The Future of Pixel Devices

Google has always been at the forefront of innovation, especially when it comes to its Pixel devices. Pixel phones are known for their amazing cameras, software features, and seamless integration with Google services. But what if Google could make its hardware, not just software, to power its Pixel devices? That is exactly what Google Tensor is all about.

Tensor comes from Google’s machine learning technology, Tensor Processing Unit (TPU), which is integrated into the chip. The TPU is a specialized hardware accelerator that can perform complex mathematical operations on large amounts of data, such as matrices and tensors, much faster and more efficiently than conventional CPUs or GPUs. The TPU enables Google Tensor to run advanced artificial intelligence applications on the device, without relying on cloud servers or internet connectivity.

Some of the main features and benefits of Google Tensor are:

Google Tensor is a system-on-chip (SoC) processor that was co-designed by Google and Samsung for Pixel phones. It debuted with the Pixel 6 in 2021 and was upgraded to Tensor G3 in the Pixel 8 series in 2024. Tensor is focused on delivering complex artificial intelligence (AI) and machine learning (ML) processes on the phone itself, rather than relying on cloud computing. This enables Pixel phones to perform faster, more personalized, and more helpful tasks, such as:

Enhanced Imaging and Video

Tensor allows Pixel phones to capture high-quality photos and videos with advanced features, such as snapping photos from multiple cameras at once, recording from multiple cameras at different exposures, faster shutter speed, and more processing happening while you record video. Tensor also enables Pixel phones to apply computational photography techniques, such as HDR+, Night Sight, Portrait Mode, and Super Res Zoom, to improve the image quality and details.

Improved Security and Performance

Tensor provides Pixel phones with powerful security and performance capabilities, such as:

  • Titan M2: A dedicated security chip that protects the device from tampering, malware, and phishing attacks. It also enables features like Locked Folder, which lets you store sensitive photos and videos in a secure space that only you can access with your fingerprint or password.
  • Tensor Security Core: A hardware module that encrypts and decrypts data on the device, ensuring that your personal information is protected even if your phone is lost or stolen.
  • Tensor Memory Engine: A technology that optimizes the memory usage and allocation on the device, resulting in faster app launch times, smoother multitasking, and longer battery life.

More Intelligent Speech Recognition

Tensor enables Pixel phones to understand and process natural language better, even without an internet connection. This allows Pixel phones to offer more intelligent speech recognition features, such as:

  • Live Translate: A feature that lets you translate speech and text in real-time across 89 languages, with the option to use the camera to scan and translate text from images.
  • Live Caption: A feature that automatically generates captions for any media playing on your device, including videos, podcasts, and phone calls.
  • Voice Typing: A feature that lets you dictate text to your device with high accuracy and punctuation, and also edit and format your text with voice commands.

Google Tensor is a key component of Google’s vision for the future of Pixel devices and the smartphone industry. By creating its chip, Google aims to differentiate its Pixel devices from the competition and to provide a unique and seamless user experience. Google Tensor also represents Google’s commitment and investment in machine learning and artificial intelligence, which are expected to play a vital role in the next generation of computing and communication.

Tensor G3: The Current Generation

The Tensor G3 is the latest iteration of Google’s custom chip for its Pixel devices. It powers the Pixel 8 and Pixel 8 Pro, launched in October 2023. The Tensor G3 is manufactured by Samsung using a 4nm process, which offers improved performance and efficiency over the previous 5nm process used for the Tensor G2.

CPU: A Mixed Bag

The Tensor G3 has a 9-core CPU configuration, consisting of one Cortex-X3 core at 2.91GHz, four Cortex-A715 cores at 2.37GHz, and four Cortex-A510 cores at 1.7GHz. These are based on Arm’s 2022 CPU designs, which offer higher performance per watt and better machine-learning capabilities than the older Cortex-X1, A78, and A55 cores used in the Tensor G2. However, the Tensor G3 does not use the latest Cortex-X4, A720, and A520 cores that are available in 2023, which means it lags behind some of its competitors in terms of raw CPU power.

GPU: A Modest Upgrade

The Tensor G3 also features a Mali-G715 GPU with seven cores, which is a 15% improvement over the Mali-G710 GPU with six cores in the Tensor G2. The Mali-G715 is based on Arm’s Valhall architecture, which supports ray tracing and variable rate shading for enhanced graphics quality and performance.

TPU: The Star of the Show

The Tensor G3 also has a third-generation Tensor Processing Unit (TPU), which is a specialized hardware accelerator for machine learning tasks. The TPU can perform complex mathematical operations on large amounts of data, such as matrices and tensors, much faster and more efficiently than conventional CPUs or GPUs. The TPU enables the Tensor G3 to run advanced artificial intelligence applications on the device, without relying on cloud servers or internet connectivity.

Pixel Features: Powered by Tensor G3

The Tensor G3 enables some of the unique capabilities and experiences that the Pixel 8 series offers. For example, the Tensor G3 powers the Magic Eraser feature, which allows users to remove unwanted objects from their photos with a single tap. The Tensor G3 also enables the Face Unblur feature, which can automatically detect and sharpen blurry faces in photos. Another feature that the Tensor G3 supports is the Live Translate feature, which can translate speech and text in real-time across different languages, even when offline.

Comparison with Other Flagship SoCs

The Tensor G3 is a competitive chip in the flagship smartphone market, but it is not the best in every aspect. Compared to other flagship SoCs, such as the Snapdragon 8 Gen 3 and the Exynos 2400, the Tensor G3 has some advantages and disadvantages. For instance, the Tensor G3 has a higher CPU frequency than the Exynos 2400, but a lower CPU frequency than the Snapdragon 8 Gen 3. The Tensor G3 also has a smaller GPU than the Exynos 2400, but a larger GPU than the Snapdragon 8 Gen 3. The G3 has a similar TPU performance to the Exynos 2400, but a lower TPU performance than the Snapdragon 8 Gen 3. The Tensor G3 also has a similar modem performance to the Snapdragon 8 Gen 3 and the Exynos 2400, supporting 5G sub-6GHz and mmWave bands.

The table below summarizes the main differences between the Tensor G3, the Snapdragon 8 Gen 3, and the Exynos 2400:

SoC CPU GPU TPU Modem
Tensor G3 1x Cortex-X3 @ 2.91GHz 4x Cortex-A715 @ 2.37GHz 4x Cortex-A510 @ 1.7GHz Mali-G715 MP7 Third-gen TPU Exynos 5300 (up to 7.3Gbps down, 3.87Gbps up)
Snapdragon 8 Gen 3 1x Cortex-X4 @ 3.3GHz 3x Cortex-A720 @ 3.2GHz 2x Cortex-A720 @ 3GHz 2x Cortex-A520 Refresh @ 2.3GHz Adreno (ray tracing support) Hexagon DSP X75 (up to 10Gbps down, 3.5Gbps up)
Exynos 2400 1x Cortex-X4 @ 3.21GHz 2x Cortex-A720 @ 2.9GHz 3x Cortex-A720 @ 2.6GHz 4x Cortex-A520 @ 2GHz Samsung Xclipse 940 Dual NPU Exynos 5300 (up to 9.64Gbps down, 2.55Gbps up)

Tensor G4: The Next Step

The Tensor G4 is the next iteration of Google’s custom chip for its Pixel devices. It is expected to power the Pixel 9 and Pixel 9 Pro, which are rumored to launch in October 2024. The Tensor G4 is still a semi-custom chip, co-designed with Samsung’s System LSI division, and will use an updated Samsung 4nm process, which is the same as the Tensor G3.

CPU

The Tensor G4 will offer minor improvements over the Tensor G3 in terms of specifications and performance. The Tensor G4 will feature one Cortex-X4 CPU core at 3.21GHz, which is the latest and fastest prime core from Arm’s 2023 CPU designs. The Cortex-X4 core will be accompanied by two Cortex-A720 cores at 2.9GHz, three Cortex-A720 cores at 2.6GHz, and four Cortex-A520 cores at 2GHz. These are the same CPU cores that are used in the Exynos 2400, Samsung’s flagship SoC for 20232. However, the Tensor G4 will have a lower CPU frequency than the Exynos 2400, which means it will have a lower raw CPU power.

GPU

The Tensor G4 will also feature a new GPU, the Immortalis-G715, which is based on Arm’s Valhall 2nd gen architecture. The Immortalis-G715 will have eight cores, which is one more than the Mali-G715 in the Tensor G3. It will also support ray tracing and variable rate shading, which are advanced graphics techniques that can improve the quality and performance of games and other applications. The Immortalis-G715 will have a similar GPU frequency to the Mali-G715, which is 848MHz. However, the Immortalis-G715 will have a higher GPU performance than the Mali-G715, due to the extra core and the new features.

TPU

The Tensor G4 will also have a fourth-generation Tensor Processing Unit (TPU), which is a specialized hardware accelerator for machine learning tasks. The TPU can perform complex mathematical operations on large amounts of data, such as matrices and tensors, much faster and more efficiently than conventional CPUs or GPUs. The TPU enables the Tensor G4 to run advanced artificial intelligence applications on the device, without relying on cloud servers or internet connectivity. The Tensor G4 will have a higher TPU performance than the Tensor G3, due to the improved design and optimization.

Pixel 9 Features

The Tensor G4 will enable some of the new and improved capabilities and experiences that the Pixel 9 series will offer. For example, the Tensor G4 will power the new camera system, which is rumored to include a 200MP main sensor, a 32MP telephoto sensor, and a 32MP ultra-wide sensor. The Tensor G4 will also enable the new video recording features, such as 8K at 60FPS, 4K at 120FPS, and HDR10+. Another feature that the Tensor G4 will support is the new voice assistant, which is rumored to be faster, smarter, and more conversational than the Google Assistant.

Comparison with Other SoCs

The Tensor G4 is a competitive chip in the flagship smartphone market, but it is not the best in every aspect. Compared to other flagship SoCs, such as the Snapdragon 8 Gen 3 and the Exynos 2400, the Tensor G4 has some advantages and disadvantages. For instance, the Tensor G4 has a higher CPU frequency than the Snapdragon 8 Gen 3, but a lower CPU frequency than the Exynos 2400. The Tensor G4 also has a smaller GPU than the Snapdragon 8 Gen 3, but a larger GPU than the Exynos 2400. The G4 has a similar TPU performance to the Snapdragon 8 Gen 3 and the Exynos 2400. The Tensor G4 also has a similar modem performance to the Snapdragon 8 Gen 3 and the Exynos 2400, supporting 5G sub-6GHz and mmWave bands.

Tensor G5: The Fully Custom Chip

The Tensor G5 is the future iteration of Google’s custom chip for its Pixel devices. It is reportedly in development for the Pixel 10 and Pixel 10 Pro, which are expected to launch in October 2025. The Tensor G5 will be Google’s first fully custom chip, meaning that it will not use any components or designs from Samsung or other third-party manufacturers. Instead, Google will design and optimize every aspect of the chip, from the CPU and GPU cores to the memory and modem modules. The Tensor G5 will also be manufactured by TSMC, the world’s leading semiconductor foundry, instead of Samsung, which has been Google’s partner for the previous Tensor chips.

The Tensor G5 will be a major leap forward for Google’s Pixel devices and the smartphone industry. By creating its own fully custom chip, Google will have more control and flexibility over the hardware and software integration of its Pixel devices, as well as the ability to tailor them for specific tasks and features. The Tensor G5 will also benefit from TSMC’s advanced manufacturing process and technology, which will offer higher performance and lower power consumption than Samsung’s process. The Tensor G5 will also leverage Google’s expertise and innovation in machine learning and artificial intelligence, which are expected to play a vital role in the next generation of computing and communication.

However, the Tensor G5 will also face some potential advantages and challenges.

Some of the possible advantages are:

  • Higher performance: The Tensor G5 will likely use the latest and fastest CPU and GPU cores from Arm’s 2024 CPU and Valhall 3rd gen GPU designs, which will offer higher performance per watt and better machine learning capabilities than the previous cores used in the Tensor G4. The Tensor G5 will also have a fifth-generation Tensor Processing Unit (TPU), which will be a specialized hardware accelerator for machine learning tasks. The TPU will be able to perform complex mathematical operations on large amounts of data, such as matrices and tensors, much faster and more efficiently than conventional CPUs or GPUs. The TPU will enable the Tensor G5 to run advanced artificial intelligence applications on the device, without relying on cloud servers or internet connectivity.
  • Lower power consumption: The Tensor G5 will likely use TSMC’s 3-nanometer manufacturing process, which will offer improved performance and efficiency over the previous 4-nanometer process used for the Tensor G4. The Tensor G5 will also use Integrated Fan-Out (InFO) packaging technology, which will reduce the thickness and increase the power efficiency of the chip by eliminating the need for a substrate layer between the chip and the board. The Tensor G5 will also have a dedicated power management unit (PMU), which will optimize the power consumption of the chip based on the workload and the battery level.
  • More differentiation and innovation: The Tensor G5 will be Google’s first fully custom chip, which will give Google more freedom and creativity to differentiate its Pixel devices from the competition, and to provide a unique and seamless user experience. The Tensor G5 will also allow Google to innovate and experiment with new features and capabilities that are not possible or feasible with other chips, such as quantum computing, neuromorphic computing, and brain-computer interfaces.

Some of the possible challenges are:

  • Compatibility issues: The Tensor G5 will be Google’s first fully custom chip, which means that it will not be compatible with the standard Android operating system or the applications that are designed for other chips. Google will have to modify and optimize the Android operating system and the Google apps for the Tensor G5, as well as provide support and updates for the chip. Google will also have to work with other app developers and service providers to ensure that their apps and services work well with the Tensor G5, or risk losing customers and market share.
  • Development and production costs: The Tensor G5 will be Google’s first fully custom chip, which means that Google will have to invest more time and money in the research and development of the chip, as well as the testing and validation of the chip. Google will also have to pay more for the manufacturing and packaging of the chip, as TSMC’s 3-nanometer process and InFO technology are more expensive and complex than Samsung’s 4-nanometer process and conventional technology. Google will have to balance the costs and benefits of the Tensor G5 and decide whether it can afford to sell the Pixel devices at a competitive price or a premium price.
  • Market and consumer acceptance: The Tensor G5 will be Google’s first fully custom chip, which means that it will be a new and unfamiliar product for the market and the consumers. Google will have to educate and persuade the market and consumers about the advantages and benefits of the Tensor G5, as well as the features and experiences that the Pixel devices offer. Google will also have to overcome the skepticism and resistance that some consumers may have towards Google’s products and services, especially in terms of privacy and security.

Conclusion

In this article, we have explored Google Tensor, the series of custom system-on-chip (SoC) processors designed by Google for its Pixel devices. We have discussed the specifications, performance, features, and benefits of the different generations of Tensor, from the first-generation Tensor that debuted on the Pixel 6 series in 2021, to the rumored fifth-generation Tensor that is expected to power the Pixel 10 series in 2025. And we have also compared Tensor with other flagship SoCs in the smartphone market, such as the Snapdragon 8 Gen 3 and the Exynos 2400, and highlighted some of the advantages and disadvantages of Tensor.

We have seen how Google Tensor is shaping the future of Pixel devices and the smartphone industry, by giving Google more control and flexibility over the hardware and software integration of its Pixel devices, as well as the ability to optimize them for specific tasks and features, especially those related to machine learning and artificial intelligence. We have also seen how Google Tensor is a key component of Google’s vision and innovation for the next generation of computing and communication, as well as its commitment and investment in machine learning and artificial intelligence.

#Future #Pixel #Devices

Leave a Comment