NVIDIA H100 NVL 94GB PCI-E 5.0 (Passive)


NVIDIA H100 NVL 94GB PCI-E 5.0 (Passive) Main Picture

The NVIDIA H100 Tensor Core GPU is a powerful, highly optimized processing platform for AI inference and other demanding GPU-accelerated applications. It supports floating-point operations with precision from FP64 to FP8 as well as INT8 (integer) calculations - making it a single accelerator for pretty much every compute workload. With 94GB of high-bandwidth memory on a massive 6016-bit bus, it also outpaces traditional consumer and even professional video cards by a wide margin when it comes to memory-intensive jobs.

This H100 NVL variant allows pairs of cards to be connected via three bridges so they can utilize NVLink: a very high-speed interconnect for improved multi-GPU operation. Each card is housed in a traditional 2-slot graphics card form factor and fits in a standard PCI-Express 5.0 x16 slot, but the fanless heatsink means these are only functional in purpose-built systems that have been designed to fit and cool such powerful GPUs.


Chipset Manufacturer NVIDIA
Product Category Data Center
Motherboard Connection PCI Express 5.0 x16
Cooling Method Passive Heatsink
Core Specifications
Core Speed 1,080 MHz
Boost Speed 1,785 MHz
Processors 14,592
Memory Specifications
Onboard Memory 94GB
Memory Type HBM3
Memory Speed 2,619 MHz
Memory Bus Width 6016-bit
Bandwidth 3,938 GB/s
Power Connectors
Plug 1 16-pin PCIe
Length 267 mm (10.5 in)
Height 112 mm (4.4 in)
Width 42 mm (1.7 in)
Net Weight 1.28 kg (2.8 lbs)

Utilizes 3x NVIDIA RTX A6000 NVLink Bridge per pair of GPUs