Can RTX 5090 32GB run NVIDIA-Nemotron-3-Super-120B-A12B-BF16?

NVIDIA · 32 GB GDDR7 · 1792 GB/s bandwidth · NVIDIA · 120.0B (active 12.0B) · MoE · model ctx up to 256K

No — NVIDIA-Nemotron-3-Super-120B-A12B-BF16 at Q4 needs ≈ 68.2 GB even at 4K context, beyond the RTX 5090 32GB's 32 GB GDDR7. It would only run with heavy CPU/disk offload. Decode at Q4/8K: ≈ 57.7 tok/s (estimated, batch 1). FP16 needs ≈ 243.4 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	68.2 GB	No	59.8 est.
Q4	8K default	69.4 GB	No	57.7 est.
Q4	32K	76.6 GB	No	47.4 est.
FP16	4K	242.2 GB	No	—
FP16	8K default	243.4 GB	No	—
FP16	32K	250.6 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	RTX 5090 32GB VRAM
Q4	66.0 GB	2.4 GB	1 GB	69.4 GB	32 GB
FP16	240.0 GB	2.4 GB	1 GB	243.4 GB	32 GB

Other GPUs that run NVIDIA-Nemotron-3-Super-120B-A12B-BF16

• B200 192GB SXM — ≈424.2 tok/s Q4 est.
• MI300X 192GB — ≈281.1 tok/s Q4 est.
• H200 141GB SXM5 — ≈254.5 tok/s Q4 est.
• H100 80GB SXM5 — ≈177.7 tok/s Q4 est.
• Gaudi 3 128GB — ≈173.7 tok/s Q4 est.
• A100 80GB SXM4 — ≈108.1 tok/s Q4 est.
• Mac Studio M3 Ultra 96GB — ≈43.4 tok/s Q4 est.
• Mac Studio M2 Ultra 192GB — ≈42.4 tok/s Q4 est.

All GPUs for NVIDIA-Nemotron-3-Super-120B-A12B-BF16 →

Other models for the RTX 5090 32GB

• MiniCPM5-1B — ≈1055.9 tok/s Q4 est.
• Qwen3.5-2B — ≈570.2 tok/s Q4 est.
• GLM-4.7-Flash — ≈380.1 tok/s Q4 est.
• Qwen3-30B-A3B-Instruct-2507 — ≈380.1 tok/s Q4 est.
• Qwen3.5-35B-A3B — ≈380.1 tok/s Q4 est.
• Qwen3.6-35B-A3B — ≈380.1 tok/s Q4 est.
• Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16 — ≈380.1 tok/s Q4 est.
• Ornith-1.0-35B — ≈380.1 tok/s Q4 est.

All models on the RTX 5090 32GB →