Google Cloud 透過 C4A 和 A4X 機器系列，在 Compute Engine 中提供一系列 Arm 技術的伺服器。ARM 架構經過最佳化處理，運作效率相當卓越，因此成本效益更高。
相較於 x86 伺服器，Arm 處理器在標準伺服器中相當常見，因為它們更節能。行動電話和筆記型電腦就是在 Arm 處理器上執行的裝置範例。使用 Arm CPU 的縮減指令集，指令越少，效能速度越快，電池和電力消耗越低。
舉例來說，C4A 會使用 Google 自訂的 Arm 處理器 Axion，該處理器以 Arm Neoverse V2 處理器為基礎。Neoverse V2 是首款搭載 Armv9 效能、電源和安全性強化功能的 V 系列 CPU。此 VM 專為高效能運算、機器學習和通用雲端運算而設計。建議您將 C4A 一般用途 Arm 虛擬機器 (VM) 用於下列任何用途：
執行需要視需要快速擴充用量的耗用大量運算資源的工作負載。
針對與 Arm 相容的工作負載，提供最佳成本效益。
以現代化開放原始碼軟體堆疊進行建構。
開發及測試使用 Arm CPU 的行動或嵌入式系統。
評估工作負載是否適合 Arm CPU。
如要使用搭載 Arm 架構 CPU 的 GPU，請選擇在 NVIDIA GB200 NVL72 平台上執行的 A4X 機器系列。使用此機器系列建立的 VM 已連結 NVIDIA GB200 Grace Blackwell 超級晶片。這個機器系列經過最佳化處理，適用於大規模平行的 Compute Unified Device Architecture (CUDA) 運算工作負載，例如機器學習 (ML) 和高效能運算 (HPC)。
對於 C4A 機器系列，每個 vCPU 都由單一核心支援，且不支援多執行緒並行 (SMT)。因此，與啟用 SMT 的 VM 相比，C4A VM 可提供更高的 vCPU 效能。雖然 SMT 可為特定工作負載帶來好處，但單執行緒核心最適合運算密集型工作負載，因為程序可以存取整個核心，而非與其他程序共用。
作業系統映像檔
C4A VM 支援公開的 Arm 架構OS 映像檔。您也可以使用公開提供的 Arm 映像檔建立自訂映像檔。
Tau T2A 機器系列
Tau T2A Arm 機器系列採用 64 核心 Ampere Altra Arm 處理器，全核心頻率為 3.0 GHz。透過 Tau T2A，您可以執行在 ARM 上執行最佳效能的工作負載，甚至是專屬工作負載。
Tau T2A 機器系列已預先定義機器類型，最多可搭載 48 個實體核心，每個 vCPU 有 4 GB 記憶體。Tau T2A 機型會在單一 NUMA 節點中執行。
Tau T2A 機型僅支援 NVMe 儲存裝置介面，以及 Google 虛擬 NIC (gVNIC) 網路。不支援 Virtio-Net 和 SCSI 介面。所有公開提供的 Arm OS 映像檔都已設定為使用 NVMe 和 gVNIC 介面。gVNIC 是專為 Compute Engine 設計的網路介面。這項功能可提供更佳效能，並支援更高的網路頻寬和處理量。
[[["容易理解","easyToUnderstand","thumb-up"],["確實解決了我的問題","solvedMyProblem","thumb-up"],["其他","otherUp","thumb-up"]],[["難以理解","hardToUnderstand","thumb-down"],["資訊或程式碼範例有誤","incorrectInformationOrSampleCode","thumb-down"],["缺少我需要的資訊/範例","missingTheInformationSamplesINeed","thumb-down"],["翻譯問題","translationIssue","thumb-down"],["其他","otherDown","thumb-down"]],["上次更新時間：2025-09-04 (世界標準時間)。"],[[["\u003cp\u003eGoogle Cloud offers Arm-powered servers in Compute Engine, including the C4A machine series built on Google's custom Axion Arm64-based CPU, which are optimized for power efficiency and deliver better price for performance.\u003c/p\u003e\n"],["\u003cp\u003eThe C4A machine series provides up to 72 vCPUs, 576 GB of DDR5-5600 memory, supports up to 100 Gbps per VM networking performance through Titanium, and is ideal for compute-intensive workloads and Arm-compatible applications.\u003c/p\u003e\n"],["\u003cp\u003eThe Tau T2A machine series, running on the Ampere Altra Arm processor, offers up to 48 physical cores, 4GB of memory per vCPU, and supports only NVMe and gVNIC interfaces, making it suitable for workloads that specifically benefit from the Arm architecture.\u003c/p\u003e\n"],["\u003cp\u003eC4A and Tau T2A machine series both use a single core for each vCPU without simultaneous multithreading (SMT), which provides higher performance per vCPU and is ideal for compute-intensive tasks.\u003c/p\u003e\n"],["\u003cp\u003eArm servers are power efficient and are a great solution for a vast range of workloads, such as ML, image encoding, web and app serving, and more, providing a great alternative to x86 servers.\u003c/p\u003e\n"]]],[],null,["*** ** * ** ***\n\nGoogle Cloud offers a range of Arm powered servers in Compute Engine through\nthe C4A and A4X machine series. Arm architecture is optimized for power efficiency, and as a result can yield better price for performance.\n\nArm processors are common in standard servers due to their power efficiency as\ncompared to x86 servers. Mobile phones and laptops are examples of devices that\nrun on an Arm processor. With an Arm CPU's reduced instruction set, fewer\ninstructions equals greater performance speed with lower battery and power\nconsumption.\n\nFor example, C4A uses Google's custom Arm processor, Axion, which is based on\nthe Arm Neoverse V2 processor. The Neoverse V2 is the first V-series CPU to\nhave Armv9 performance, power, and security enhancements. It is designed for\nhigh performance computing, machine learning, and general-purpose cloud\ncomputing. Consider using C4A general-purpose Arm virtual machines (VMs) for\nany of the following purposes:\n\n- Run compute-intensive workloads that require the ability to scale usage quickly when needed.\n- Optimize for price-performance on Arm-compatible workloads.\n- Build on modern, open source software stacks.\n- Develop and test mobile or embedded systems which use an Arm CPU.\n- Evaluate whether your workload is suitable for an Arm CPU.\n\nTo use GPUs with an Arm-based CPU, choose the A4X machine series, which runs on\nthe NVIDIA GB200 NVL72 platform. VMs created using this machine series have\nattached NVIDIA GB200 Grace Blackwell Superchips. This machine series is\noptimized for massively parallelized Compute Unified Device Architecture (CUDA)\ncompute workloads, such as machine learning (ML) and high performance computing\n(HPC).\n\nA4X machine series\n\n[A4X](/compute/docs/accelerator-optimized-machines#a4x_series) is the first Compute Engine VM with both Arm-based CPUs and attached GPUs. A4X\noffers machine types that have up to 140 vCPUs,\nand 884 GB of memory. A4X uses NVIDIA GB200 GPUs, which offer\n180 GB memory per GPU. A4X has two sockets with NVIDIA Grace Arm CPUs\nconnected to four B200 GPUs with fast chip-to-chip (NVLink C2C) communication.\nA4X is available in the `a4x-highgpu-4g` machine type.\n\nStorage options for A4X instances\n\nA4X can be used with Google Cloud Hyperdisk attached storage and comes with\n12,000 GiB of [Local SSD](/compute/docs/disks/local-ssd). Compute Engine\nautomatically attaches the Local SSD disks to your A4X instances during instance\ncreation.\n\nOS images\n\nA4X instances support public Arm-based [OS images](/compute/docs/images/os-details).\nYou can also [create custom images](/compute/docs/images/create-custom)\nusing a public Arm-based OS image.\n\nC4A machine series\n\n[C4A](/compute/docs/general-purpose-machines#c4a_series) is the first\nArm-based VM built on Google's Axion\nArm64-based CPU. C4A offers machine types with up to 72 vCPUs and 576 GB of\nDDR5-5600 memory. C4A is available in `standard`, `highmem`, and `highcpu`\nmachine types.\n\nC4A is built on [Titanium](/titanium) which uses network offloads and\nenables per VM Tier_1 networking performance of up to 100 Gbps with the\n[gVNIC networking interface](/compute/docs/networking/using-gvnic).\nC4A also supports the NVMe disk interface with Hyperdisk Balanced and Hyperdisk Extreme disks.\n\nSimultaneous multithreading\n\nFor the C4A machine series, each vCPU is backed by a single core with no\n[simultaneous multithreading (SMT)](/compute/docs/instances/set-threads-per-core).\nThus, C4A VMs deliver greater performance per vCPU compared to a VM with SMT\nenabled. While SMT provides benefits to certain workloads, single-threaded cores\nare ideal for compute-intensive workloads because the processes can access the\nentire core instead of sharing it with other processes.\n\nOS images\n\nC4A VMs support public Arm-based [OS images](/compute/docs/images/os-details).\nYou can also [create custom images](/compute/docs/images/create-custom)\nusing a publicly-available Arm-based image.\n\nTau T2A machine series\n\nThe [Tau T2A Arm](/compute/docs/general-purpose-machines#t2a_machines)\nmachine series runs on the 64 core Ampere Altra Arm processor at 3.0 GHz\nall-core frequency. Tau T2A makes it possible to run workloads that run best,\nor exclusively, on Arm.\n\nThe [Tau T2A machine series](/compute/docs/general-purpose-machines#t2a_machines)\nhas predefined machine types of up to 48 physical cores with 4 GB of memory per\nvCPU. Tau T2A machine types run within a single\n[NUMA](https://www.kernel.org/doc/html/v4.18/vm/numa.html)\nnode.\n\nTau T2A machine types support only the NVMe interface for storage, and\n[Google virtual NIC (gVNIC)](/compute/docs/networking/using-gvnic) for\nnetworking. Virtio-Net and SCSI interfaces are not supported. All\npublicly-available Arm [OS images](/compute/docs/images/os-details)\nare configured to use the NVMe and gVNIC interfaces.\ngVNIC is a network interface that is designed specifically for\nCompute Engine. It provides better performance and supports higher\nnetwork bandwidths and throughput.\n\nFor this machine series, each vCPU is backed by a single core with no\nsimultaneous multithreading (SMT).\n\nWorkload recommendations\n\nThe C4A machine series is an excellent choice for a wide range of scale-out and\ncompute-intensive workloads, especially when price performance is a key concern.\nConsider C4A when you are deploying workloads such as the following:\n\n- ML data processing\n- ML inferencing and model serving\n- App serving, web serving, and game serving\n- Embedded systems development\n- Development on CI/CD on Arm\n- Video and image encoding, transcoding, and processing\n- Digital advertising exchanges and serving\n- Cache servers\n- Computational drug discovery\n- Android development\n- Autonomous or conventional automotive software development\n\nWhat's next\n\n- Review the specifications and features of the [A4X machine series](/compute/docs/accelerator-optimized-machines#a4x_series).\n- Review the specifications for the [C4A machine series](/compute/docs/general-purpose-machines#c4a_series).\n- Learn about available [CPU platforms](/compute/docs/cpu-platforms) for Google Cloud.\n- [Create and start a Compute Engine instance](/compute/docs/instances/create-vm-from-public-image) using an Arm OS image."]]
