H100
NVIDIA's flagship AI accelerator GPU; the de facto standard for training and serving large language models.
Think of it like
The gold standard; if you have the money, you buy H100s.
Example
Training a 70B-parameter model typically uses 16–128 H100 GPUs depending on strategy.
How it actually works
H100 has 80GB HBM3 memory, 141 TFLOPS FP8, and NVLink for scaling. Prices are $40k per GPU, but cloud pricing is lower.
For product teams
Supply-constrained; huge profit margins for NVIDIA.
For engineers
H100 is the baseline assumption for LLM infrastructure; A-100s are budget option, H100s are premium.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome