Skip to content

NVIDIA® H100 80 GB SXM5 InfiniBand

Location

Pricing

2048 cards available

Duration Cost Upfront payment
On-demand (hourly) $4.95/GPU-hr $0
Month-to-month $3.75/GPU-hr 100% of total
6 months $3.50/GPU-hr 25% of total
1 year $2.75/GPU-hr 25% of total
2 years $2.51/GPU-hr 25% of total
3 years $2.25/GPU-hr 25% of total

Configuration

Component Details
GPU Model NVIDIA® H100 80GB SXM5
Architecture NVIDIA® Hopper™
Motherboard Asus
GPUs per node 8
CPU Model Intel® Xeon™ Platinum 8480+ 2
CPU Cores 112
VRAM Type HBM3
VRAM Size 80GB (640GB total)
RAM 2TB
RAM type DDR5
Memory bandwidth
  • 4th-generation NVIDIA® NVLink™
  • 900 GB/s total bandwidth for multi-GPU I/O ops
Form factor HGX with a 7U form factor, leveraging SMX5 GPUs
Thermal Design Power (TDP) 9.19 kW
Processing Power TensorCore GPU & Transformer engine
Interconnect Inter-GPU communication utilizes the 4th-generation NVIDIA® NVLink™, boasting a total bandwidth of 900 GB/sec for multi-GPU I/O ops. Communication between GPUs across nodes occurs through InfiniBand Quantum 2 switches, offering a speed of 3200 GB/sec.
Cooling 45KW High-Density Racks utilize RDHx technology to efficiently manage cooling, addressing heat dissipation while ensuring optimal efficiency and reliability.
Deep Learning Framework Support Seamlessly integrated with the NVIDIA NGC registry, which encompasses all the leading deep learning frameworks
Compatibility Linux-based operating systems
Local storage
  • 2x1.92TB NVMe U.2 Gen5
  • 8x7.68TB NVMe U.2 Gen5