NVIDIA H100 PCIe 80GB HBM2e Data Center GPU | 900-21010-0000-000 / 90SKC000-M7LAN0
$21,834.25
NVIDIA H100 PCIe 80GB Hopper data center accelerator with HBM2e memory. Passive, full-height dual-slot design; up to 350 W with qualified power configuration. Confirm complete OEM label, server compatibility, cooling, cabling and supplied package. Separate from H100 SXM and NVL variants.
Explore Related
Related products, categories and research.
Graphics Card
NVIDIA H100 PCIe
The NVIDIA H100 PCIe is a Hopper data center accelerator with 80GB HBM2e memory. Its passive cooling and dual-slot form factor require a suitable server configuration. This page identifies the PCIe 80GB model; H100 SXM and H100 NVL are separate variants. Confirm the full board label and OEM reference against the offered unit.
Product overview
NVIDIA H100 PCIe accelerator overview
The NVIDIA H100 PCIe is a Hopper data center accelerator with 80GB HBM2e memory. Its passive cooling and dual-slot form factor require a suitable server configuration. This page identifies the PCIe 80GB model; H100 SXM and H100 NVL are separate variants. Confirm the full board label and OEM reference against the offered unit.
The product record and technical documentation should be used together to confirm the exact model, compatibility, included accessories and configuration before installation or deployment.
Three checks before purchase
1
Identify the exact model
Match the manufacturer model, part number and product-label details to the quotation.
2
Confirm compatibility
Review system, power, software, cooling and integration requirements that apply to this product type.
3
Approve the supplied scope
Confirm condition, included items, warranty, documentation and delivery terms before order approval.
Product images
Product reference images
Use the product images together with the final quotation to confirm the supplied model, configuration and condition.




NVIDIA H100 PCIe 80GB hardware highlights
- Hopper architecture with 80GB HBM2e memory.
- PCI Express Gen5 x16 host interface.
- Passive, full-height, full-length dual-slot design.
- Multi-Instance GPU support for up to seven instances.
Use the exact PCIe model documentation for system planning. Do not substitute memory, power or interconnect specifications from an SXM module or an NVL configuration.
AI and compute workload planning
The NVIDIA H100 PCIe can be evaluated for supported AI training, inference and high-performance computing workloads. Begin with the application framework, memory demand and host platform rather than assuming a particular performance result from the model name. CPU resources, storage, networking and software configuration also affect the complete system.
For shared workloads, review the supported MIG profiles and the software requirements for each application. Partitioning changes the resources available to each instance. Validate the intended deployment using representative workloads before selecting a production configuration.
NVIDIA H100 PCIe specifications
| Model | NVIDIA H100 PCIe 80GB |
|---|---|
| Architecture | Hopper |
| Memory | 80GB HBM2e; 5120-bit |
| Peak memory bandwidth | 2000 GB/s in NVIDIA product brief |
| Host interface | PCIe Gen5 x16 |
| Cooling | Passive; host airflow required |
| Form factor | Full-height, full-length 10.5-inch; dual-slot |
| Board power | Up to 350 W; depends on qualified power configuration |
| Power connection | One PCIe 16-pin auxiliary connector |
| MIG | Up to seven GPU instances |
| Catalog references | 900-21010-0000-000 / 90SKC000-M7LAN0; verify supplied label |
These are PCIe model-level reference values. The exact OEM label, firmware and host qualification must be checked for the supplied board. A catalog reference does not establish that two OEM packages have identical contents.
Server, power and cooling compatibility
- Confirm support for the PCIe 80GB accelerator in the intended server.
- Check dual-slot clearance, retention hardware and airflow through the passive heatsink.
- Verify qualified auxiliary power cabling and the intended board-power configuration.
- Match the operating system, driver and CUDA/framework releases using current NVIDIA and server-vendor documentation.
- Confirm any multi-GPU bridge and topology requirements separately.
The 16-pin power connection is part of a qualified server power design. Connector appearance alone does not establish that a cable or adapter supports the intended operating mode. Follow the integration instructions for the exact host and accelerator.
Official product reference and supplied package
NVIDIA H100 PCIe product brief (PDF) describes the 80GB HBM2e model. Consult NVIDIA MIG documentation for partitioning requirements.
Confirm condition, provenance, included power adapters, retention parts, bridges and warranty in the written offer. Software support described in model literature does not establish licence entitlement for this individual unit. Request the complete NVIDIA/OEM label before accepting a configuration.
Buyer questions
Frequently asked questions
Questions about the offered PCIe accelerator.
What is NVIDIA H100 PCIe?
The NVIDIA H100 PCIe is a Hopper data center accelerator with 80GB HBM2e memory. Its passive cooling and dual-slot form factor require a suitable server configuration. This page identifies the PCIe 80GB model; H100 SXM and H100 NVL are separate variants. Confirm the full board label and OEM reference against the offered unit.
What should be checked before ordering?
Exact manufacturer model / part number and GPU variant; VRAM capacity and memory type; PCIe slot generation, slot width and chassis clearance.
How is the exact configuration confirmed?
Hardware revision, firmware or software, cooling, accessories, interfaces and included modules can vary by version. Confirm the exact supplied configuration in the written quotation.
Which items are included?
Only items explicitly listed in the quotation and packing scope should be treated as included. Product images may show accessories or system components that are not part of the final offer.
B2B quotation
Request a PCIe 80GB quote
Include delivery country, quantity, required configuration and target purchase date for a destination-specific offer.
Recently viewed
Add AI Robot Supplier as a preferred source on Google.