-G78 MP20Adreno 740
CachesUnknown4MB CPU L3
8MB system level
4MB CPU L3
8MB system level
8 MB level 3 cache
RAMLPDDR5 @ 5500MHzLPDDR5 @ 4200MHzLPDDR5X @ 4200MHz
Machine LearningThird-gen Tensor Processing UnitNext-gen Tensor Processing UnitTensor Processing UnitQualcomm Hexagon
Media DecodeH.264, H.265, VP9, AV1H.264, H.265, VP9, AV1H.264, H.265, VP9, AV1H.264, H.265, VP9, AV1
Modem4G LTE
5G sub-6Ghz and mmWave
4G LTE
5G sub-6Ghz and mmWave
4G LTE
5G sub-6Ghz and mmWave
4G LTE
5G sub-6Ghz and mmWave
WiFiWiFi 7WiFi 6EWiFi 6WiFi 7
Bluetooth5.35.25.25.3

The Google Tensor G3 has a significant upgrade compared to its predecessors. The G2 was barely an upgrade from the original Tensor, so the Pixel 8 series feels like a proper upgrade from the Pixel 6 I was previously using.

The 9-core design is particularly interesting, and I would assume it has been implemented this way for efficiency.

As someone who writes about networking and WiFi a lot, I am glad to see that WiFi 7 is included. The Qualcomm Snapdragon 8 Gen 2 may support it, but it is not guaranteed, my Honor Magic5 Pro doesn’t even have WiFi 6E.

Benchmarks

Antutu

AntutuTotalCPUGPUMEMUX
Snapdragon 8 Gen 3
iQOO 12
2084308446149906269400131331759
Snapdragon 8 Gen 3
Samsung Galaxy S24 Ultra
1945797445818817997373210308772
Snapdragon 8 Gen 2
Honor Magic5 Pro
V10
1367374298680583069259642225983
Google Tensor G3
Pixel 8 Pro
V10
1144447355374381544177818229711
Dimensity 8200-Ultra
Xiaomi 13T
V10
896827260471230211192345213800
MediaTek Dimensity 8100
Realme GT Neo 3
V10
871235260194215150180938214956
Google Tensor G2
Pixel 7 Pro
V10
848876231239289398141067187172
Google Tensor
Pixel 7 6
V10
845450249800236824157931200895
MediaTek Dimensity 8050
Infinix Note 30 VIP
V10
732420205125184374158126184795

In Antutu, the Google Tensor G3 on the Pixel 8 Pro, scores a respectable score of 1,144,447, but this falls short of 1,367,374 Qualcomm by 16%.

The score breakdown reveals some interesting differences: the CPU score on the Tensor G3 is 19% higher, but the GPU score is 35% lower.

There is a significant improvement from the Tensor and Tensor G2 with a score that’s 35% higher than the Pixel 7 Pro.

Geekbench 6

ChipsetSingle CoreMulti Core
Snapdragon 8 Gen 3
iQOO 12
22506909
Snapdragon 8 Gen 3
Samsung Galaxy S24 Ultra
21036640
Snapdragon 8 Gen 219495235
Tensor G317634409
Tensor G214103410
Tensor12643283
Dimensity 8200-Ultra11193733

For Geekbench 6, I initially thought the Google Tensor G3 may perform better than it did due to the 9-core design. I at least thought it would score a higher multi-core score. The Qualcomm chipset may have fewer cores, but all the cores are clocked quite a bit higher than the Tensor G3.

The end result is that the Qualcomm chipset scores 11% and 19% higher than the Tensor G3 for single and multi-core, respectively.

You have a substantial improvement of 25% and 29% going from the G2 to G3.

3DMark Wildlife Stress Testing / Thermal Throttling

3DMarkChipsetWild Life High Wild Life LowStability TemperatureBattery
iQOO 12 RetestSnapdragon 8 Gen 3183461214466.2%26° to 42°87% to 73%
iQOO 12 First TestSnapdragon 8 Gen 318087754941.7%25° to 39°96% to 82%
Samsung Galaxy S24 UltraSnapdragon 8 Gen 317580982855.9%24° to 42°79% to 66%
Honor Magic5 Pro Snapdragon 8 Gen 212726856767.3%19° to 38°100% to 92%
Honor Magic4 Pro Snapdragon 8 Gen 1 10189644563.3%24° to 44°86% to 72%
OPPO Find X5 ProSnapdragon 8 Gen 1 9537606063.5%21° to 44°33% to 23%
Pixel 8 ProGoogle Tensor G38434559966.4%27° to 44°83% to 72%
Pixel 7 ProGoogle Tensor G26527476673%
Pixel 6Google Tensor6908348750.5%
Xiaomi 13T Dimensity 8200-Ultra 6485587990.728° to 42°83% to 74%
OnePlus 9 Pro Snapdragon 8885775348960.3%
OPPO Find X3 ProSnapdragon 8885765380466%
Samsung Galaxy S21 UltraExynos 21005466640462.3%
Xiaomi Mi 11Snapdragon 8885550504590.9%
Realme GTSnapdragon 8885850323455.3%
Oppo Find X3 NeoSnapdragon 8655038494298.1%
Realme GT Neo 3Dimensity 8100 5446539999.1%32° to 38°100% to 94%
Infinix Note 30 VIPDimensity 80504565450698.731° to 46°100% to 91%
OnePlus Nord 2Dimensity 12004161293270.5%

For the 3DMark Wildlife Stress Testing, I have reused an old table with quite a lot of phone benchmarks on it.

The Qualcomm chipsets are clearly better for gaming than the Google Tensor chipsets. The Snapdragon 8 Gen 2 scores 51% and 53% higher for the low and high scores with a similar stability level.

PC Mark Performance and Work 3.0 Battery

Performance ScoreWork 3.0 Battery
Honor Magic5 Pro 1431013h 38 mins
Honor Magic4 Pro 1467010h 32 mins
Samsung Galaxy S21 Ultra
Exynos 2100
1363812h 46min
Pixel 8 Pro
1209112 hours 56 mins
OPPO Find X3 Pro119329 hours
Pixel 7 Pro
Tensor G2
1155011h 39min
Pixel 6
Tensor
114439h 45min

I always find that the PC Mark results are all over the place as they are highly dependent on how a manufacturer handles different workloads. Some brands score suspiciously high scores.  

I think the important thing here is that the Pixel series has improved the battery times with each generation, with a significant improvement since the Pixel 6. However, the performance score has only improved negligibly, with a 5% improvement going from the G2 to G3.

AI Benchmarks

For the AI benchmarks, I only had access to the Pixel 8 Pro, Pixel 6 and Honor Magic 5 Pro.

With the AI benchmarks, things can get a little confusing. The Google Tensor G3, uses the Android Neural Networks API (NNAPI) that is built into Android.

Qualcomm chipsets have their own neural network library, and the problem here is that a developer needs to apply to Qualcomm to be allowed to use it. This ends up meaning the Snapdragon 8 Gen 2 can perform significantly better than the Tensor G3 if the application has access to those libraries.

Geekbench ML

  • Pixel 8 Pro: 2122
  • Pixel 6: 1989
  • Honor Magic 5 Pro: 562

In the case of Geekbench ML, it appears the Snapdragon 8 Gen 2 on the Honor falls back to CPU rather than using NNAPI, which results in a significantly worse performance than either of the Pixel phones.

As for the difference between the Pixel 6 and 8 Pro, the 6.7% uplift in performance is quite poor, considering the chipsets are two generations apart.  

AI Benchmark

  • Pixel 8 Pro: 718
  • Pixel 6: 337
  • Honor Magic 5 Pro using Qualcomm Libraries: 1870
  • Honor Magic 5 Pro using NNAPI: 135

AI Benchmark has access to the correct neural network library for Qualcomm, which allows it to achieve a substantially higher score than the Google Tensor chipsets.

The Honor Magic 5 Pro score is over 2.5 times higher than the Pixel 8 Pro.

Toggling off the Qualcomm libraries and attempting to use the same NNAPI as the Pixel phones, the score plummets 93% down to 135, which is also 81% lower than the Pixel 8 Pro. Switching to GPU acceleration improved the score to 359.

However, at least the Pixel 8 Pro scores a significantly larger score than the Pixel 6 with over double the score.

Overall

Based on benchmarks alone, the Google Tensor G3 is certainly inferior to its Qualcomm counterpart.

There is more to it than that, though. Google don’t seem to make quite a big deal about their chipset being the absolute best in raw performance like other chipset brands. If you look at any of the marketing pages from Google for the Pixel 8 Pro or the Google Tensor G3, there is no mention of things like gaming; they are not advertising things like Ray Tracing or anything like that.

The focus is predominantly on AI and photography. While many people think the focus of on-device AI is a bit of a scam, there is no denying the prowess of Pixel phones for photography.

From my personal perspective, I don’t game much, and my phone is mainly used for productivity and photography. Phones are so powerful nowadays I don’t really feel a great deal of benefit about new chipsets; until recently, I was quite content with my Pixel 6, though I was getting frustrated with its battery.

I have since moved to the Honor Magic5 Pro, which I love. Now, I am using the Pixel 7 Pro, which, according to the above benchmarks, has an inferior chipset than the Honor. However, I have noticed absolutely no difference in perceived performance (it is early days, though).

So far, I am more than happy with the Pixel 7 Pro, it is a huge improvement from the Pixel 6, mainly because I absolutely hated the fingerprint sensor on the Pixel 6, while the Pixel 7 Pro works perfectly and has the added benefit of facial recognition.

Similar Posts

Leave a Reply

Your email address will not be published. Required fields are marked *