The Pixel 6 and the Pixel 6 Pro are the first two devices to come with Google’s custom Tensor silicon chip instead of the mainstream Snapdragon 888. At the Pixel 6 launch event, Google devoted most of its efforts detailing the new Tensor system-on-a-chip (SoC). Hailing it as the most powerful mobile chipset, Google said it has applied its Machine Learning (ML) knowledge to bring onboard A.I. capabilities to a smartphone with the new chipset. The claim shall be put to test eventually when reviewers compare it with Qualcomm’s premium tier chipset — the Snapdragon 888 and Snapdragon 888 Plus — and Apple’s latest A15 Bionic chip.
With the Pixel 6, Google may finally be ready to take on Apple, and a vital weapon in this face-off will be its custom Tensor system-on-chip. But before it can challenge the big dog, first, we need to see how the Google Tensor stacks up versus the Qualcomm Snapdragon 888.
Why Tensor in the first place?
- 1 Why Tensor in the first place?
- 2 Diving into the hardware
- 3 Google Tensor SoC uses a 20-core GPU and a Samsung 5G modem
- 4 Google Tensor is big on security
- 5 A.I. is Tensor’s primary purpose
- 6 Why is Tensor essential for Pixel?
The Google Pixel 6 was never a well-kept secret. Ahead of the launch, there were plenty of compelling leaks and official certifications revealing key details about the upcoming smartphones. Google even formally announced the Tensor chip over two months before the launch and later teased the design of Pixel 6 and the Pixel 6 Pro at its offline store in New York City. Therefore, Google instead used most of its time at the launch event talking about Tensor’s virtues.
The Pixel — as evidenced by its name — has been devoted to not only improving photography on smartphones in the lineup but also opening up APIs for other manufacturers to adopt for better photography on their devices. While the entire smartphone industry has been relying on bigger camera sensors and higher megapixel count on their flagship smartphone camera, Google has always emphasized its computational photography algorithms can outpace advancements in terms of camera hardware throughout the history of the Pixel family.
But despite the advanced software features, Google’s hesitation to upgrade the camera sensors on its flagship devices led to a quick decline in interest in the Pixel phones. The tech giant is finally making conscious efforts to address this by opting for much-improved camera hardware to complement its outstanding camera software. Nonetheless, all of these efforts would not be as effective as they are with Google’s custom chipset that allows it to maximize the performance efficiency of the new Pixel phones.
Diving into the hardware
The Google Silicon team outlined tidbits of the new Tensor SoC, including its design, core count, and dedicated security features. This goes on to confirm many of the leaks and speculations we know of the Tensor chip, which was previously addressed to by its codename, “Whitechapel.” The following paragraphs discuss its details.
Tri-cluster, 8-core CPU with an edge
Like most other chipmakers, Google has licensed IP from ARM to design a custom mobile silicon. The Google Tensor is equipped with an eight-core CPU consisting of two ARM Cortex-X1 cores, two Cortex-A76 cores, and four Cortex-A55 cores that are based on a 5nm design, the company revealed to ArsTechnica.
Based on this information, we can see why the Google Tensor is touted to have an edge over other competitor chipsets such as Samsung’s Exynos 2100 and the Snapdragon 888 or Snapdragon 888 Plus. Both of the other chipsets also feature a tri-cluster design, like Tensor, but come with a single ARM Cortex-X1 core along with three Cortex-A78 cores and four Cortex-A55 cores.
Here’s a quick comparison of CPU core configuration and clock speeds for different cores on the Google Tensor, Snapdragon 888, Snapdragon 888 Plus, and Exynos 2100 chipsets:
|SoC||Google Tensor||Qualcomm Snapdragon 888/888 Plus||Samsung Exynos 2100|
Tensor prioritizes efficiency
Phil Carmack, a VP at Google and the general manager of Google Silicon, told ArsTechnica the company’s reasoning behind choosing two of ARM’s Cortex-X1 cores instead of just one. Carmack says the CPU will be able to divide the load between the two Cortex-X1 cores, even for moderately significant tasks, and this will contribute to more efficient performance.
Carmack illustrates a use case by sharing a camera example. From recording to rendering, and from Google Lens detection to machine learning function, multiple tasks are happening all at once when the camera is being used. As a result, several components of the SoC are required to work in harmony. Besides the camera hardware, the CPU, the GPU, ISP (Image Signal Processor), and the ML processing unit all combine forces to contribute to a lag-free camera experience.
If Google were to stick with a single performance Cortex-X1 core on the Tensor — as is the case with its Snapdragon and Exynos counterparts, this workload would fall back to the “medium” Cortex-A76 cores running at full capacity but still lag. In contrast, two Cortex-X1 cores can execute the same workload with greater efficiency and lower power consumption than the medium cores. A higher power efficiency while performing tasks translates to lower heat generation and a better battery backup.
Notably, the Pixel 5 or the Pixel 4a 5G, which used the Snapdragon 765G chipset, were plagued with severe heating issues, especially while using the camera. A custom CPU architecture, therefore, should — in theory — allow the Pixel 6 and the Pixel 6 Pro to allocate resources more optimally.
On one hand, while Google chooses to go all-in with two Cortex-X1 cores instead of one, it is a bit shocking to see Tensor using at least three-generations-old medium cores. The Snapdragon 888 and the Exynos 2100 use medium cores based on Cortex-A78, which is relatively more efficient than the Cortex-A76 deployed on Tensor. Google, unfortunately, did not bother to offer any sound reasoning for this.
Furthermore, for low-intensity operations like maintaining the Always-On Display (AOD) and Now Playing, the Google Tensor has a special Context Hub. Once again, a dedicated unit for tasks with low power consumption is a step towards more power efficiency.
Google Tensor SoC uses a 20-core GPU and a Samsung 5G modem
Alongside the tweaked CPU design, Google Tensor was earlier reported to feature a Mali-G78 GPU — the same as the Exynos 2100. Google says this is a 20-core graphics processor, which is specially designed to deliver premium gaming performance. It also claims the GPU has 370% better performance than the one on the Pixel 5. The real-world performance will only be known once we have the devices to run graphics benchmarks and test games on them.
The Google Tensor is likely to rely on Samsung’s Exynos 5123 modem for its 5G capabilities in most markets instead of opting for a Qualcomm modem. Cues pointing to the existence of a Samsung modem on the Google Pixel 6 and the Pixel 6 Pro were first spotted in Android 12 beta by XDA and later confirmed in a report by Reuters.
The Exynos modem supports both Sub-6GHz and mmWave 5G frequencies. But recent findings suggest that only certain carrier-locked variants of the Pixel 6 support both types of 5G signals while the unlocked models support only Sub-6GHz 5G. This means, not all Pixel 6 models will be created equally but Digital Trends’ Erika Rawes says that it really doesn’t matter.
So, the unlocked Google Pixel 6 does NOT support mmWave 5G. It's sub-6GHz only. The Verizon model (not sure about AT&T and T-Mo yet) does include mmWave in the Pixel 6, which is why it costs $100 more than the unlocked model. # GooglePixel6Pro #GooglePixel
— Z (@ericmzeman) October 19, 2021
Google Tensor is big on security
The Google Tensor chipset features the second generation of its dedicated security chip — the Titan M2. The Titan M2 is the successor to the first-gen Titan security chip that has been present on the premium Pixel smartphones since the Google Pixel 3. Google says the new security chip is designed to protect sensitive data such as passwords and PINs against online breaches as well as physical attack techniques including “electromagnetic analysis, voltage glitching, and even laser fault injection.”
Alongside the Titan M2 chip, the Pixel 6 smartphones will also feature a Tensor Security Core — a CPU-based sub-system that is specially designed to run sensitive tasks in isolation so other apps do have access to this data.
Despite claims about its performance, Google did not build a custom silicon to offer a higher power efficiency than Qualcomm or other competitors. The main reason, as Google shared unapologetically, is to provide a stable and secure platform to execute artificial intelligence (A.I.) and machine learning (ML) tasks on the smartphone itself, without relying on a cloud infrastructure. In fact, the chipset draws its name from Google’s Tensor Processing Units or A.I.-accelerated processors used in its data centers.
In hindsight, Google could be dropping hints about a custom SoC by introducing dedicated A.I.-centric chips, including the Pixel Visual Core and the Pixel Neural Core.
Besides the optimized CPU, Google Tensor SoC also features a dedicated TPU — commonly known as an NPU or a neural processing unit — to perform A.I.-based applications on the Pixel 6 and the Pixel 6 Pro. Because of its nature and Google’s expertise with machine learning, the Tensor is designed to run machine learning models on the devices themselves.
This advanced architecture allows Tensor to accomplish complex tasks such as Automatic Speech Recognition (ASR), which will actively translate any other language to your phone’s default language in apps like Messages, WhatsApp, and Recorder or even tools such as Live Caption. Additionally, improved speech recognition also allows Tensor to interpret pauses and punctuations in speech more accurately and using only half the amount of power as previous Pixel phones.
In addition to better speech processing, the Tensor brings significant improvements to photography. First of all, the chipset now facilitates computational videography — besides photography — using Google’s HDRNet. This machine-learning algorithm ensures the Pixel 6 and the Pixel 6 Pro capture the most vivid and accurate colors in each frame. Tensor also facilitates features like Face Unblur — to fix blurred faces in moving photos, Magic Eraser — to patch unwanted objects from images, and a better perception of skin tones for people of color.
Why is Tensor essential for Pixel?
As Google relentlessly repeated throughout the Pixel 6 launch event, Tensor guarantees that Google’s latest advancements in A.I. can be delivered directly on its latest and upcoming mobile phones. This would be difficult to achieve with a generic SoC such as the Snapdragon 888, especially with limited control over Qualcomm’s chipset design process.
Another reason Google chose a custom SoC with two ARM Cortex-X1 cores instead of just one is to ensure more power efficiency and fewer heat-related losses. Unlike previous Google smartphones such as the Pixel 5, the new Pixel 6 smartphones are less likely to heat up while running routine tasks such as capturing 4K video. The Snapdragon 888 and Exynos 2100 have also been criticized for poor heat management to compensate for initial higher performance. However, greater amounts of heat for prolonged periods can lead to throttling and eventually reduce performance, thereby forfeiting the main goal of higher performance.
One last reason behind Google’s choice of a custom SoC is to draw the world’s attention towards its endeavors to reclaim its lost dominance in the smartphone world. The biggest smartphone brands including Samsung, Apple, and Huawei, already make their own custom chipsets, while OPPO has also been working on its custom chipset reportedly. All this makes it essential for Google to go the extra mile and prove its competence to stay relevant in the smartphone industry.
Above article first published by Source link . We curated and re-published.