Can Mac Studio M3 Ultra 96GB run OLMo-2-1124-13B-Instruct?

Apple · 96 GB LPDDR5 · 819 GB/s bandwidth · allenai · 13.7B · model ctx up to NoneK

METAL

Yes — the Mac Studio M3 Ultra 96GB (96 GB LPDDR5) can run OLMo-2-1124-13B-Instruct. At Q4 with the default 8K context it needs ≈ 11.28 GB VRAM (weights 7.54 GB + KV cache 2.74 GB + 1 GB runtime overhead), which fits within 96 GB. Decode at Q4/8K: ≈ 38.0 tok/s (estimated, batch 1). FP16 (≈ 31.14 GB) also fits, at ≈ 10.5 tok/s (estimated). Estimates come from memory-bandwidth math; rows tagged measured override estimates. Relative ranking is more reliable than absolute tok/s. Methodology. Buy Mac Studio M3 Ultra 96GB on Amazon →

Fit & speed by quant and context

Quant	Context	VRAM needed	Fits?	tok/s (decode)
Q4	4K	9.91 GB	Yes	38.0 ± est.
Q4	8K default	11.28 GB	Yes	38.0 ± est.
Q4	32K	19.5 GB	Yes	38.0 ± est.
FP16	4K	29.77 GB	Yes	10.5 ± est.
FP16	8K default	31.14 GB	Yes	10.5 ± est.
FP16	32K	39.36 GB	Yes	10.5 ± est.

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	Mac Studio M3 Ultra 96GB VRAM
Q4	7.54 GB	2.74 GB	1 GB	11.28 GB	96 GB
FP16	27.4 GB	2.74 GB	1 GB	31.14 GB	96 GB

Other GPUs that run OLMo-2-1124-13B-Instruct

• B200 192GB SXM — ≈371.6 tok/s Q4 est.
• MI300X 192GB — ≈246.2 tok/s Q4 est.
• H200 141GB SXM5 — ≈223.0 tok/s Q4 est.
• H100 80GB SXM5 — ≈155.6 tok/s Q4 est.
• Gaudi 3 128GB — ≈152.2 tok/s Q4 est.
• A100 80GB SXM4 — ≈94.7 tok/s Q4 est.
• RTX 5090 32GB — ≈83.2 tok/s Q4 est.
• RTX 4090 24GB — ≈46.8 tok/s Q4 est.

All GPUs for OLMo-2-1124-13B-Instruct →

Other models for the Mac Studio M3 Ultra 96GB

• MiniCPM5-1B — ≈482.6 tok/s Q4 est.
• Qwen3.5-2B — ≈260.6 tok/s Q4 est.
• DeepSeek-R1-0528-Qwen3-8B-layer-mix-bpw-3.8-mlx — ≈226.6 tok/s Q4 est.
• Qwen3.6-35B-A3B — ≈173.7 tok/s Q4 est.
• Qwen3.5-35B-A3B — ≈173.7 tok/s Q4 est.
• GLM-4.7-Flash — ≈173.7 tok/s Q4 est.
• Qwen3-30B-A3B-Instruct-2507 — ≈173.7 tok/s Q4 est.
• Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16 — ≈173.7 tok/s Q4 est.

All models on the Mac Studio M3 Ultra 96GB →