RTX 4090
高性价比推理 / 微调Value inference / fine-tune
VRAM 24 GB
乌兰察布、上海等可用Available in Ulanqab, Shanghai & more
¥2.15¥2.15元/小时 起¥/hr 起
CUDAPyTorch推理
咨询开通InquireGPU 云算力 · AI 训练与推理GPU Cloud · AI Train & Infer
一站式 AI 云服务One-stop AI Cloud
随开随用、精确计费Instant provision · Precise billing
港通金屬樞紐 为企业与开发者提供弹性 GPU 实例、训练集群与低延迟推理能力。从实验到生产，按量取用，专注业务而非机房。Gangtong Metal Hub delivers elastic GPU instances, training clusters, and low-latency inference — from experiment to production, pay for what you use.
海量 GPUMassive GPU Fleet
消费级到数据中心级多规格随开随用Consumer to datacenter GPUs, provision on demand
秒级计费Per-Second Billing
用多少付多少，精确到秒Pay only for what you use, billed by the second
低延迟推理Low-Latency Inference
弹性扩缩，稳定支撑在线服务Elastic scale for stable online serving
训练集群Training Clusters
多卡多机，高带宽互联可选Multi-GPU / multi-node with high-bandwidth options
热门实例Popular Instances
按需选择 GPU 规格Pick the GPU that fits
按量计费起步价 · 闲置 / 无卡模式 ¥0.15/小时On-demand starter rates · Idle / no-GPU mode ¥0.15/hr
RTX 5090
最热门 · 新上卡型Hottest · New SKU
VRAM 32 GB
¥3.32¥3.32元/小时 起¥/hr 起
训练多模态Agent
咨询开通InquireA800
数据中心训练级Datacenter training class
VRAM 80 GB
¥6.23¥6.23元/小时 起¥/hr 起
大模型多机NVLink
咨询开通InquireH20
高性能训练 / 推理High-end train / infer
VRAM 96 GB
¥7.50¥7.50元/小时 起¥/hr 起
SFT推理大显存
咨询开通Inquire实例闲置 / 无卡模式仅 ¥0.15/小时，保留环境与数据，有卡时按 GPU 规格计费。Idle / no-GPU mode from ¥0.15/hr — keep env & data; billed at GPU rates when attached.
平台能力Platform
弹性 · 按量 · 开箱即用Elastic · Metered · Ready to run
镜像模板、高速存储与私有网络，覆盖从实验到生产的完整链路。Image templates, fast storage, and private networking — experiment to production.
弹性伸缩Elastic Scale
按负载自动扩缩，忙时加卡、闲时释放，降低空转成本。Scale with load — add GPUs when busy, release when idle.
按量计费Pay-as-you-go
秒级计费与预付容量可选，适配实验、生产与项目制采购。Per-second or reserved capacity for labs, production, and projects.
镜像模板Image Templates
预装 CUDA、主流框架与常用工具链，一键拉起开发环境。Prebuilt CUDA, frameworks, and toolchains — launch in one click.
存储与网络Storage & Network
高速盘、对象存储与私有网络，支撑检查点与数据吞吐。Fast disks, object storage, and private networking for checkpoints.
核心产品Core Products
GPU 实例 + 模型推理GPU instances + model inference
GPU 云实例GPU Cloud Instances
从 4090 到 A100/H100 级规格，适合实验、微调、推理与中小规模训练。随开随用，按量计费。From 4090 to A100/H100-class SKUs for experiments, fine-tuning, inference, and mid-scale training.
查看实例规格 →View instance SKUs →模型 API / 推理端点Model API / Inference
低延迟推理部署与 API 接入，支持扩缩容、鉴权与可观测指标。Low-latency inference with API access, scaling, auth, and observability.
了解推理服务 →Explore inference →快速上手Get Started
三步开始使用Three steps to get going
- 01
告诉我们需求Tell us your needs
GPU 规格、时长、训练或推理场景 — 邮件即可。GPU SKU, duration, training or inference — just email us.
- 02
开通与配置Provision & configure
匹配实例、镜像与网络，协助环境就绪。Match instances, images, and networking to get you running.
- 03
开始训练 / 推理Train or serve
按量使用，随时扩容或释放资源。Use on demand — scale up or release anytime.
可信交付Trust
为生产工作负载而设计Built for production workloads
以下为示意性能力区间，正式对外披露数据以合同约定为准Indicative capacity ranges; formal figures subject to contract
99.9%
目标可用性（示意）Target availability (indicative)
秒级
计费粒度Billing granularity
多规格
GPU 实例族GPU instance families
< 1 日
咨询响应Inquiry response
AWSMicrosoft AzureNVIDIA GPU基础设施与技术生态对齐Infrastructure & ecosystem alignment
准备好开通 GPU 算力了吗？Ready to spin up GPU compute?
试用咨询、实例报价或专属训练集群 — 邮件即可，我们通常在一个工作日内回复。Trials, instance quotes, or dedicated training clusters — just email us; we typically reply within one business day.