CybserSecurity News

Tag: Tensor G4

  • The yield rate of Exynos 2400 is currently at 60%

    In October of the preceding year, Samsung heralded the advent of its latest mobile processor innovation, the Exynos 2400, marking a significant leap forward from its Exynos 2200 predecessor. This new chipset boasts a substantial 70% enhancement in CPU performance and a staggering 14.7-fold acceleration in artificial intelligence (AI) workloads. Additionally, it introduces the RDNA 3 architecture to its GPU, significantly elevating gaming experiences and ray tracing capabilities.

    Manufactured using Samsung’s advanced 4LPP+ process, recent revelations by enthusiasts suggest a yield rate of 60%, slightly trailing behind TSMC’s N4P process at 70%. Yet, considering Samsung’s yield was a mere 25% approximately 12 to 18 months prior, this represents a monumental stride in progress.

    Despite the 4LPP+ process not equating to the N4P’s efficiency, and the Exynos 2400 not outperforming Snapdragon 8 Gen 3 by a wide margin, the disparity has considerably narrowed. Samsung is poised to amplify production, maintaining a 60% yield rate. Rumors hint at Google’s forthcoming Tensor G4 potentially employing the 4LPP+ manufacturing process later this year, bearing similarities to the Exynos 2400, raising anticipation for further improvements in yield rates.

    The Exynos 2400’s CPU configuration features a sophisticated quad-cluster architecture, comprising a super-core (Cortex-X4@3.20GHz), two high-frequency large cores (Cortex-A720@2.90GHz), three medium-frequency large cores (Cortex-A720@2.60GHz), and four efficiency cores (Cortex-A520@2.00GHz), totaling ten cores. It also incorporates a “17K MAC” NPU and an integrated 5G modem capable of delivering downlink speeds of up to 12.1Gbps and uplink speeds of up to 3.67Gbps, supporting Sub-6GHz with downlink/uplink speeds of 9.64/2.55Gbps.

  • Google Tensor G5 may use TSMC 3nm process node

    This year, Google unveiled the Tensor G3, its inaugural smartphone SoC supporting the AV1 codec, which is employed in its flagship products, the Pixel 8 and Pixel 8 Pro. However, in comparison to Qualcomm’s third-generation Snapdragon 8 and MediaTek’s Dimensity 9300, the Tensor G3 significantly lags, with a widening performance gap that has been a source of disappointment and has compelled Google to seek new directions.

    According to media reports, Google is contemplating adopting TSMC’s 3nm process node, such as the N3E technology, for its Tensor G5. This move would at least bring Google’s SoC manufacturing process in line with Qualcomm and MediaTek. Beyond the manufacturing technique, Google plans to abandon Samsung’s SoC design, focusing instead on a fully customized solution, and may even introduce an internally developed new GPU to enhance the overall performance of the SoC.

    Moreover, Google’s focus extends beyond CPU and GPU performance. How to improve energy efficiency and AI capabilities are also pivotal considerations, as it aims to develop a chip that synergizes more effectively with its software requirements. However, these advancements are not expected until 2025 at the earliest, meaning that Google’s first fully customized SoC might not debut until the Pixel 10 and Pixel 10 Pro.

    As for the upcoming “Zuma Pro” Tensor G4, Google will continue its collaboration with Samsung. Rumored to be a minor update, it will be featured in the Pixel 9 and Pixel 9 Pro. Reports have indicated that the Tensor G4 prototype has been operational on a development platform named “Ripcurrent Pro,” with Samsung’s development department assisting. It might also incorporate elements of the Exynos 2400 design, with both being manufactured using the 4LPP+ process.

    Google has no plans to switch its foundry for the Tensor G4, not only due to timing but also due to cost considerations, as the expense of transitioning to TSMC’s 3/4nm process at this stage is prohibitively high.