8x NVIDIA HGX H200 Liquid-Cooled 5U AI Training Server — Dual Xeon

A 5U liquid-cooled rack server with eight NVLink-connected NVIDIA HGX H200 GPUs and dual 5th Gen Intel Xeon processors for large-scale AI training and HPC.

A partir de$ 27 400
Estados Unidos (USD)

Alterar moeda

A moeda dos preços depende do país. Escolha um país para trocá-la.

Pedido personalizado
Feito sob medida para os seus requisitos. Você escolhe o tamanho e a configuração.

This is a 5U rack-mount server built to train large AI models and run HPC workloads at data-center scale. Eight NVIDIA HGX H200 GPUs sit fully interconnected over NVLink at 900 GB/s, so a model can span all eight accelerators without choking on inter-GPU traffic — the bottleneck that slows LLM training, machine learning and generative AI work on loosely coupled hardware. Two 5th Gen Intel Xeon Scalable processors (up to 350 W each) handle the host side: data loading, preprocessing and feeding the GPUs, which keeps the accelerators busy instead of idling on I/O.

Direct Neptune liquid cooling pulls up to 95% of the heat off the CPUs, GPUs and NVLink switches, so the machine holds full clocks under sustained load rather than throttling the way an air-cooled chassis does on long training runs. That matters for the venues that buy these — AI research labs, HPC facilities and colocation operators running training clusters — where a thermal ceiling directly costs compute hours.

  • Memory and storage: up to 4 TB of DDR5 at 5600 MHz across 32 DIMMs, up to 12 hot-swap U.2/U.3 NVMe SSDs, plus 2x M.2 for boot — bandwidth and fast local scratch to cut I/O waits on large datasets.
  • Expansion and power: up to 10 PCIe Gen5 x16 FHHL slots for networking and fabric, and up to 8 hot-swap power supplies for full N+N redundancy.
  • Management: XClarity Controller 2 (XCC2) for remote deployment and monitoring.
  • OS support: RHEL, Ubuntu, AlmaLinux, Rocky Linux and ESXi.

Ficha técnica

Form factor
5U rack
Processor
2x 5th Gen Intel Xeon Scalable, up to 350W each
GPU
8x NVIDIA HGX H200, NVLink 900 GB/s
Max memory
4 TB
Memory type
DDR5 up to 5600 MHz, 32 DIMMs
Storage
Up to 12x U.2/U.3 hot-swap NVMe SSD + 2x M.2 boot
I/O expansion
Up to 10x PCIe Gen5 x16 FHHL
Power supplies
Up to 8x hot-swap, N+N redundancy
Cooling
Direct (Neptune) liquid cooling, up to 95% heat removal
Management
XClarity Controller 2 (XCC2)
OS support
RHEL, Ubuntu, AlmaLinux, Rocky Linux, ESXi

Perguntas frequentes

How are the eight H200 GPUs connected inside this server?

All eight NVIDIA HGX H200 GPUs are fully interconnected over NVLink at 900 GB/s, so they operate as one pooled accelerator. A single large model can span all eight without the inter-GPU bandwidth limits you hit when GPUs talk only over PCIe or a slower fabric.

Why does this server use liquid cooling instead of fans?

Direct liquid cooling removes up to 95% of the heat from the CPUs, GPUs and NVLink switches, which an air-cooled chassis cannot match at this density. That keeps the hardware at full clocks through long training runs and HPC simulations rather than throttling once temperatures climb, so you get sustained compute instead of a thermal ceiling.

What memory and storage can this AI server hold?

It scales to 4 TB of DDR5 at 5600 MHz across 32 DIMMs, plus up to 12 hot-swap U.2/U.3 NVMe SSDs and 2x M.2 for boot. The combination gives enough bandwidth and fast local scratch space to reduce I/O stalls while loading and preprocessing large training datasets.

Which operating systems does this server support?

The platform supports RHEL, Ubuntu, AlmaLinux, Rocky Linux and ESXi, and it ships with XClarity Controller 2 (XCC2) for remote deployment and monitoring.

What workloads is this 5U server built for?

It targets large language model training, generative AI development, machine learning and large-scale HPC simulation — workloads that need many tightly coupled GPUs and high host throughput from the dual 5th Gen Intel Xeon processors.

Does the server have redundant power for continuous operation?

Yes. It takes up to 8 hot-swap power supplies configured for full N+N redundancy, so it keeps running through a PSU failure, and the units can be swapped without downtime.

O frete está incluído no preço?

Nos itens em estoque, ou seja, tudo o que dá para adicionar à especificação, o frete já está no preço anunciado. Nos itens feitos sob encomenda, a entrega é grátis (isso aparece na página do produto) ou calculada no pedido, conforme o destino e o volume.

Quanto tempo leva a entrega?

Um item em estoque costuma chegar em uma a três semanas; um item sob encomenda leva de três a seis, porque trabalhamos direto com os fabricantes e a produção e a logística precisam de tempo antes do envio. Depois do pedido, o gerente responsável pela sua conta confirma um prazo mais preciso. Atrasos no desembaraço aduaneiro do país de destino não são de nossa responsabilidade.

Os impostos e taxas de importação estão incluídos?

Na maioria dos casos, sim. Enviamos de preferência no regime DDP (Delivered Duty Paid): os impostos e as taxas de importação já ficam resolvidos e você não paga nada a mais na entrega. Em casos raros o envio pode ser DDU (Delivered Duty Unpaid), e aí você acerta as taxas de importação direto com a transportadora, mas só se isso tiver sido combinado com você antes.

Os produtos têm garantia?

Sim. A maioria dos equipamentos tem garantia do fabricante de um a dois anos contra defeitos de fábrica e falhas técnicas em uso normal, e o prazo exato está na página de cada produto. Se aparecer um defeito, providenciamos a troca ou o reembolso conforme a nossa Política de trocas e devoluções, e a nossa equipe acompanha você em cada etapa da solicitação.

Solicitar orçamento