Hey. I don’t usually post in the Android or Pixel Boards, but given the popularity/hype around the new Pixel 7s/Tensor G2s, I decided to give a barebones benchmark, similar to PC Hardware.
TLDR: Performance is great. Throttling is 82% of max_p. Battery life is 2.5 hours gaming.
My prior phone was an ASUS ROG2 (a high performance “gaming phone”), so my primary concern was performance, speed, heat, and frame rates.
For the benchmarks/scores, I tested both phones on a gigantic heatsink, so the silicon beneath could run at full tilt without any thermal throttling.
The CPU BURN tests were done specifically the opposite, in the original configuration, upright/freestanding, with no heatsinks of any kind (your hands count as a heatsink). Both are using the same type of case and material – clear plastic. Pixel has a Google approved/partnered case, to showcase “real world” conditions.
The first test was to establish the maximum TDP/power draw of the processor and GPU package together, **to determine real-world gaming battery life.** This was done by checking 3Dmark real-time power consumption via AccuBattery during benchmarking.
**[Pixel 7 Pro TDP]**
Battery Voltage: 4.0V
Full Load (mA): 1500 1777 1971 1491 1660 1465 1398 1452 1596 1906 1592 1625 1750 1507 1437
Average Draw: 1608 mA / 6.432 W
Maximum Draw: **7.884 W**
Runtime at Max Load: **2.44 hours** (2 hours, 26 minutes)
**[ASUS ROG2 TDP]**
Battery Voltage: 3.9V
Full Load (mA): 1379 1434 1523 1500 1457 1441 1418 1360 1333 1331 1374 1363 1553 1474 1699 1729
Average Draw: 1460 mA / 5.694 W
Maximum Draw: **6.743 W**
Runtime at Max Load: **3.425 hours** (3 hours, 26 minutes).
The second test was to push the CPU/Package as hard as possible by doing number crunching. This is to test the cooling system, and to see if the package throttles under intense load.
**[Thermal Throttling / CPU Burn Tests]**
[Pixel 7 Pro CPU BURN Test Result](https://imgpile.com/images/T3SqQE.png)
Test Length: 5 minutes
Burn Method: Linpack
Average Performance: 251,588 GIPS
Average versus Maximum: 82%
CPU Throttling: YES
CPU Throttling Threshold: 2m 30s
CPU Throttling Level: Minor (-18%)
[ASUS ROG2 CPU BURN Test Result](https://imgpile.com/images/T3SANr.jpg)
Test Length: 5 minutes
Burn Method: Linpack
Average Performance: 199,083 GIPS
Average Performance: 95%
CPU Throttling: NO
CPU Throttling Threshold: N/A
CPU Throttling Level: Insignificant (-5%)
Finally, we look under the hood, compare some specs, run GeekBench5, 3DMark without heat being an issue, and see if it’s a good price/performance ratio:
***Pixel 7 Pro***
Release Date: October 7, 2022
MSRP: $900
SoC/Chipset: Tensor G2
Processor: ARM Cortex Octo-Core, Revision R1P1
Subset: ARM Cortex A-78 (x2) @ 2.35 GHz
A-55 (x4) @ 1.80 GHz
X-1 (x1) @ 2.85 GHz
GPU: Mali-G710
RAM: 12 GB, 11422 MB usable
Flash/Storage: 128 GB, 109.61 GB usable
Battery: 5000 mah / 19.25 Wh [*still undergoing durability testing*]
Display Resolution: 3120 x 1440
Display Refresh: 120 Hz
[GeekBench5](https://imgpile.com/images/T3S97o.png)
[3DMark](https://imgpile.com/images/T3SCPC.png)
[CPU-Z](https://imgpile.com/images/T3SDLS.png)
Single Core — **1060**
Multi Core — **3234**
3DMark WildLife Score: **6440**
Average FPS: 38.50
Average Temperature: 25C
***ASUS ROG 2***
Release Date: July 22, 2019
MSRP: $800
SoC/Chipset: Snapdragon 855+
Processor: Qualcomm Octo-Core, Revision R13P14
Subset: Kryo 485 @ 2.96 GHz
GPU: Adreno 640
RAM: 12 GB, 11476 MB usable
Flash/Storage: 512 GB, 461.71 GB usable
Battery: 6000 mah / 23.1 Wh
Display Resolution: 2340 x 1080
Display Refresh: 120 Hz
[GeekBench5](https://imgpile.com/images/T3SlOM.jpg)
[3DMark](https://imgpile.com/images/T3Sr44.jpg)
[CPU-Z](https://imgpile.com/images/T3SzG2.jpg)
Single Core — **744**
Multi Core — **2630**
3DMark WildLife Score: **3464**
Average FPS: 20.70
Average Temperature: 25C
In Conclusion:
Why compare a smartphone/SoC from 2019 against one made in 2022?
A. The ROG2 was sold as “a flagship”, with performance being priority/the prime selling point.
B. $800 in 2019 is $928 in today’s 2022 dollars. Current “flagships” / 8+ Gen-1 SoCs are $1200-$1400 in 2022 dollars.
C. To evaluate if there were any performance and energy efficiency improvements in SoC tech across the years (if Moore’s Law is still working), identify heat issues, and compare non-dedicated performance hardware against mass market (“gaming vs. non-gaming”).
E. I own both, and wanna see if it’ll burn my hands lol
How does the Pixel 7 / Pixel 7 Pro stack up with the rest in terms of performance:
A. The $600 Pixel-7 is a far more compelling deal if it has the same processor/package as the $900 Pixel 7 Pro.
B. The cooling system with a single graphene layer seems pretty good, but the processor definitely can’t handle a sustained full load for more than about 2-3 minutes without minor thermal throttling (-18%). This is a type of phone you probably want to run caseless, or in an aluminum shell of some sort for maximum performance.
C. For the normal workloads of these phones, they’re unlikely to throttle, at least within the first year. Year 3-5, it’s unknown how the battery, and CPU will respond/degrade, and what level of hardware demand will be asked of by the software.
D. The software is everything with these phones (Pixel). CPU-Z shows that the Pixel7Pro is using a custom CPU governor (sched_pixel), likely based on (sched_util). Google’s various AI-based quirks/software optimizations and upgrades to Android, may just take it a step beyond in the future.
@EcureuilHargneux asked about the Pixel 6A, I happened to be getting one in the mail, so I benchmarked it also for comparison:
***Pixel 6A***
Release Date: July 28th, 2022
MSRP: $450
SoC/Chipset: Tensor G1
Processor: ARM Cortex Octo-Core, Revision Unknown
Subset:
2x Cortex-X1 @ 2.8 GHz
4x Cortex-A76 @ 2.25 GHz
2x Cortex-A55 @ 1.8 GHz
RAM: 6 GB, 5580 MB usable
Flash/Storage: 128 GB, 110.27 GB usable
Battery: 4410 mah / 16.98 Wh
Display Resolution: 2400 x 1080
Display Refresh: 60 Hz
Governor: sched_pixel
[GeekBench5](https://imgpile.com/images/TpdmBi.png)
[3DMark](https://imgpile.com/images/Tpdzuu.png)
[CPU-Z](https://imgpile.com/images/TpdlcF.png)
Single Core — **971**
Multi Core — **2501**
3DMark WildLife Score: **6440**
Average FPS: 38.60
Average Temperature: 25C
TDP
Battery Voltage: 3.80 V
Measurements (mA): 2886 2563 2634 2311 2801 2119 2677 2838 3058 2944 2067 1988
Average Draw: 2573 mA / 9.77 W
Maximum Draw: 11.62 W
Runtime at Max Load: *1.461* hours** (1 hours, 27 minutes)
I was discouraged by the high power draw of the Tensor G1 chip during the 3Dmark Gaming Testing, so I decided to run the BURN TEST without a case to give it the best possible shot in terms of thermals. That however is unscientific, so I also ran one with a Google approved/Pixel standard 6A case for comparison reasons (same, clear plastic).
[PIXEL 6A BURN TEST CASELESS](https://imgpile.com/images/TpdFAg.png)
Test Length: 5 minutes
Burn Method: Linpack
Average Performance: 219,814 GIPS
Average versus Maximum: 88%
CPU Throttling: YES
CPU Throttling Threshold: N/A
CPU Throttling Level: -12%
[PIXEL 6A BURN TEST CASED](https://imgpile.com/images/Tpd3VW.png)
Test Length: 5 minutes
Burn Method: Linpack
Average Performance: 221,007 GIPS
Average versus Maximum: 90%
CPU Throttling: YES
CPU Throttling Threshold: N/A
CPU Throttling Level: -10%
Paradoxically, running the Pixel6A with a case was better, than without, by about 2%.
The max TDP of the Tensor G1 was 11.62 W, the max TDP of the Tensor G2 was 7.88W.
**BONUS ROUND — MACHINE LEARNING BENCHMARKS**
A lot of the posters in the thread have mentioned that they’re not interested in the raw compute scores or gaming performance, so I decided to go ahead and run TensorFlow Lite using GPU+NNAPI on all three devices – you know, if you wanted to do machine learning on a smartphone, for some reason.
**ASUS ROG2 (Snapdragon 855+)**
CPU Score: 352
[GPU Score: **1213**](https://imgpile.com/images/TpdyLC.jpg)
NNAPI: 623
**Pixel 7 Pro (Tensor G2)**
CPU Score: 319
GPU Score: 1016
[NNAPI: **2197**](https://imgpile.com/images/TphTPu.png)
**Pixel 6A (Tensor G1)**
CPU Score: 306
[GPU Score: **1362**](https://imgpile.com/images/TpdZFx.png)
[NNAPI: 1326](https://imgpile.com/images/TpdSfL.png)
A. Why GPU rather than NNAPI ONLY?
A lot of smartphones (or computers even) don’t have dedicated neural processing chips, and have to use things like CUDA or OpenCL to run Machine Learning APIs/programs on graphics processing hardware. In the case of the ROG2, the score running on the GPU was roughly double than running in NNAPI mode.
B. Why did the Pixel6A score higher on GPU versus NNAPI score?
I have no idea, but this underscores the reasoning that raw hardware beneath is super important – if the API doesn’t exist or optimizations/implementations don’t exist, your barebones determines raw speed.