Fact-checked by Grok 2 weeks ago

Google Tensor

Google Tensor is a family of custom-designed, ARM-based system-on-chip (SoC) processors developed by exclusively for its Pixel smartphones, emphasizing advanced (AI) and capabilities to deliver seamless on-device experiences. First unveiled in October 2021 with the and Pixel 6 Pro, the original Tensor SoC marked Google's inaugural effort in mobile chip design, integrating a dedicated (TPU) to accelerate AI tasks such as , voice recognition, and real-time translation. Subsequent generations have built upon this foundation, iteratively improving performance, power efficiency, and AI integration. The Tensor G2, introduced in the Pixel 7 series in 2022, enhanced security with the Titan M2 module and supported expanded AI features like advanced face unlock. In 2023, the Tensor G3 powered the Pixel 8 lineup, featuring upgraded ARM Cortex CPU cores, a more efficient GPU, and a next-generation TPU for running generative AI models. The Tensor G4, debuting in the Pixel 9 series in 2024, optimized for Gemini AI models, delivered faster on-device processing for features like live video summaries and enhanced accessibility tools. By 2025, the Tensor G5 in the Pixel 10 series shifted to TSMC's node, providing up to 60% more AI processing power via its fourth-generation , 34% faster CPU performance, and new capabilities such as motion deblur in imaging and proactive AI assistants. These SoCs distinguish themselves through tight hardware-software integration with and Google's , prioritizing AI-driven functionalities over raw performance, while incorporating robust to protect user data. Early generations were fabricated by , but the transition to for Tensor G5 reflects Google's push for greater efficiency and customization in mobile AI . Overall, the Tensor series has enabled devices to lead in innovative features like Magic Editor for photos, Call Screen for spam reduction, and on-device Gemini Nano for privacy-focused AI interactions.

Development

Background

Google's Pixel smartphones initially relied on Qualcomm's Snapdragon system-on-chips (SoCs) for their processing power. The original and , released in 2016, featured the Snapdragon 821. This was followed by the Snapdragon 835 in the series of 2017, the Snapdragon 845 in the series of 2018, the Snapdragon 855 in the series of 2019, and the Snapdragon 765G in the of 2020. These off-the-shelf processors provided reliable performance but limited Google's ability to fully optimize hardware for its software ecosystem, particularly in areas like and . To address this, shifted to custom silicon with the introduction of the Tensor SoC series, debuting in the and Pixel 6 Pro smartphones launched in October 2021. Development of Tensor began in , following Google's acquisition of key talent from HTC's chip design team, as part of a broader effort to build in-house expertise in mobile design. This move aligned with industry trends toward custom silicon, exemplified by Apple's A-series processors since the in 2010 and Samsung's SoCs starting around the same period, which allowed device makers to tailor hardware for specific features and efficiency. Google first teased Tensor on August 2, 2021, positioning it as the company's first custom-built mobile designed exclusively for devices, with a strong emphasis on accelerating on-device and tasks. To expedite development, partnered with for both design collaboration—leveraging Samsung's expertise for elements like CPU cores—and fabrication on a node. Tensor was marketed as a premium mobile enabling -exclusive capabilities, such as advanced for features like Magic Eraser and real-time voice recognition enhancements.

Design

The Google Tensor series employs an ARMv8-based architecture, featuring custom-designed components optimized for integration with Google's software ecosystem and services. This foundation allows for tailored enhancements in areas like acceleration and system efficiency, enabling the SoC to prioritize tasks specific to and Google apps. A prominent element is the embedded TPU (Tensor Processing Unit), which serves as a dedicated for on-device inference. Central to the Tensor design are goals that emphasize on-device AI capabilities over conventional measures of CPU or GPU computational power. This approach shifts resources toward efficient execution of neural network models, reducing latency and enhancing user privacy by minimizing cloud dependency for tasks like image recognition and . Dedicated hardware support for Lite models exemplifies this focus, providing optimized pathways for deploying lightweight AI inferences in power-constrained mobile scenarios. The series adopts a modular that supports progressive refinements across generations, enabling targeted upgrades without requiring a complete redesign. Examples include scaling core configurations for better parallelism and adopting smaller process nodes to improve , allowing the to evolve in alignment with advancing algorithms and device requirements. Security forms an integral part of the Tensor design philosophy, featuring a dedicated Tensor Security Core that works with the Titan M2 chip for robust hardware-rooted protections. This dedicated security module handles encryption of sensitive data and enforces verified boot to prevent unauthorized modifications, safeguarding against sophisticated threats. Lab validations confirm Titan M2's resistance to attacks such as voltage glitching, electromagnetic analysis, and laser-based fault injection.

Manufacturing

The first-generation Google Tensor (G1) SoC was manufactured by Samsung Foundry using its 5nm process node, incorporating design elements inspired by the Samsung Exynos 2100 but with customizations tailored by Google for Pixel devices. Samsung Foundry continued production for subsequent generations, with the Tensor G2 also on the 5nm node, while the G3 and G4 shifted to a 4nm process to improve density and efficiency. These Samsung-fabricated encountered challenges, including lower yields on the 5nm that contributed to constraints and throttling issues under sustained loads, affecting overall device performance stability. In a significant shift, the Tensor G5 marked Google's transition to for fabrication on its 3nm N3E or N3P process node, representing the company's first fully in-house designed without Samsung's co-design involvement, which enabled enhanced power efficiency and reduced thermal concerns compared to prior generations. This change incorporated advanced (EUV) across multiple layers for both Samsung's earlier nodes and TSMC's 3nm process, allowing for finer patterning and higher integration densities. The foundry switch to also mitigated risks from Samsung's yield variability, supporting smoother production timelines for the 2025 lineup by leveraging TSMC's higher-volume capacity and reliability.

Architecture

Processor Cores

The processor cores in Google Tensor SoCs adopt an octa-core ARM Cortex-based design, leveraging a big.LITTLE configuration to dynamically balance high-performance computing demands with power efficiency for mobile applications. This heterogeneous setup combines high-performance cores for intensive tasks, mid-tier cores for general workloads, and efficiency cores for background operations, allowing the system to scale resources based on real-time needs. The architecture relies on ARM's DynamIQ technology, which facilitates the mixing of diverse core types within a shared cluster, enabling advanced power gating and frequency scaling to minimize thermal throttling and battery drain. Over successive Tensor generations, the core lineup has progressed through ARM's Cortex-A series, starting with models like the Cortex-A76 and Cortex-A78 for performance clusters in earlier iterations, and advancing to more efficient variants such as Cortex-A720 and A725 in later designs. For example, the Tensor G5 features an unusual 1x Cortex-X4 prime core, 5x Cortex-A725 performance cores, and 2x Cortex-A520 efficiency cores. Prime cores, typically the highest-performance units (e.g., from the Cortex-X series), are optimized for sustained high-load scenarios like multitasking or computational bursts, while the overall configuration maintains eight cores to support Android's multi-threaded ecosystem. This evolution prioritizes integration and software optimization over raw peak performance, aligning with Google's focus on AI-driven features and user experience fluidity. Clock speeds for prime cores range up to 2.8 GHz in early generations and reach 3.78 GHz in the Tensor G5, with efficiency cores operating at lower frequencies to conserve energy, supported by per-core caches and a shared L3 cache of approximately 4 MB for faster data retrieval and reduced . The cores connect to the via 's interconnect fabric, which ensures coherent memory access and efficient task migration between core clusters. Additionally, the CPU subsystem integrates with dedicated memory controllers that support LPDDR5 RAM, delivering high (up to 51.2 GB/s in dual-channel setups) essential for handling large datasets in imaging and pipelines.

Graphics and AI Accelerators

The (GPU) in Google Tensor SoCs up to the G4 is based on ARM's architecture, while the G5 uses ' PowerVR DXT-48-1536 GPU, enabling efficient rendering for mobile gaming and visual effects. Early models, such as Tensor G1, integrate the Mali-G78 with 20 execution units, supporting the API for high-performance graphics and basic ray tracing capabilities in later iterations like the Immortalis-G715 in Tensor G3. This configuration allows for smooth and higher gaming experiences, with optimizations for power efficiency in demanding titles. Complementing the GPU, Tensor incorporates a dedicated Tensor Processing Unit (TPU), an edge-optimized variant of Google's , supporting INT8 and FP16 precision for on-device inference. This TPU, evolved across generations, handles up to 12 in recent models and is tailored for compact neural networks like Gemini Nano, enabling seamless AI processing without cloud dependency. The architecture prioritizes low-latency operations for tasks, distinguishing it from general-purpose GPUs by focusing on matrix multiplications central to . These accelerators power key AI features in Pixel devices, including hardware-accelerated for secure face unlock, which uses the for rapid biometric matching, and for on-device voice interactions. Photo editing tools like Magic Eraser leverage the to intelligently remove or inpaint objects in images, applying generative models directly on the device for privacy-preserving enhancements. The GPU and TPU integrate via a shared system memory architecture, allowing efficient data exchange with the CPU for low-latency inference and rendering pipelines. This unified design minimizes data transfer overhead, enabling hybrid workloads where AI-enhanced graphics, such as real-time style transfers in camera apps, benefit from direct memory access across components.

Modem and Other Components

The Google Tensor system-on-chip () integrates a -derived 5G , starting with the Exynos 5123 in the first-generation Tensor for devices, which supports both sub-6 GHz and mmWave bands with peak download speeds of up to 7.5 Gbps and upload speeds of 3 Gbps. Subsequent generations evolved this component for improved efficiency, such as the Exynos 5300 in Tensor G2 and G3, which maintains sub-6 GHz and mmWave compatibility while enhancing power consumption and for better real-world connectivity in and 8 series. By Tensor G4 in the Pixel 9 lineup, the Exynos 5400 achieves theoretical download speeds exceeding 10 Gbps in optimized conditions, though practical prioritizes over maximum throughput. Beyond the modem, the Tensor SoC incorporates an integrated Image Signal Processor (ISP) tailored for advanced camera systems, enabling support for up to 50 MP sensors in early models like the Pixel 6's primary camera and scaling to multi-camera arrays with features such as real-time + processing. In later iterations, such as Tensor G5 for the Pixel 10 series, the fully custom-designed ISP handles 10-bit video capture at 120 , motion deblur in low light, and up to 200 MP resolution for high-end sensors, optimizing noise reduction and dynamic range without relying on external processing. The SoC also includes a display controller that drives up to QHD+ resolutions at 144 Hz refresh rates, as seen in Pro implementations, ensuring smooth visuals with support for power efficiency. Audio components feature an on-chip audio processor (AoC) that supports high-resolution codecs like LDAC and , facilitating spatial audio rendering and noise cancellation directly within the Tensor for Pixel devices. For storage and input/output, Tensor SoCs support UFS 3.1 in initial generations ( through 9), delivering sequential read speeds up to 2,100 MB/s and write speeds up to 1,200 MB/s, with inline hardware via AES-XTS for secure data handling. The Tensor G5 advances to UFS 4.0 compatibility in the 10, doubling to 4,200 MB/s reads and 2,800 MB/s writes while incorporating zoned storage for workloads, though PCIe interfaces remain limited to internal high-speed links rather than external expansion. is handled through integrated that leverage the SoC's process advancements, such as the 4nm in Tensor for 20% better efficiency over predecessors, enabling adaptive optimization that extends usage by predicting and throttling non-essential tasks. In the 3nm Tensor G5, these ICs achieve up to 30% power savings compared to 5nm designs, supporting over 30 hours of mixed-use life in 10 models through dynamic voltage and -driven states. Security features center on the Titan M series co-processor, with Titan M2 integrated alongside Tensor's dedicated security core—a RISC-V-based subsystem that verifies boot integrity and isolates sensitive operations in a hardware enclave. This combination performs end-to-end encryption for payments, biometric authentication, and firmware updates, earning Common Criteria EAL4+ certification for resistance to tampering, as implemented across Pixel 6 and later devices. The Titan M2 handles secure key storage and rollback protection independently of the main CPU, ensuring that even if the OS is compromised, critical data like private keys remains protected within the chip's isolated memory.

Models

Tensor G1

The Google Tensor G1 (also known as GS101) was announced on October 19, 2021, as part of the series launch event, marking Google's entry into custom system-on-chip () design for its smartphones. It powers the , Pro, and later the mid-range released in July 2022. Fabricated on Foundry's node, the Tensor G1 represented a shift from third-party processors like Qualcomm's Snapdragon, enabling tighter integration of Google's software and hardware for improved and imaging capabilities. The features an octa-core CPU configuration with two high-performance cores clocked at 2.80 GHz, two mid-range cores at 2.25 GHz, and four efficiency-focused cores at 1.80 GHz, providing a balance of power and for mobile workloads. is handled by the Arm Mali-G78 MP20 GPU, supporting advanced rendering for and . A custom Google Processing Unit (TPU) accelerates on-device tasks, while the integrated 5123 modem delivers connectivity, including sub-6 GHz and mmWave support, along with integrated features like and 5.2. As the inaugural custom silicon for devices, the Tensor G1 introduced key innovations centered on -driven experiences, such as Real-Time +, which applies enhancements during video recording at up to fps for more vibrant colors and dynamic range without post-processing delays. It also enabled on-device Live Translate, allowing real-time translation of conversations in apps like Messages and using less power than previous cloud-dependent methods, enhancing privacy and speed for multilingual users. These features underscored Google's focus on integrating its expertise directly into hardware for seamless, always-on capabilities.

Tensor G2

The (GS201) is the second-generation system-on-chip () developed by in collaboration with , debuting in October 2022 with the and Pixel 7 Pro smartphones. Announced on October 6 and released on October 13, it represents an iterative refinement over the original Tensor, retaining the Samsung 5nm fabrication process while incorporating optimizations for enhanced thermal management and power efficiency. This design choice allowed to focus on software-hardware integration rather than a node shrink, enabling targeted improvements in processing without major redesign costs. At its core, the Tensor G2 features an octa-core CPU configuration comprising 2× prime cores clocked at 2.85 GHz for high-performance tasks, 2× performance cores at 2.35 GHz, and 4× efficiency cores at 1.80 GHz. The (GPU) upgrades to the Arm Mali-G710 (Immortalis-G710 MP7), delivering approximately 20% better performance and efficiency compared to the Mali-G78 in the prior generation, supporting advanced rendering for gaming and . The (TPU) is refined for workloads, rated at 12 (trillion operations per second) in INT8 precision with improved architecture that boosts relevant AI tasks by up to 60% while reducing power draw by about 20%. These enhancements contribute to overall power consumption reductions of 15-20% over the Tensor G1 in mixed workloads, aiding longer battery life in AI-intensive scenarios. A hallmark of the Tensor G2 is its emphasis on on-device AI innovations, powering features like Photo Unblur, which leverages to restore sharpness in out-of-focus images captured by the Pixel's camera system, and an expanded Live Translate capability that supports real-time voice and text translation across more than 50 languages in additional apps such as and . These capabilities stem from the TPU's optimized and acceleration, enabling seamless integration of generative AI without cloud dependency. The SoC also includes an upgraded image signal processor (ISP) for faster photo and , further streamlining user experiences in the ecosystem. The Tensor G2 is deployed in several Google devices, including the , (both launched in 2022), the mid-range (2023), and the (2023). Its balanced architecture prioritizes and camera optimizations over raw computational speed, aligning with Google's strategy for intelligent .

Tensor G3

The Google Tensor G3 is a (SoC) designed by Google and manufactured by using a 4 nm process node. It was announced on October 4, 2023, and powers the , , and later the Pixel 8a smartphone lineup. The Tensor G3 features a nine-core CPU configuration consisting of one prime core clocked at 2.91 GHz for high-performance tasks, four Cortex-A715 performance cores at 2.37 GHz, and four Cortex-A510 efficiency cores at 1.7 GHz. Its graphics processing is handled by an Arm Immortalis-G715 MP7 GPU, which supports advanced rendering and workloads. The also includes a dual-core (NPU) based on Google's custom (TPU) architecture, optimized for on-device AI inference and enabling seamless integration with models like Gemini Nano for tasks such as text summarization and image understanding. Key innovations in the Tensor G3 emphasize generative AI capabilities, marking the first implementation of advanced on-device processing for features like Video Boost, which uses cloud-assisted AI to upscale videos to and enhance stabilization on the 8 Pro. It also introduces improved generative editing tools, such as expanded Magic Editor for photo manipulation and Audio Magic Eraser for video sound cleanup, all powered by the enhanced to prioritize efficiency and privacy through local computation. These advancements focus on integrating generative AI directly into everyday device interactions, setting the stage for more sophisticated on-device intelligence in the Pixel ecosystem.

Tensor G4

The Tensor G4 is the fourth-generation system-on-chip () developed by in collaboration with , debuting in August 2024 alongside the Pixel 9 series smartphones. It powers the Pixel 9, Pixel 9 Pro, Pixel 9 Pro XL, and Pixel 9 Pro Fold, marking the final Tensor iteration manufactured using Samsung's 4nm process node with specific optimizations for power efficiency. This chip emphasizes on-device processing while maintaining 's focus on custom tailored for Pixel devices. The Tensor G4 features an octa-core CPU configuration consisting of one prime core clocked at 3.1 GHz for high-performance tasks, three Cortex-A720 performance cores at 2.6 GHz, and four Cortex-A520 efficiency cores at 1.92 GHz. The GPU is an Arm Mali-G715, continuing the of Immortalis-series graphics in recent Tensor chips for improved rendering in AI-driven applications. Additionally, the integrated supports multimodal AI capabilities, enabling advanced on-device processing for features like Gemini Nano models that handle text, images, and audio inputs simultaneously. Key innovations in the Tensor G4 include enhanced satellite connectivity via the integrated , allowing Emergency SOS messaging in areas without cellular or coverage. It also powers an advanced version of Audio Magic Eraser, which uses to isolate and remove unwanted sounds from videos, such as crowd noise or wind, for cleaner audio output. Furthermore, the chip incorporates improved thermal management through optimized packaging and process refinements, resulting in better sustained performance and reduced throttling during prolonged use compared to prior generations.

Tensor G5

The Google Tensor G5 is the fifth-generation system-on-chip (SoC) developed by Google, marking the company's first fully custom-designed processor without reliance on third-party architectures like Samsung's Exynos base. Announced on August 20, 2025, alongside the Pixel 10 series, it powers the Pixel 10, Pixel 10 Pro, and Pixel 10 Pro XL smartphones. Manufactured on TSMC's 3 nm process node, this shift from previous Samsung Foundry production enables improved power efficiency and performance scaling. The Tensor G5 features an eight-core CPU configuration consisting of one high-performance prime core clocked at 3.78 GHz, five Arm Cortex-A725 performance cores at 3.05 GHz, and two efficiency cores at 2.25 GHz. This setup delivers an average 34% CPU performance improvement over the Tensor G4, enhancing multitasking and responsiveness in everyday tasks. The GPU is the PowerVR IMG DXT-48-1536, optimized for mobile gaming and graphics rendering, though it omits hardware-accelerated ray tracing to prioritize efficiency. Complementing these, the integrated (), referred to as the fourth-generation (), provides up to 60% greater computational power than its predecessor, enabling faster inference for on-device AI models. A key innovation in the Tensor G5 is its deep integration with Google's Gemini Nano large language model, allowing advanced on-device generative features such as real-time summarization and proactive assistance without cloud dependency. This enhancement supports over 20 on-device experiences, including enhanced photo editing and voice processing, while maintaining through local computation. Overall, these advancements reflect Google's focus on -centric hardware tailored for the ecosystem.

Future Models

Google's Tensor G6 is anticipated to power the 11 series, expected to launch in 2026. This next-generation represents a continuation of Google's shift toward greater customization, building on the Tensor G5's foundation by emphasizing power efficiency over raw performance gains. The Tensor G6 is reportedly set to be fabricated on TSMC's 2nm process node, a significant advancement from the 3nm node used in prior models, enabling improved thermal management and energy efficiency. Additionally, Google plans to integrate a M90 , supporting up to 12 Gbps download speeds and dual-active connectivity, marking a departure from previous modems to enhance connectivity reliability. A variant of the Tensor G6 is expected to extend to non-smartphone devices, such as the Pixel Tablet 3 and 11a, though reports as of late 2024 suggested potential cancellation of the Pixel Tablet 3. Long-term, has secured production capacity for Tensor chips through at least the Pixel 14, projected for 2029, underscoring a commitment to in-house design for sustained advancements. Developing custom SoCs like the Tensor series involves substantial upfront costs, estimated in the billions for design and tooling, which must balance against the need for seamless integration with the ecosystem and third-party components. This approach allows tailored optimizations for Google services but requires ongoing investments to match competitors' off-the-shelf efficiency without compromising compatibility.

Reception and Impact

Performance Reviews

The Google Tensor series has demonstrated steady performance evolution across its generations, with each iteration focusing on balancing computational power, efficiency, and specialized capabilities tailored for devices. Early models like the Tensor G1 prioritized integration of accelerators over raw speed, achieving moderate CPU scores in standard benchmarks while laying the foundation for on-device processing. Subsequent generations built on this by incorporating process node improvements and architectural tweaks, resulting in measurable gains in both single- and multi-core tasks, though consistently trailing flagship competitors in peak throughput. In GPU benchmarks such as GFXBench, the Tensor series has shown progressive enhancements, with the G1 scoring around 57 fps in the Manhattan ES 3.1 test on the , while later models like the G5 achieve up to 2x higher frame rates in offscreen scenarios due to optimized GPU configurations and better thermal management. For AI-specific evaluations, the Tensor excels in mobile inference tasks; for instance, in AI using TensorFlow Lite CPU, the G5 delivers single-scene scores of 2176 and multi-scene scores of 3458, marking a 18-25% improvement over the G4's 1836 and 2778, respectively. These results highlight the series' strength in real-world ML workloads like image recognition and , where dedicated accelerators provide consistent performance without excessive power draw. Compared to Qualcomm's Snapdragon 8 Gen series, Tensor chips generally lag in raw computational power—for example, the G5's score of approximately 1,030,000 falls short of the Snapdragon 8 Elite's 2,721,000 by over 2x—but outperform in -optimized scenarios. The G5 demonstrates superior efficiency in on-device transcription and tasks, processing models up to 2.6x faster than predecessors while using 2x less energy, enabling features like real-time call screening without cloud dependency. This contrast underscores Tensor's design philosophy, prioritizing integrated ecosystem optimizations over brute-force metrics seen in Snapdragon's gaming-focused architecture. Expert analyses from outlets like Android Authority praise the Tensor's generational strides, noting 20-30% uplifts in sustained load performance per iteration, such as the G5's 34% average CPU speed increase over the G4 in prolonged workloads like video encoding. Reviews from PhoneArena highlight Pixel-exclusive efficiencies, with 2025 evaluations of the G5 emphasizing its 2x power savings in Gemini-related tasks compared to the G4, allowing for extended battery life during AI-heavy usage. Qualcomm's own reports on Snapdragon benchmarks indirectly affirm Tensor's niche, where it holds advantages in low-latency AI inference despite lower peak scores. Overall, these reviews position Tensor as a reliable enabler for Google's software-driven features rather than a benchmark leader.

Criticisms and Limitations

Early generations of the Google Tensor , particularly the G1 through models, have been criticized for significant throttling during intensive tasks like , where the chip's temperature rises rapidly, leading to performance degradation of up to 50% in sustained workloads on devices such as the 9 with the G4. This overheating has been identified as the primary reason for phone returns, negatively impacting in high-demand scenarios and prompting complaints about inadequate heat management compared to competitors like Qualcomm's Snapdragon chips. While the Tensor G5 in the 2025 10 series shows some mitigation, with throttling limited to under 20% in tests, critics note that the issue persists in prolonged sessions, throttling faster than expected for a 3nm-process chip. Efficiency shortcomings have also drawn scrutiny, with Tensor chips exhibiting higher power consumption relative to rivals; for instance, the G4 demonstrates elevated battery drain in CPU-intensive tests, contributing to shorter runtime in multimedia applications compared to the Snapdragon 8 Gen 3. The G5 introduces efficiency gains via its TSMC 3nm fabrication, promising better power management overall, yet 2025 reviews highlight ongoing modem inefficiencies from the Exynos integration, which continue to affect connectivity and battery life in real-world use despite prior updates. The Tensor's heavy reliance on Google's ML Kit for on-device has been noted as a limitation for third-party app developers, as the framework's custom model constraints and optimization focus on Google's ecosystem can hinder broader app performance and integration on devices. This software dependency ties into Google's seven-year update policy for and later models, which extends OS and security support but raises concerns about hardware longevity, as Tensor's thermal and efficiency issues may render devices less viable before the full support period elapses. These factors have collectively been perceived as constraining the Pixel lineup's competitiveness in high-end and , where Tensor's GPU limitations and immaturity result in underperformance against Snapdragon-powered flagships, limiting frame rates and graphical fidelity in demanding titles.

References

  1. [1]
    Google Tensor is a milestone for machine learning
    Oct 19, 2021 · Google Tensor is the first ever system on a chip made by Google, unlocking new experiences in Pixel 6 and Pixel 6 Pro.
  2. [2]
    Google Tensor debuts on the new Pixel 6 this fall
    Aug 2, 2021 · Google Tensor is our first custom-built SoC specifically for Pixel phones, and it will power the Pixel 6 and Pixel 6 Pro later this fall.
  3. [3]
    The latest and greatest: meet Pixel 7 and Pixel 7 Pro - The Keyword
    Oct 6, 2022 · The redesigned Pixel 7 and Pixel 7 Pro are here. Powered by our next-generation Google Tensor G2 processor and shipping with Android 13.
  4. [4]
    Google Tensor G3: The new chip that gives your Pixel an AI upgrade
    Oct 4, 2023 · It includes the latest generation of ARM CPUs, an upgraded GPU, new ISP and Imaging DSP and our next-gen TPU, which was custom-designed to run ...
  5. [5]
    14 new AI features in Google's Pixel 9 series devices - The Keyword
    Aug 13, 2024 · Google Tensor G4 is our fastest, most powerful chip yet. It's custom-built for advanced AI like Gemini, and helps Pixel run quickly and ...
  6. [6]
    5 reasons why Google Tensor G5 is a game-changer for Pixel
    Aug 20, 2025 · You can expect faster on-device AI, proactive features, and the best Pixel camera yet. Plus, your Pixel 10 battery will last all day.
  7. [7]
    Pixel by Google Smartphone with a Snapdragon 821 processor
    The Pixel by Google Smartphone is powered by a Snapdragon 821 processor. Explore features, read device specs and learn where to buy.
  8. [8]
    Google Pixel 5 - Full phone specifications - GSMArena.com
    Chipset, Qualcomm SM7250 Snapdragon 765G 5G (7 nm) ; CPU, Octa-core (1x2.4 GHz Kryo 475 Prime & 1x2.2 GHz Kryo 475 Gold & 6x1.8 GHz Kryo 475 Silver) ; GPU, Adreno ...Google Pixel 5 long-term review · Pictures · Compare · Review
  9. [9]
    Tensor is Google's most important launch of 2021, not the Pixel 6
    Nov 3, 2021 · Monika Gupta, senior director of the Google Silicon team, says Tensor ... Tensor development began in 2017, around the same time Google ...
  10. [10]
    Google's Tensor chip is only the first step toward truly great Pixel ...
    Aug 4, 2021 · As it stands now, though, the main difference between Tensor and a Snapdragon 888 looks like it'll just be some fancy neural network tricks. But ...
  11. [11]
    Nikkei: Google Tensor is a 5nm chip and its produced by Samsung
    Aug 4, 2021 · The new 5nm fabrication process is more efficient than the 7nm process and could improve battery life and performance. Google has also mentioned ...
  12. [12]
    Google Unveils The New Tensor SoC, The Heart Of The New Pixel 6 ...
    Oct 20, 2021 · The new Tensor SoC features two Arm Cortex X1 cores, clock speed unknown but likely lower than contemporaries, paired with two Arm Cortex A76 cores, and a quad ...
  13. [13]
    Google Tensor vs Snapdragon 888 series: How the Pixel 6 chip ...
    However, the first-generation Tensor is designed to compete with 2021's Qualcomm Snapdragon 888 series and Samsung Exynos 2100 flagship chipsets. So we'll use ...
  14. [14]
    Tensor G2 confirmed as 5nm processor, not 4nm as expected
    Oct 11, 2022 · According to new information, the Tensor G2 is, in fact, built upon a 5nm manufacturing process and not 4nm as was expected ahead of launch.
  15. [15]
    Google's Tensor G1, G2, and G3 chipsets explained: Core Pixel ...
    Oct 27, 2023 · The Tensor G3 chipset arrived alongside the Pixel 8 series in October 2023. It's a nona-core chipset, built on Samsung's 4nm manufacturing process.
  16. [16]
    Google Tensor G4 explained: Everything you need to know about ...
    Aug 22, 2024 · It's powered by Arm CPU and GPU cores, and is manufactured using one of Samsung's 4nm processes. It is rumored to be the same 4LPP+ process as ...Missing: G1 | Show results with:G1
  17. [17]
    Is the Google Pixel 8's Tensor G3 chip any good? We find out
    Oct 18, 2023 · ... issues that marred the performance of the Tensor G2 and other chips manufactured on Samsung's 5nm process last year. In 2022, Samsung ...
  18. [18]
    Tested: Are Google's Tensor chips really that hot and inefficient?
    Nov 17, 2024 · Google's Tensor has gone from Samsung's dubious 5nm LPE to a much improved Samsung 4nm (possibly 4LPP+) process with the Tensor G4, and the ...
  19. [19]
    Google Pixel 9 Pro XL's Tensor G4 Chip Shows Thermal Throttling ...
    Aug 21, 2024 · This isn't the first time Google's Tensor processors have faced criticism for temperature concerns. The Tensor series, which drives Google's ...
  20. [20]
    Google switches from Samsung to TSMC — Pixel 10 and Tensor G5 ...
    Aug 21, 2025 · It is the first Pixel processor manufactured by TSMC, moving away from Samsung's fabrication, which historically lagged in efficiency. This ...
  21. [21]
    First fully custom Pixel chip, Tensor G5, coming in 2025 from TSMC
    Jul 6, 2023 · According to a new report today, Google is indeed working to replace Samsung with TSMC for Tensor G5 development.
  22. [22]
    Google Tensor G5 Chip: TSMC's 3nm Power Brings Big Changes
    Aug 27, 2025 · Google claims that the fourth-generation TPU delivers up to 60% more AI processing power compared to the previous version, which perfectly ...
  23. [23]
    Google announces Tensor G5: What's new with the Pixel 10 ...
    Aug 20, 2025 · For starters, the Tensor G5 is made by TSMC on its 3nm process. Google didn't identify the specific process used, but this switch from Samsung ...Missing: 5nm | Show results with:5nm
  24. [24]
    Google shift to TSMC for Pixel 10 chip was a 'shock' to Samsung
    Jun 18, 2025 · According to a new report, Google's switch from Samsung's foundry to TSMC for the Tensor G5 chip in the upcoming Pixel 10 series was a “shock” for Samsung.Google Tensor G4 in Pixel 9... · It's not just Google, Samsung... · Per recent leaksMissing: manufacturing | Show results with:manufacturing
  25. [25]
    Pixel 10 powered by Tensor G5 on TSMC 3nm with 34% CPU gain ...
    Aug 21, 2025 · Pixel 10 powered by Tensor G5 on TSMC 3nm with 34% CPU gain and 60% faster AI ... Alphabet Inc. has introduced its latest Pixel 10 series ...
  26. [26]
    The “Google Silicon” team gives us a tour of the Pixel 6's Tensor SoC
    Oct 19, 2021 · This is Google's first main SoC in a smartphone, and the chip has a unique CPU core configuration and a strong focus on AI capabilities.
  27. [27]
    Google Tensor Processor - Benchmarks and Specs - Notebookcheck
    The Google Tensor SoC (G1, first generation) is the first smartphone processor from Google. ... Announcement Date, 11/01/2021 = 1467 days old. Product Link ( ...
  28. [28]
    DynamIQ: Revolutionizing Multicore Computing - Arm
    Enabling DynamIQ for Armv8.2 AE CPUs, which enables a mixture of big and LITTLE CPUs to be configured in a single cluster. When Hybrid mode is enabled, DSU-AE ...Missing: Google Tensor
  29. [29]
    Google Tensor G4: specs and benchmarks - NanoReview
    CPU ; Architecture, 1x 3.1 GHz – Cortex-X4 3x 2.6 GHz – Cortex-A720 4x 1.92 GHz – Cortex-A520 ; Cores, 8 ; Frequency, 3100 MHz ; Instruction set, ARMv9.2-A ; Process ...
  30. [30]
    Google Tensor SoC in Pixel 6 series features two Cortex-X1 cores ...
    Oct 19, 2021 · Google Tensor SoC in Pixel 6 series features two Cortex-X1 cores and 20-core GPU along with 4 MB L3 cache, built "not to win benchmarks but to ...<|control11|><|separator|>
  31. [31]
    Pixel phone hardware tech specs - Google Help
    Check out the tech specs for Pixel phones. If you're interested in buying a Pixel phone, visit the Google Store. Pixel.
  32. [32]
    Benchmarked: The Google Tensor G2 versus the competition
    Mar 21, 2023 · The original Tensor employed a Mali-G78 GPU with a 20-core design, while the Tensor G2 brings a Mali-G710 GPU with seven shader cores. Our own ...
  33. [33]
    Pixel 8's Tensor G3 gets leaked detailing all of its main specs
    Jun 5, 2023 · Google's Tensor G3 might come with a nona-core chipset supporting MTE · Ray-tracing and Immortalis graphics · Tensor G3 might be the first ...Ray-Tracing And Immortalis... · Other Tensor G3 Improvements · Ufs 4.0 Support<|separator|>
  34. [34]
    A New Foundation for AI on Android - Android Developers Blog
    Dec 6, 2023 · The new foundation for AI on Android uses Gemini Nano, optimized for mobile, and Android AICore, a system service for on-device access.
  35. [35]
    How Google Tensor Helps Google Pixel Phones Do More
    Google Tensor: the brains behind Pixel phones. Designed by Google, the chip helps Pixel take better pictures, supercharge your productivity, and so much more.
  36. [36]
    AI features accelerated by Google Tensor that make Pixel 6 and ...
    Oct 20, 2021 · Magic Eraser allows easy removal of "photobombs" or stray objects or people in photos after they have been taken. Users can mark which objects ...
  37. [37]
    Does tflite utilize unified memory architecture of SoC?
    May 13, 2022 · Generally, TFLite allots memory for tensors independently for every device (CPU, GPU, and NPU) that is utilized during inference.
  38. [38]
    How Google built the Pixel 10's Tensor G5 without Samsung's help
    Mar 18, 2025 · The upcoming Tensor G5 will be Google's first fully custom chip, and thanks to a leak from Google we now know how it achieved this feat.
  39. [39]
    The Google Pixel 9 series has UFS 3.1 storage, but does it matter?
    Aug 22, 2024 · The Google Pixel 9 and Pro models all come with UFS 3.1 storage, even if you pick one of the expensive memory options.<|separator|>
  40. [40]
    Google Pixel 10 series finally upgrades to faster UFS 4.0 storage ...
    Aug 20, 2025 · UFS 3.1 supports 2,100 MB/s read and 1,200 MB/s write, where UFS 4.0 jumps up to 4,200 MB/s read and 2,800 MB/s write. On top of that, UFS 4.0 ...
  41. [41]
    Tensor explained: Why Google built a custom chip for Pixel 6
    Oct 20, 2021 · The Tensor chip is specifically designed to offer Google's latest advances in AI directly on a mobile device.
  42. [42]
    Google's Pixel 7 Packs a Beefed-Up Security Chip - WIRED
    Oct 11, 2022 · Google spent three years subjecting Titan M2 to testing by the third-party lab SGS Brightsight. The chip now has an array of Common Criteria ...
  43. [43]
    Google Pixel 6 - Full phone specifications - GSMArena.com
    Google Pixel 6 · Released 2021, October 28 · 207g, 8.9mm thickness · Android 12, up to 15 · 128GB/256GB storage, no card slot.
  44. [44]
    Report: Google Tensor Chip Will Use 5nm Process, Samsung to ...
    Aug 4, 2021 · Report: Google Tensor Chip Will Use 5nm Process, Samsung to Manufacture. This chip will come in the new Pixel 6 and 6 Pro later this year, but ...
  45. [45]
    Google Tensor: specs and benchmarks - NanoReview
    Google Tensor – an 8-core chipset that was announced on October 19, 2021, and is manufactured using a 5-nanometer process technology.
  46. [46]
    Google Tensor G2 chip: Everything you need to know
    Jan 22, 2024 · Moreover, Google told Android Authority that its Tensor G2 chipset is still based on Samsung Foundries' 5nm process, which isn't as efficient ...Missing: G1 | Show results with:G1
  47. [47]
    Google Pixel 7 launches for $599 with Tensor G2, face unlock
    Oct 6, 2022 · The Google Pixel 7 just went official as Google's “affordable flagship” for 2022. What's new? Tensor G2, an updated camera setup, and more, but all for the ...
  48. [48]
    Google Tensor G2: specs and benchmarks - NanoReview
    Google Tensor G2 ; GPU: Mali-G710 MP7 ; Cores: 8 ; Clock: 2850 MHz.Tensor G4 vs Tensor G2 · Tensor G3 vs Tensor G2 · Snapdragon 888 vs Tensor G2Missing: ARM | Show results with:ARM
  49. [49]
    Google Tensor G2: All the details on the Pixel 7's custom chipset
    May 30, 2023 · The new GPU also helps the onboard machine-learning-focused Tensor Processing Unit (TPU), giving it an up to 35% boost in applicable processes.
  50. [50]
    What is the Tensor G2 SoC on the Google Pixel 7 series? How is it ...
    Oct 18, 2022 · In the case of Google Tensor G2, Google says that the TPU is both 60% more powerful and 20% more efficient. Those are some pretty big ...
  51. [51]
    Google Pixel 7a: Features, colors, pricing
    May 10, 2023 · It's built with Google Tensor G2, our flagship processor, and Titan M2, our dedicated security chip, making it faster, more efficient and more secure.Missing: specs | Show results with:specs
  52. [52]
    Google Tensor G3: All you need to know about Pixel 8 processor
    May 26, 2024 · The Pixel 8's Tensor G3 sports a single powerhouse Arm Cortex-X3 CPU core clocked at 2.91GHz, four Cortex-A715 cores at 2.37GHz, and four little Cortex-A510 ...
  53. [53]
    Google Tensor G3: specs and benchmarks - NanoReview
    Google Tensor G3 – an 9-core chipset that was announced on October 4, 2023, and is manufactured using a 4-nanometer process technology.Tensor G4 vs Tensor G3 · Tensor G3 vs Tensor G2 · Exynos 2400 vs Tensor G3 · VS
  54. [54]
    Google Tensor G3 Processor - Benchmarks and Specs
    Oct 26, 2023 · The CPU of the Tensor G3 is composed of three clusters. The first houses a single powerful Cortex-X3 core, which can operate at up to 2.91 GHz.
  55. [55]
    Google Tensor G4 Processor - Benchmarks and Specs
    The Tensor G4 is the processor (SoC) powering Pixel 9 series phones that debuted in August 2024. The CPU part of the SoC is made up of 8 cores.
  56. [56]
    Google's Tensor G4 vs. Qualcomm's Snapdragon 8 Gen 4
    Aug 15, 2024 · It has an ARM Mali-G715 GPU with a clock speed of 940 MHz, designed to improve AI capabilities and privacy features via a specialized Titan M2 ...<|separator|>
  57. [57]
    Satellite SOS on Pixel 9 – When it launches, how it works - 9to5Google
    Aug 16, 2024 · Satellite SOS is a new feature available to the PIxel 9, 9 Pro, 9 Pro XL, and 9 Pro Fold. It's thanks to the new Exynos 5400 modem on the Tensor G4 chip.
  58. [58]
    Pixel 9 Tech Specs, Features, Screen Size & Battery Life
    Google Tensor G4, Pixel's most advanced chip, and 12 GB RAM brings power and speed to the advanced AI features on Pixel 9.Missing: innovations | Show results with:innovations
  59. [59]
    Google's Tensor G4 SoC to have improved heat management and ...
    Mar 6, 2024 · The Tensor G4 will apparently have improved heat management, along with better performance and higher efficiency. So it looks like it will be a ...Missing: thermal | Show results with:thermal
  60. [60]
    Google Pixel 10 - Full phone specifications - GSMArena.com
    Google Pixel 10 5G Android smartphone. Announced Aug 2025. Features 6.3″ display, Google Tensor G5 chipset, 4970 mAh battery, 256 GB storage, 12 GB RAM, ...
  61. [61]
    Google Tensor G5: specs and benchmarks - NanoReview
    Google Tensor G5 – an 8-core chipset that was announced on August 20, 2025, and is manufactured using a 3-nanometer process technology.
  62. [62]
    Pixel 10's Tensor G5 deep dive: All the info Google didn't tell us
    Aug 27, 2025 · Before getting into the specifics, one of the biggest changes with the Tensor G5 is that Google has moved manufacturing from Samsung Foundry to ...
  63. [63]
    Google Tensor G5 vs Snapdragon 8 Gen 3: Benchmarks and Specs
    Aug 20, 2025 · Google Tensor G5 vs Snapdragon 8 Gen 3: Specs Comparison ; CPU Cores. 1x Cortex-X4 (3.78GHz), 5x Cortex-A725 (3.05GHz), 2x Cortex-A520 (2.25GHz).
  64. [64]
    Here's what Google's Pixel lineup has in store for 2025 and beyond
    Nov 30, 2024 · The Pixel Tablet 3 will be powered by the same Google Tensor G6 processor destined for the Pixel 11, but with a slight tweak. It'll be mounted ...
  65. [65]
    Google Pixel 11 rumors: Release date, features, and more
    Jul 30, 2025 · The Tensor G6 will adopt a single unannounced ARM Cortex-X930 core and six Cortex-A730 cores, which will allow the chip to focus on power ...<|control11|><|separator|>
  66. [66]
    Google Tensor G6 in Pixel 11 reportedly using TSMC 2nm process
    Jun 24, 2025 · According to a new report, Google is planning to jump to TSMC's 2nm process for Tensor G6 faster than it has in prior years.
  67. [67]
    Google reportedly testing MediaTek modem for Tensor G6, Pixel 11
    Oct 3, 2025 · The Tensor G6 for the Pixel 11 is rumored to switch to a MediaTek modem ... MediaTek announced the M90 5G Modem at MWC 2025 in February.Missing: Armv9 partnerships
  68. [68]
    Google Secures Tensor Chip Supply from TSMC Until Pixel 14
    May 26, 2025 · Google secures Tensor chip supply from TSMC until Pixel 14. It will ensure long-term Pixel performance and supply stability through the Pixel 14 generation.
  69. [69]
    Google Pixel 11's Tensor G6 might be a downgrade, but could also ...
    Nov 8, 2024 · Obviously, there are some upfront costs associated with making a custom chip, but with that big of a per-unit price difference, it suggests ...Missing: challenges | Show results with:challenges
  70. [70]
    This is why Snapdragon phones keep outperforming Google's Pixel
    but cost consciousness and an AI focus explain why Google's okay with ...
  71. [71]
    Question - Tensor 4 thermal throttling | XDA Forums
    Aug 18, 2024 · Tensor G4 running in Google's top-end Pixel 9 Pro XL throttles to 50 percent of its maximum performance, indicating poor thermal efficiency.
  72. [72]
    Google Tensor heating woes are the top reason for Pixel phone ...
    Nov 9, 2024 · The highest number of customers return their Pixel phones due to bad thermal output. The high temperatures are also said to ruin the experience of running ...
  73. [73]
    Google Pixel 10 series review: Don't call it an Android - Ars Technica
    Aug 29, 2025 · With the more powerful Tensor G5 to run larger Gemini Nano on-device models, Google has woven AI into even more parts of the OS. Google's ...
  74. [74]
    The Flaw In Google Pixel's New Tensor G5 Chip - Wccftech
    Oct 12, 2025 · In contrast, Google's Tensor G5 uses off-the-shelf Arm Cortex CPU cores that are not as fine-tuned or deeply optimized to work within the Pixel ...
  75. [75]
    Snapdragon 8 Gen 3 vs Tensor G4: tests and benchmarks
    We put Qualcomm Snapdragon 8 Gen 3 against Google Tensor G4 to find out which SoC is better. See performance comparison in benchmarks and games.
  76. [76]
    Tensor G5 vs Tensor G4: Benchmarks and specs - Gizmochina
    Cortex-X4 5x 3.05 GHz — Cortex-A725 2x 2.25 GHz — Cortex-A520, 1x 3.1 GHz — Cortex-X4 3x 2.6 GHz — Cortex-A720 4x 1.92 GHz — ...
  77. [77]
    Google ML Kit: Everything you need to know - Android Police
    Apr 26, 2023 · And with the Google Tensor chip powering these devices now, they continue to surpass many hardware limitations, as they emphasized all those ...
  78. [78]
    7 years of software updates for the Pixel 8 series
    Oct 4, 2023 · Google will also extend hardware support and make parts available for seven years to match the software commitment. The result? You can be ...
  79. [79]
    Google's extended Pixel updates prove just how important Tensor is
    Dec 8, 2024 · The Pixel 8 series and onwards already offer extensive seven-year update policies that'll probably outlast their hardware. Unfortunately, we ...
  80. [80]
    The Google Pixel 10's gaming reputation isn't fair - Android Police
    Oct 27, 2025 · It's a common refrain among reviewers that the Google Pixel 10 is limited in GPU performance and gaming.
  81. [81]
    Google Pixel 10 Gaming Woes: Outdated Drivers Hit Tensor G5 ...
    Oct 26, 2025 · Google's Pixel 10 series, featuring the Tensor G5 chip with PowerVR GPU, has underperformed in gaming and graphics tasks compared to rivals due ...