Skip to content

Private deployments only · info@hanqing.com仅提供私有化部署 · info@hanqing.com

Five GPU series, ready to deploy五类 GPU 现已可部署

Private cloud computefor AI video generation.We build it. You run it.AI 视频生成的私有云算力。我们搭建,你来掌控。

Hanqing builds and runs private GPU clusters for AI video workloads — B300, B200, H200, RTX PRO 6000 and H100 — in your data centre or a hosted one. Root stays with you.瀚擎为 AI 视频负载搭建并运营私有 GPU 集群 —— B300、B200、H200、RTX PRO 6000、H100 —— 部署在你的机房或托管机房,root 权限始终在你手里。

GPU seriesGPU 系列

5

Per-GPU memory, up to单卡显存,最高

288 GB

Memory bandwidth, up to显存带宽,最高

8 TB/s

01What you get你得到什么

AI video, on GPUs that are yoursAI 视频生成,跑在属于你的 GPU 上

A private GPU cluster built for video workloads. We build and run it. You own it.一套为视频负载而建的私有 GPU 集群。我们搭建和运维,你拥有它。

  • 01

    Dedicated GPUs专属 GPU

    B300, B200, H200, RTX PRO 6000 and H100, sized to your workload and shared with no one.B300、B200、H200、RTX PRO 6000、H100,按你的负载选型,不与任何人共享。

  • 02

    Built for video为视频而建

    Text-to-video, image-to-video, batch rendering and fine-tuning, tuned on your real jobs.文生视频、图生视频、批量出片与微调,按你的真实任务调优。

  • 03

    Your data stays home数据不出域

    Weights, prompts and renders never leave your hardware. Air-gapped works too.权重、提示词与成片都留在你的硬件上,也支持完全离网。

  • 04

    Run by us, owned by you我们运维,你掌控

    Kubernetes scheduling, round-the-clock monitoring and a spare-parts SLA. Root stays with you.Kubernetes 调度、全天候监控与备件 SLA。root 权限始终在你手里。

02Showcase案例

Rendered on Hanqing在瀚擎算力上生成

Two films and five frames, all rendered on Hanqing compute.两支成片、五帧画面,全部在瀚擎算力上生成。

Featured film精选成片

Orbit crew → brand planet太空舱 → 品牌星球

1080P · 21s · with sound1080P · 21 秒 · 含声音

Featured film精选成片

Wuxia short武侠短片

1080P · 30s · 12 shots1080P · 30 秒 · 12 镜

01

Cyberpunk neon rain赛博霓虹雨夜

Vertical 1080P · 10s竖屏 1080P · 10 秒

02

Bamboo grove at dawn竹林晨雾

Vertical 1080P · 10s竖屏 1080P · 10 秒

03

Brand product close-up品牌产品特写

Vertical 1080P · 6s竖屏 1080P · 6 秒

04

City aerial timelapse城市航拍延时

Vertical 1080P · 15s竖屏 1080P · 15 秒

05

Studio portrait relight棚拍人像重打光

Vertical 1080P · 8s竖屏 1080P · 8 秒

Drag to explore拖动查看
01 / 05
03Private cloud私有云

From racks to models, delivered从机柜到模型,交钥匙交付

We built our own AI data centre to run video generation. Now we build and run yours.我们为自己的视频生成负载建了一座 AI 数据中心,现在也为你建、为你运维。

How it is delivered交付形态

  • Your data centre, or a hosted one你的机房,或托管机房
  • Root stays with youroot 权限始终在你手里
  • Air-gapped if you need it需要时可完全离网

What we stand up交付的技术栈

01

Your facility你的机房

Power, cooling, network电力 · 制冷 · 网络

02

GPU racksGPU 机柜

B300, B200, H200, RTX PRO 6000 and H100B300、B200、H200、RTX PRO 6000、H100

03

InfiniBand fabricInfiniBand 网络

Validated before handover验证通过后移交

04

KubernetesKubernetes 调度

Jobs routed across the fleet任务跨集群路由

05

Your models and API你的模型与 API

In your domain在你的域内

  1. 01

    Survey & design勘测与设计

    We audit power, cooling and network, then design racks that pass your compliance review.勘测电力、制冷与网络,输出能通过合规评审的机柜方案。

  2. 02

    Build & stress-test搭建与压测

    Hardware racked and cabled, then run at full load until the InfiniBand fabric passes.上架布线后满载压测,InfiniBand 网络验证通过后移交。

  3. 03

    Deploy & tune部署与调优

    Kubernetes goes in, your video models land in-domain, throughput is tuned on your real jobs.部署 Kubernetes,视频模型在域内落地,按真实任务调优吞吐。

  4. 04

    Operate持续运维

    Round-the-clock monitoring, a spare-parts SLA and regular capacity reviews. Root stays with you.全天候监控、备件 SLA、定期容量复盘。root 权限始终在你手里。

04GPUs算力

Five GPU series, each with a job五类 GPU,各司其职

Pick by workload. Delivered as a private deployment.按负载选型,私有化交付。

  • B300

    Blackwell Ultra

    288GB HBM3e

    Memory bandwidth显存带宽
    8 TB/s
    Interconnect互联
    NVLink 5 · 1.8 TB/s

    Best for适用

    Very large model training · inference acceleration超大模型训练 · 推理加速

  • B200

    Blackwell

    180GB HBM3e

    Memory bandwidth显存带宽
    8 TB/s
    Interconnect互联
    NVLink 5 · 1.8 TB/s

    Best for适用

    Large-model training · high-throughput inference大模型训练 · 高吞吐推理

  • H200

    Hopper

    141GB HBM3e

    Memory bandwidth显存带宽
    4.8 TB/s
    Interconnect互联
    NVLink 4 · 900 GB/s

    Best for适用

    Large models · high-memory inference大模型 · 大显存推理

  • RTX PRO 6000

    Blackwell

    96GB GDDR7

    Memory bandwidth显存带宽
    1.6 TB/s
    Interconnect互联
    PCIe 5.0

    Best for适用

    Inference · video generation & rendering推理 · 视频生成与渲染

  • H100

    Hopper

    80GB HBM3

    Memory bandwidth显存带宽
    3.35 TB/s
    Interconnect互联
    NVLink 4 · 900 GB/s

    Best for适用

    Model training · fine-tuning模型训练 · 微调

Swipe for more左右滑动查看更多

Routing调度

Every job goes to the GPU that fits it — by model size, resolution, duration and latency.每个任务都按模型规模、分辨率、时长和时延,路由到最合适的 GPU。

  • Model size模型规模
  • Resolution分辨率
  • Duration时长
  • Latency时延
  • Token utilisationtoken 利用率
  • Generation quality生成质量
  • Generation speed生成速度
Scheduler调度器Illustration示意

Incoming jobs任务队列

Text-to-video · 1080P · 30s文生视频 · 1080P · 30 秒Image-to-video · 1080P · 10s图生视频 · 1080P · 10 秒Brand LoRA · fine-tune品牌 LoRA · 微调Long-context inference大显存推理 · 长上下文Batch render · 200 clips批量出片 · 200 条Model training · multi-node模型训练 · 多节点Frontier model training · multi-node超大模型训练 · 多节点
  • B300Very large model training · inference acceleration超大模型训练 · 推理加速
  • B200Large-model training · high-throughput inference大模型训练 · 高吞吐推理
  • H200Large models · high-memory inference大模型 · 大显存推理
  • RTX PRO 6000Inference · video generation & rendering推理 · 视频生成与渲染
  • H100Model training · fine-tuning模型训练 · 微调

Pricing报价

Quoted per project. Tell us your models, cluster size and timeline.按项目报价。告诉我们你的模型、集群规模和时间表。

Get a quote获取报价

Compute, quietlyextraordinary.算力,安静而非凡。

Private deployments, quoted per project. Write to us.私有化部署,按项目报价。欢迎来信。

Contact联系方式info@hanqing.com

Private deployments only仅提供私有化部署

No public sign-up, no shared console. Each cluster is built for one customer. For a quote or a conversation, write to us.没有公开注册,也没有共享控制台,每套集群只为一家客户搭建。报价或洽谈,请发邮件。

Email邮箱
info@hanqing.com
Available GPUs可部署算力
B300 · B200 · H200 · RTX PRO 6000 · H100