Professional HardwareMore Performance

New · Teklif ile tedarik

NVIDIA DGX H100

Brand: NVIDIA SKU: AI-NVIDIA-DGX-H100

The NVIDIA DGX H100 is the reference AI system of the Hopper generation, with eight H100 Tensor Core GPUs and 640 GB of total GPU memory. Widely used for model training and inference with 32 petaFLOPS of FP8 performance, this platform is starting to give way to H200 and Blackwell systems, so its availability and configuration are verified separately at the quotation stage. It is also suitable for those who want to expand existing DGX H100 clusters.

  • Garanti ve servis koşulları teklif aşamasında üretici ve tedarik kanalına göre netleşir.
  • Delivery and installation terms are finalised at the quotation stage
  • Your card details are not stored on our site

The NVIDIA DGX H100 is the reference AI system of the Hopper generation, with eight H100 Tensor Core GPUs and 640 GB of total GPU memory. Widely used for model training and inference with 32 petaFLOPS of FP8 performance, this platform is starting to give way to H200 and Blackwell systems, so its availability and configuration are verified separately at the quotation stage. It is also suitable for those who want to expand existing DGX H100 clusters.

NVIDIA DGX H100 overview

The DGX H100 is NVIDIA's first DGX system to bring the Hopper architecture to enterprise AI infrastructure. It forms the basis of many organisations' production AI clusters.

Hardware architecture

According to the NVIDIA datasheet, the system has 8 × H100 Tensor Core GPUs (640 GB of total GPU memory) and 4 × NVSwitch; FP8 performance is 32 petaFLOPS. On the network side, four OSFP ports serve eight single-port NVIDIA ConnectX-7 VPI cards. Maximum system power is given as 10.2 kW in the NVIDIA datasheet (~11.3 kW in one reseller source; the discrepancy has been reported).

Use cases in AI workloads

Model training and fine-tuning, inference, HPC, data analytics and expansion of existing Hopper clusters.

Business advantages

Broad software compatibility and a mature ecosystem reduce development and operating costs.

Scalability and infrastructure requirements

It requires rack installation and more than 10 kW of power; it scales as a DGX BasePOD / SuperPOD with InfiniBand. DGX H200 or DGX B200 should also be considered for new projects.

Who is the NVIDIA DGX H100 for?

Suitable for organisations that want to expand an existing DGX H100 cluster with the same model or that have standardised on the Hopper software environment.

Key features

  • 8 × NVIDIA H100 Tensor Core GPU
  • 640 GB total GPU memory
  • FP8: 32 petaFLOPS
  • 4 × NVIDIA NVSwitch
  • 8 × tek portlu ConnectX-7 VPI (4 × OSFP)
  • NVIDIA DGX OS

Technical Specifications

MarkaNVIDIA
ModelDGX H100
Product typeRaf tipi yapay zekâ sunucusu
GPU8 × NVIDIA H100 Tensor Core GPU
Total GPU memory640 GB
FP8 performance32 petaFLOPS
NVSwitch4 pcs
Network4 × OSFP ports, 8 × single-port NVIDIA ConnectX-7 VPI
Maximum system power10.2 kW (NVIDIA datasheet)
SoftwareNVIDIA DGX OS, NVIDIA AI Enterprise
Tedarik durumuVerified at the quotation stage due to new-generation systems

Procurement and corporate quotation process

The NVIDIA DGX H100 is supplied by special order. Once quantity, network and storage configuration, installation and NVIDIA support options are determined, a corporate quotation including price and delivery time is prepared. Stating your intended use and delivery country in the quote form speeds up the process.

Similar Products