内容来源：https://aibusiness.com/data/google-cloud-nvidia-virtual-machines

内容总结：

谷歌云携手英伟达近日发布新一代G4虚拟机，搭载全新RTX Pro 6000 Blackwell架构GPU，为企业级人工智能应用注入强劲动力。该系列产品在运算性能上实现突破性提升，较前代产品吞吐量提升高达九倍，可支持参数规模从300亿至1000亿的大型语言模型微调与推理。

此次合作标志着双方战略伙伴关系的深化。G4虚拟机采用创新架构设计，不仅通过第五代张量核心提升AI运算效率，更凭借第四代RT核心实现影院级图形渲染。其独特的分区GPU技术允许单卡拆分为四个独立计算单元，支持多任务并行处理，有效优化资源配置。

值得关注的是，谷歌云平台同步上线英伟达Omniverse虚拟仿真平台与Isaac Sim机器人模拟架构。这些工具与G4虚拟机结合，可为工业数字孪生、机器人仿真等场景提供完整的云端解决方案。企业用户现可通过谷歌云市场直接部署这些服务，构建高精度虚拟仿真环境。

业界专家指出，此次技术升级将显著加速生成式AI在多媒体推理、可视化设计等领域的落地进程，为制造业、研发机构等用户群体带来更高效的数字化工具支持。

中文翻译：

由谷歌云为您呈现
如何选择首个生成式AI应用场景
开启生成式AI之旅时，建议优先关注能够提升人类信息交互体验的领域。

谷歌云正式推出搭载NVIDIA RTX Pro 6000 Blackwell GPU的G4虚拟机，专注于人工智能、机器人技术及企业级应用。此次通过推出面向企业高端AI应用的新一代虚拟机，谷歌云进一步深化了与NVIDIA的战略合作。G4虚拟机采用NVIDIA RTX Pro 6000 Blackwell服务器版GPU，据谷歌表示这将实现重大技术突破。

同期，谷歌云应用市场已正式上架NVIDIA Omniverse虚拟机镜像及机器人仿真架构参考方案NVIDIA Isaac Sim。谷歌云技术专家Roy Kim与Dai Vu在博文中阐释：“G4虚拟机带来性能跃升，其吞吐量可达G2实例的九倍，这将为多模态AI推理、超逼真设计可视化、基于NVIDIA Omniverse的机器人仿真等多样化工作负载带来革命性进展。”

首批G4虚拟机支持配置1/2/4/8块RTX Pro Blackwell GPU选项，总计搭载768GB GDDR7显存。后续将很快推出分体式GPU配置方案。NVIDIA表示，该架构融合第五代张量核心（实现AI性能跨越）与第四代RT核心（支持影院级画质与照片级仿真）。

谷歌云指出，客户将在多个领域受益：在生成式AI方面，G4的计算能力可加速大语言模型微调与推理，支撑多模态及文生图等实时应用；在能效层面，单块GPU可划分为四个独立实例，各配备专属高速显存与计算核心，实现单虚拟机并发处理多类轻量任务；此外还具备卓越弹性，可运行参数规模从不足300亿至超1000亿的AI模型。

针对NVIDIA Omniverse的上线价值，Kim与Vu补充道：“G4虚拟机为企业数字孪生所需的实时渲染与物理精准仿真提供核心基础设施，结合Omniverse共同构建可扩展的云环境，助力工业数字孪生与机器人仿真应用的开发部署及交互。”

对NVIDIA而言，此次合作拓展构建了基于Blackwell的端到端平台，其统一架构有望在单一云生态内为工作负载加速提供无缝体验。

拓展阅读：

英文来源：

Sponsored by Google Cloud
Choosing Your First Generative AI Use Cases
To get started with generative AI, first focus on areas that can improve human experiences with information.
Google Cloud launches G4 VMs with Nvidia RTX Pro 6000 Blackwell GPUs aimed at AI, robotics and enterprise applications.
Google Cloud has further expanded its partnership with Nvidia via a new lineup of virtual machines aimed at advanced AI applications in the enterprise.
The G4 VMs are powered by Nvidia's RTX Pro 6000 Blackwell Server Edition GPUs, and, according to Google, constitute a significant advance.
At the same time, Google has made Nvidia Omniverse generally available as a virtual machine image on Google Cloud Marketplace as well as Nvidia Isaac Sim, a reference robotic simulation architecture.
In a blog post, Google Cloud's Roy Kim and Dai Vu explained: "The G4 VM offers a profound leap in performance, with up to nine times the throughput of G2 instances, enabling a step-change in results for a wide spectrum of workloads, from multimodal AI inference, photorealistic design and visualization, and robotics simulation using applications developed on Nvidia Omniverse."
Initially, G4 VMs can be configured with one, two, four and eight RTX Pro Blackwell GPU options -- totaling 768 GB of GDDR7 memory. Fractional GPU options will be coming soon.
Its design combines fifth-generation tensor cores that deliver a leap in AI performance, and fourth-generation RT cores that enable cinematic-quality graphics and photorealistic simulations, according to Nvidia.
Google Cloud stated that customers will benefit in several areas, including generative AI, where G4's compute accelerates large language model fine-tuning and inference, enabling real-time applications such as multimodal and text-to-image models.
There are potential efficiency gains, too. Because the G4 allows a single GPU to be split into four isolated instances, each with its own high-bandwidth memory and compute core, multiple smaller distinct workloads can be run concurrently off a single VM.
Plus, it brings flexibility and scaling, with the ability to run AI models from less than 30 billion to more than 100 billion parameters.
Google Cloud's Kim and Vu also described how the availability of Nvidia Omniverse on Google Cloud Marketplace will benefit customers. "G4 VMs provide the necessary infrastructure to run the demanding real-time rendering and physically accurate simulations required for enterprise digital twins. Together [with Omniverse], they provide a scalable cloud environment to build, deploy, and interact with applications for industrial digital twins or robotics simulation."
For Nvidia, the expanded partnership establishes an end-to-end platform on Blackwell, with the unified architecture providing what is claimed to be a seamless experience for accelerating workloads within a single cloud ecosystem.
You May Also Like