Skip to content
港通金屬樞紐Gangtong Metal Hub

产品与服务Products & Services

GPU 云算力产品组合GPU cloud compute portfolio

从弹性云实例到专属训练集群，再到推理端点与私有部署 — 随开随用、按量计费。From elastic instances to dedicated clusters, inference endpoints, and private deployments — provision on demand, pay as you go.

实例规格一览（按量起步价）Instance SKUs (on-demand starter rates)

单卡基础价 · 询价请联系 support@aivon.funSingle-GPU base rates · Inquire: support@aivon.fun

RTX 3090

24 GB · 入门推理 / 实验Entry infer / labs

¥1.19

元/小时 起¥/hr 起

RTX 4090 24G

24 GB · 推理 / 微调Infer / fine-tune

¥2.15

元/小时 起¥/hr 起

RTX 4090 48G

48 GB · 大显存微调High-VRAM fine-tune

¥3.30

元/小时 起¥/hr 起

RTX 5090 32G

32 GB · 多模态 / AgentMultimodal / Agent

¥3.32

元/小时 起¥/hr 起

A800 80G

80 GB · 训练 / 多卡Train / multi-GPU

¥6.23

元/小时 起¥/hr 起

H20 96G

96 GB · 高性能训练 / 推理High-end train / infer

¥7.50

元/小时 起¥/hr 起

A100 80G

80 GB · 大模型训练LLM training

¥10.75

元/小时 起¥/hr 起

01

GPU 云实例GPU Cloud Instances

弹性、可计费的高性能 GPU 实例，覆盖实验、微调与生产推理前置负载。Elastic, metered high-performance GPU instances for experiments, fine-tuning, and pre-production inference.

  • 多规格 GPU 实例族（面向训练 / 推理的不同显存与算力档位）Multiple GPU instance families (tiers for training vs. inference memory/compute)
  • 按需开通与释放，支持项目制与月度容量规划On-demand provision/release; project-based and monthly capacity planning
  • 预装或可定制 CUDA / 主流深度学习框架环境Prebuilt or customizable CUDA / deep-learning framework environments
  • 与对象存储、镜像仓库及 CI 流水线对接能力Integration paths for object storage, image registries, and CI pipelines
02

专属 AI 训练集群Dedicated AI Training Clusters

为多节点分布式训练提供可预期的网络、存储与调度能力。Predictable networking, storage, and scheduling for multi-node distributed training.

  • 高带宽 / 低延迟节点互联方案（按项目选型）High-bandwidth / low-latency interconnect options (project-selected)
  • 作业队列与优先级策略，减少共享干扰Job queues and priority policies to reduce noisy-neighbor effects
  • 检查点存储与训练日志归档建议实践Checkpoint storage and training-log archival best practices
  • 专属客户成功与容量扩容路径Dedicated customer success and capacity expansion paths
03

推理服务端点Inference Endpoints

面向在线与批处理推理的端点部署，强调稳定性与可观测性。Online and batch inference endpoints focused on stability and observability.

  • API 接入与鉴权集成API access and authentication integration
  • 水平扩缩容与流量峰值预案Horizontal scaling and peak-traffic playbooks
  • 延迟 / 吞吐 / 错误率等基础可观测指标Baseline observability: latency, throughput, error rate
  • 版本发布与回滚流程建议Release and rollback process guidance
04

私有云 / VPCPrivate Cloud / VPC

租户级网络隔离与安全边界，满足数据驻留与企业治理要求。Tenant-level network isolation and security boundaries for residency and governance.

  • VPC / 私有网络与安全组策略VPC / private networking and security-group policies
  • 访问控制（IAM 风格角色与最小权限）Access control (IAM-style roles and least privilege)
  • 审计日志与操作留痕能力建设Audit logging and operational trail capabilities
  • 可选专线 / VPN 对接企业内网Optional dedicated line / VPN to enterprise networks

配套能力Add-ons

镜像 · 存储 · 网络 · SLAImages · Storage · Network · SLA

镜像模板Image Templates

预装 CUDA、PyTorch、TensorFlow 与常用工具链，一键拉起。Prebuilt CUDA, PyTorch, TensorFlow, and toolchains — one-click launch.

高速存储Fast Storage

本地 SSD 与对象存储路径，支撑检查点与大数据集吞吐。Local SSD and object-storage paths for checkpoints and large datasets.

私有网络Private Networking

租户隔离、安全组与可选 VPN / 专线对接。Tenant isolation, security groups, optional VPN / dedicated links.

SLA 框架SLA Framework

可用性目标、事件响应与变更维护可写入合同（示意 99.9%）。Availability targets, incident response, and change windows in contract (indicative 99.9%).

咨询开通 / 报价Inquire / Get a quote

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.

开通试用Request Trial联系我们Contactsupport@aivon.fun