We take cryptocurrency — so the coins you hold can buy real hardware.Pay in crypto, spend it on real hardware. Free insured shipping worldwide over $399. Plain, unbranded boxes, sent from Germany.Free insured shipping over $399. Your rate is locked for 30 minutes once checkout opens.Rate locked 30 minutes. Every item sealed, serial-checked and covered by a 24-month warranty.Sealed, 24-month warranty.

Saved items Sign in
Full-height dual-slot PCIe card with passive silver heatsink and 16-pin power connector

NVIDIA

NVIDIA H100 NVL 94GB HBM3 AI accelerator card 900-21010-0020-000

SKU CH-H57X-1JFYZ MPN 900-21010-0020-000

Dual-slot FHFL accelerator with 94 GB HBM3 and Tensor cores delivering 3,341 teraFLOPS FP8 for AI workloads

  • InterfacePCI Express 5.0 x16
  • Chipset ManufacturerNVIDIA
  • GPUH100 Tensor Core GPU
  • Core ClockBase: 1080 MHz Boost: 1785 MHz
  • Memory Size94GB
  • Memory Clock2619 MHz
  • 94 GB HBM3 memory enables massive model training
  • 3 341 teraFLOPS FP8 Tensor performance accelerates inference
  • PCI Express 5.0 x16 interface delivers high bandwidth
  • Fanless dual-slot design fits dense server racks
  • Configurable 200 W to 400 W power range matches deployment needs

H100 NVL 94GB AI accelerator for servers

The NVIDIA H100 NVL is a full-height, full-length PCIe 5.0 x16 card built for data-centre AI training and inference. It carries 94GB of HBM3 memory and a passive dual-slot cooler that relies on server airflow. The GPU operates at a 1080 MHz base clock and a 1785 MHz boost clock.

Fanless design and configurable power modes

A fanless cooler removes moving parts and reduces acoustic load in dense racks. The 16-pin auxiliary connector supports three cable-strapped power envelopes: a 450 W or 600 W mode with a 400 W default maximum, and a 300 W mode with a 310 W default maximum. Both modes share a 200 W minimum and a 310 W compliance limit.

Suited for HPC clusters needing high memory

This card targets workloads that benefit from 94GB HBM3 capacity and Tensor-core throughput such as FP8 at 3341 teraFLOPS. It is not for workstations lacking FHFL 10.5-inch clearance or auxiliary 16-pin power delivery. Buyers who require active cooling or a single-slot form factor should look elsewhere.

Highlights

  • 94 GB HBM3 memory enables massive model training
  • 3 341 teraFLOPS FP8 Tensor performance accelerates inference
  • PCI Express 5.0 x16 interface delivers high bandwidth
  • Fanless dual-slot design fits dense server racks
  • Configurable 200 W to 400 W power range matches deployment needs

Specifications

BrandNVIDIA
ModelH100 NVL
Part Number900-21010-0020-000
InterfacePCI Express 5.0 x16
Chipset ManufacturerNVIDIA
GPUH100 Tensor Core GPU
Core ClockBase: 1080 MHz
Boost: 1785 MHz
Memory Size94GB
Memory Clock2619 MHz
Memory TypeHBM3
CoolerFanless
AI FeaturesAI Accelerated
Operating Systems SupportedWindows Hardware Quality Lab (WHQL):
- Windows 10, Windows 11
- Windows Server 2019, Windows Server 2022
Auxiliary Power Connector16-pin auxiliary power connector
PowerPCIe 16-pin cable strapped for 450 W or 600 W power mode:
> 400 W maximum (default)
> 310 W power compliance limit1
> 200 W minimum
PCIe 16-pin cable strapped for 300 W power mode:
> 310 W maximum (default)
> 310 W power compliance limit
> 200 W minimum
Form FactorFull-height, full-length (FHFL) 10.5"
Slot WidthDual-slot
EnvironmentAmbient operating temperature: 0°C to 50°C
Ambient operating temperature (short term): -5°C to 55°C
Storage temperature: -40°C to 75°C
Operating humidity (short term): 5% to 93% relative humidity
Operating humidity: 5% to 85% relative humidity
Storage humidity: 5% to 95% relative humidity
Shipping weight1.47 kg
Package size406 × 229 × 89 mm

Questions about this item

When does the fanless H100 NVL cooler need servicing or replacement?

The card uses a passive fanless cooler with no moving parts, so there is no fan to replace; thermal paste may need attention after several years of continuous operation.

What power ceiling limits sustained compute on the 16-pin connector?

In the default 450 W or 600 W cable mode the card is capped at 400 W maximum with a 310 W compliance limit, so the 310 W limit is reached first under sustained load.

How much noise and heat does the passive card produce at the 310 W compliance limit?

With no fans the card generates zero acoustic noise; all 310 W becomes heat that must be removed by the server's airflow, and the ambient operating range is 0 °C to 50 °C.

New to this? Buying server hardware with crypto — what to check, what it costs to pay, and what is never asked for.

Also in this aisle

People compared these

Same shelf, same checkout — eight coins and a 30-minute rate lock.