Can B200 192GB SXM run AI21-Jamba-Large-1.5?

NVIDIA · 192 GB HBM3e · 8000 GB/s bandwidth · ai21labs · 398.6B · model ctx up to NoneK

CUDA

No — AI21-Jamba-Large-1.5 at Q4 needs ≈ 260.09 GB even at 4K context, beyond the B200 192GB SXM's 192 GB HBM3e. It would only run with heavy CPU/disk offload. Decode at Q4/8K: ≈ 14.9 tok/s (estimated, batch 1). FP16 needs ≈ 877.92 GB — does not fit on this card. Estimates come from memory-bandwidth math; rows tagged measured override estimates. Relative ranking is more reliable than absolute tok/s. Methodology.

Fit & speed by quant and context

Quant	Context	VRAM needed	Fits?	tok/s (decode)
Q4	4K	260.09 GB	No	19.9 ± est.
Q4	8K default	299.95 GB	No	14.9 ± est.
Q4	32K	539.11 GB	No	4.6 ± est.
FP16	4K	838.06 GB	No	—
FP16	8K default	877.92 GB	No	—
FP16	32K	1117.08 GB	No	—

Fits = weights + KV(ctx) + 1 GB overhead ≤ 95% of VRAM. Measured anchors are context-agnostic; the fit verdict is recomputed per context. A missing tok/s means the model is far beyond this card (offload-only territory).

VRAM breakdown at 8K context

Quant	Weights	KV cache	Overhead	Total needed	B200 192GB SXM VRAM
Q4	219.23 GB	79.72 GB	1 GB	299.95 GB	192 GB
FP16	797.2 GB	79.72 GB	1 GB	877.92 GB	192 GB

Other GPUs that run AI21-Jamba-Large-1.5

• DGX Spark (GH200 480GB) — ≈0.8 tok/s Q4 est.

All GPUs for AI21-Jamba-Large-1.5 →

Other models for the B200 192GB SXM

• MiniCPM5-1B — ≈4713.8 tok/s Q4 est.
• Qwen3.5-2B — ≈2545.5 tok/s Q4 est.
• DeepSeek-R1-0528-Qwen3-8B-layer-mix-bpw-3.8-mlx — ≈2213.4 tok/s Q4 est.
• Qwen3.6-35B-A3B — ≈1697.0 tok/s Q4 est.
• Qwen3.5-35B-A3B — ≈1697.0 tok/s Q4 est.
• GLM-4.7-Flash — ≈1697.0 tok/s Q4 est.
• Qwen3-30B-A3B-Instruct-2507 — ≈1697.0 tok/s Q4 est.
• Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16 — ≈1697.0 tok/s Q4 est.

All models on the B200 192GB SXM →