GPU × use-case guide · Vision-language

Is the GB200 NVL72 (per GPU) a good GPU for vision-language?

The GB200 NVL72 (per GPU) is a blackwell NVIDIA GPU with 192 GB HBM3e (8000 GB/s), 2250 TFLOPS BF16 / 4500 TFLOPS FP8, and a 1200 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 GB200 NVL72 (per GPU) measures up.

What models fit on a single GB200 NVL72 (per GPU)?

Weights only, reserving ~25% of the 192 GB for KV cache, activations and fragmentation. ✓ = fits on one card.

Largest single-card fit: Qwen 2.5 72B at BF16, Qwen 2.5 72B at FP8, Qwen 2.5 72B at INT4. Bigger models need tensor-parallel across 72 cards.

GB200 NVL72 (per GPU) for vision-language, specifically

Vision-language is context-heavy, so the KV cache — not the weights — is what fills the 192 GB. On the GB200 NVL72 (per GPU) 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.

GB200 NVL72 (per GPU) pricing across providers

Verdict

With 192 GB, the GB200 NVL72 (per GPU) is a data-center-class card that comfortably handles vision-language for models up to Qwen 2.5 72B at full precision on a single card — a strong pick if your budget supports ~$8.5/hr.