After a slew of leaks and rumours, we saw Google finally unveil the Pixel 6 and the Pixel 6 Pro. As expected, the Pixel 6 Series comes with an in-house Google Tensor chipset, which is custom-designed by Google in collaboration with Samsung.
With Google’s first-ever custom silicon launch, the company has involuntarily become a part of the smartphone chipset wat against Apple and Qualcomm. So in this article, we compare the Google Tensor, Snapdragon 888, and A15 Bionic to see if Google can stand its ground in the raging SoC battle.
- 1 Google Tensor Vs. Snapdragon 888 Vs. A15 Bionic: A comprehensive comparison(2021)
- 2 Google Tensor Vs Snapdragon 888 Vs A15 Bionic: CPU
- 3 Google Tensor Vs Snapdragon 888 Vs A15 Bionic: GPU
- 4 Google Tensor Vs Snapdragon 888 Vs A15 Bionic: ISP
- 5 Google Tensor Vs Snapdragon 888 Vs A15 Bionic: AI and ML
- 6 Google Tensor Vs Snapdragon 888 Vs A15 Bionic: Connectivity
- 7 Conclusion: Who takes it all?
Google Tensor Vs. Snapdragon 888 Vs. A15 Bionic: A comprehensive comparison(2021)
In this comparison between Google Tensor, Snapdragon 888, and the A15 Bionic, we have discussed the CPU performance, GPU, ISP, AI, and ML capabilities, and more.
Google Tensor Vs. Snapdragon 888 Vs. A15 Bionic: Specifications:
|Google Tensor||Snapdragon 888||A15 Bionic|
|CPU||Octa-core CPU||Kryo 680, Octa-core CPU||Hexa-Core CPU|
|CPU Cores||2x 2.8GHz (Cortex-X1) 2x 2.25GHz (Cortex A76) 4x 1.8GHz(Cortex A55)||1x 2.84GHz(Cortex-X1) 3×2.4GHz(Cortex A78) 4x 1.8GHz(Cortex A55)||2x 3.2 Avalanche( High-performance) 4x 1.82 Blizzard(Energy-efficient)|
|Process Technology||Samsung’s 5nm process||Samsung’s 5nm process||TSMC’s 2nd-gen 5nm process|
|GPU||20-core Mali G78 GPU||Adreno 660||Apple-designed five-core GPU|
|Machine Learning and AI||Google Custom TPU||6rh-gen AI accelerator featuring Hexagon 780 DSP;26 TOPS||16-core Neural Engine; 15.8 TOPS|
|ISP||Google Custom ISP||Spectra 580||Apple-designed New Image Signal Processor|
|Camera capability||50MP zero-shutter lag Motion Mode||Up to 200MP; 28MP triple camera with Zero Shutter Lag; Triple 14-bit ISPs||N/A|
|Video Capability||4K 60FPS on rear camera 4K 30FPS on Front camera Google HDRnet||[email protected] 30FPS, 4K HDR Dolby @ 60FPS, 720p @ 960FPS||4K HDR Dolby Vision @60FPS|
|Modem||Samsung Exynos 5123 modem 7.3GBps Peak download and 3.6GBps Peak upload||Qualcomm X60 5G modem 7.5 GBps Peak, Download, 3GBps Peak upload||Qualcomm X60 5G modem(speculated) 7.5GBps peak download 3GBps Peak upload|
|Wi-Fi Support||Wi-Fi 6E||Wi-Fi 6E-ready||Wi-Fi 6-ready|
Google Tensor Vs Snapdragon 888 Vs A15 Bionic: CPU
Let us talk about the Google Tensor’s CPU capability. Of late, we are seeing companies putting less focus on CPU and assigning more resources to GPU, AI, and ML co-processors.
We observed the same trend during our in-depth comparison between Apple M1 vs M1 Pro vs M1 Max. And a similar trend is at play with Google’s Tensor chipset. Right off the bat, Google is not chasing numbers with its Tensor chip and instead focused on delivering “experiences” with day-to-day tasks.
Google Tensor does pack an octa-core CPU with two power-hungry Cortex-X1 cores, but both are restricted IP to 2.8GHz frequency, which is pretty unusual in the SoC landscape.
Another unusual bit is that Google has chosen the older, dual Cortex-A76 core in place of the newer Cortex-A78 cores.
Finally, you have the traditional four power-efficient A55 cores on the Google Tensor SoC. All of this unusual but might sound rookie, but in an interview with ArsTechnica, the Google Silicon team explained the rationale behind this approach.
As per Phil Carmack, the Vice President and General Manager of Google Silicon, the idea is to choreograph the interplay between the CPU, GPU, ISP, AI, and ML co-processors rather than completing a task by putting a high-performant core into action.
Google is calling it “Heterogeneous computing,” where all co-processors come together to achieve a task.
For example, if you open the camera app, all processor modules are called into action for computing, scene and face detection, handling buffers, read/write operations, and more.
In the scenario, the dialled-down Cortex-X1 core works at a medium workload, consumes less power, and improves the battery life overall.
The focus seems to be on providing sustained performance by allocating the task intelligently and not adversely affecting the battery life.
Similarly, Google is using the older A76 core because it offers 20% more sustained performance in the same power envelope as the A78. In computing, when a task is called into action, processors tend to finish it as soon as possible so that they can go into a lower-power state to save battery life.
Google is not taking this approach, instead, it has utilized medium-level cores to deliver sustained performance over a longer period of time. This way, less heat is generated, and battery life is preserved on the Pixel 6 and Pixel 6 Pro.
This is corroborated by Ron Amadeo of ArsTechnica, where he recorded a [email protected] 60FPS video on the Pixel 6 for 20 minutes straight and faced no overheating issues.
With an 80% overhead, it seems Google Tensor comes close to Snapdragon 870 SoC, which is frankly not a bad standing, more so when you are promised 24-48 hours of battery life.
In comparison to Snapdragon 888 (1100 single-core Geekbench score), Google Tensor’s CPU(1,000 single-core Geekbench scores) is a notch below and significantly behind Apple’s A15 Bionic(1,700 single-core Geekbench scores).
But again, we will have to wait for real-life tests and benchmark numbers to ascertain how well Google’s approach has paid off in its SoC design.
Google Tensor Vs Snapdragon 888 Vs A15 Bionic: GPU
Now coming to GPU, where most of the action is happening nowadays. It seems Google has gone all-in and packed a powerful 20-core Mali G78 GPU on its first Tensor chip.
To give you a perspective, even Samsung has used the 14-core Mali-G78 S21 Ultra, but Google wants to have more power in its hand drive GPU-intensive tasks and deliver a fluid and smooth experience without a hiccup.
Google says GPU on Tensor is 370% faster than Pixel 5, which packed the decent Adreno 620 GPU. By estimation, it seems the Google Tensor’s GPU will perform much better than Snapdragon 888’s 660 GPU.
We say this because the 14-core Mali G78 GPU on the S21 Ultra already rivals the Snapdragon 888 SoC with very similar scores.
And with a 20-core GPU, Google may have a powerhouse on its hand and could certainly surpass Snapdragon 888 both in synthetic tests and real-life usage.
If we pit Google Tensor’s GPU against A15 Bionic’s custom GPU, assuming Google Tensor is better than Snapdragon 888 due to the 20-core GPU, I think it will match Apple’s latest A-Series mobile chip.
Snapdragon 888 was already 10-15% behind the A15 Bionic, and with a beefy GPU on Google Tensor, it is highly likely we have a winner in the Android world.
Google Tensor Vs Snapdragon 888 Vs A15 Bionic: ISP
Google has not shared many details about its ISP, except it can now perform computational photography on each frame of a video(touted as Google HDRnet), which is mind-blowing, and yes, we are talking about a [email protected] 60FPS video. You will be able to shoot HDR videos at 4K @60FPS in the signature Pixel look.
Apart from that, the ISP on Google Tensor can take a zero-shutter lag image of 50MP, which is pretty good. It can handle two streams of cameras and produce sharp images of moving objects, a feature that Google is calling “Motion mode.”
On the other hand, the Spectra 580 ISP on Snapdragon 888 is pretty powerful and can handle HDR concurrent streams from three cameras. It allows the ISP to collect more light and quickly create an image stack for a better shot.
The Spectra 580 ISP can shoot images up to 200MP and record videos up to 8K at 30FPS.
Coming to the A15 Bionic, it has a newly-designed ISP that does a great job with Apple’s computational photography features, including the new Cinematic mode, live filters, and more.
The ISP here is also capable of shooting 4K HDR Dolby Vision videos at 60FPS, which is just mind-blowing, not to mention, the ISP can also handle ProRes RAW footage at 4K @ 30FPS.
Google Tensor Vs Snapdragon 888 Vs A15 Bionic: AI and ML
Google says that one of the major reasons for developing its own custom processor was to realize its vision of AI and ML capabilities on smartphones.
Google felt that despite having all the AI smarts and advanced ML models, it was unable to deliver the experience it was envisioning for Pixel users.
So Google partnered with Samsung and its in-house Tensor team to create a custom TPU or Tensor Processing Unit. In a way, TPU is the heart and soul of the new Google Tensor chip.
The company wants to do more meaningful things with AI, thus improving the user experience and helping users perform day-to-day tasks with ease.
For instance, the on-device translation uses an offline ML model to live translate various languages. Then, there is scene and face detection, hands-free Gboard dictation with punctuation support, Motion mode, Direct My Call feature, Tone Mapping, and a plethora of things, all powered by Google’s TPI unit.
There’s also a new Magic Eraser feature that lets you remove photobombers or distractions from the picture with a click. And it also works with not just pictures you click on the Pixel 6 but also on older pictures as well.
As a matter of fact, Google can pull off HDR videos on the Pixel 6 Series, due to the Tensor Processing Unit. The Google Tensor also packs in the Tensor Security Core and Titan M2 Security chip for increased security, privacy, and on-device AI and ML operations.
Talking about the Snapdragon’s 888 AI accelerator, the Hexagon 780 DSP can deliver up to 26 TOPS. Again, it is important to note that all these numbers are meaningless if you do not find features that can actually take advantage of AI and ML capabilities.
This is where Google shines because it controls both hardware and software verticals.
In comparison, the new 16-core Neural Engine on the A15 Bionic can perform 15.8 TOPS. And as you may already know, Apple is also using its neural engine for a host of things, including computational photography, live filters, and more.
However, when it comes to AI and ML, Google Tensor will let you get an experience of an AI-driven world.
Google Tensor Vs Snapdragon 888 Vs A15 Bionic: Connectivity
Instead of Qualcomm’s modem, Google has integrated Samsung’s Exynos 5123 modem in its first custom Tensor chip. It supports both mmWave and sub-6GHz 5G networks.
Theoretically, it supports download speeds of up to 7.3GBps and upload speeds of up to 3.6GBps. In addition to this, you have Wi-Fi 6E and Bluetooth 5.2. Moving to Snapdragon 888 and A15 Bionic, both come with Qualcomm’s X60 5G mode, which has a peak download speed of 7.5GBps and a peak upload speed of 3GBps.
The Snapdragon X60 5G modem supports both mmWae and sub-6GHz bands, along with carrier aggregation. You also get the latest Bluetooth 5.2 and Wi-Fi 6E technology support on the Snapdragon 888.
The A15 Bionic comes with Bluetooth 5 and Wi-Fi 6E. So yeah, all three chipsets are evenly matched on the connectivity front.
Conclusion: Who takes it all?
If we just talk about the CPU, Apple’s A15 Bionic is the best, followed by the Snapdragon 888 and the Tensor. However, computing has gotten pretty complex in today’s world and there are a lot of other factors we should keep in mind before assessing mobile chipsets.
Google Tensor is a promising new entrant that prioritizes offering a meaningful experience rather than chasing benchmark numbers.
Which chip is the best in your opinion?