Home / GPU / H100
H100 GPU Capacity Sourcing
The NVIDIA H100 is built on the Hopper architecture and ships in SXM5 and PCIe form factors, with up to 80GB of HBM3 memory per GPU. NVLink and NVSwitch connect multiple H100s for multi-GPU training and inference. FindMeAGPU sources H100 capacity from cloud and colocation providers on request.
Specifications
- Architecture
- Hopper
- Memory
- Up to 80GB HBM3
- Form factors
- SXM5, PCIe
- Interconnect
- NVLink / NVSwitch
Typical workloads
- LLM inference
- Training
- Fine-tuning
How sourcing works
- 1.
Tell us what you need
Submit GPU model, quantity, region, start date, and duration.
- 2.
We check across providers
We take your requirement to GPU cloud and colocation providers and check current capacity.
- 3.
You receive quotes
We contact you directly with quotes from providers that can meet your requirement.
FAQ
What is the NVIDIA H100 used for?
The H100 is used for large language model training, fine-tuning, and inference, along with other GPU-accelerated compute such as scientific simulation and recommendation systems.
Is H100 capacity available now?
Availability varies by provider, region, and time and changes frequently. Submit a request and we check current capacity across our provider network.
What form factors does the H100 come in?
The H100 ships in SXM5 (higher power and memory bandwidth) and PCIe (single or dual-slot) variants. Cloud providers most commonly offer SXM5 nodes for multi-GPU clusters.
How many H100s can be connected together?
NVLink and NVSwitch connect up to eight H100 GPUs within a single server node, and InfiniBand networking connects multiple nodes into larger clusters.
Does FindMeAGPU sell H100 hardware directly?
No. FindMeAGPU is a sourcing service — we take your requirement to multiple GPU cloud and colocation providers and help you find available capacity. We do not own or sell hardware.