GPU × use-case guide · Vision-language

Is the H100 NVL a good GPU for vision-language?

The H100 NVL is a hopper NVIDIA GPU with 94 GB HBM3 (3938 GB/s), 835 TFLOPS BF16 / 1671 TFLOPS FP8, and a 400 W TDP. Vision-language workloads care most about extra VRAM for the vision encoder and image tokens on top of the language model. Here's how the H100 NVL measures up.

What models fit on a single H100 NVL?

Weights only, reserving ~25% of the 94 GB for KV cache, activations and fragmentation. ✓ = fits on one card.

Largest single-card fit: Gemma 2 27B at BF16, Llama 3.3 70B at FP8, Qwen 2.5 72B at INT4. Bigger models need tensor-parallel across 8 cards.

H100 NVL for vision-language, specifically

Vision-language is context-heavy, so the KV cache — not the weights — is what fills the 94 GB. On the H100 NVL you'll trade context length against batch size: long prompts mean fewer concurrent requests. Because it runs offline, batch aggressively to push tokens-per-dollar down. Size it precisely on the calculator.

H100 NVL pricing across providers

Verdict

With 94 GB, the H100 NVL is a data-center-class card that comfortably handles vision-language for models up to Gemma 2 27B at full precision on a single card — a strong pick if your budget supports ~$4.1/hr.