Can Titan X Pascal 12GB run DeepSeek-R1?

NVIDIA · 12 GB GDDR5X · 480 GB/s bandwidth · DeepSeek · 684.5B (active 37B) · MoE · model ctx up to 128K

CUDA

No — DeepSeek-R1 at Q4 needs ≈ 377.94 GB even at 4K context, beyond the Titan X Pascal 12GB's 12 GB GDDR5X. It would only run with heavy CPU/disk offload. FP16 needs ≈ 1370.92 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	377.94 GB	No	—
Q4	8K default	378.4 GB	No	—
Q4	32K	381.18 GB	No	—
FP16	4K	1370.46 GB	No	—
FP16	8K default	1370.92 GB	No	—
FP16	32K	1373.7 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	Titan X Pascal 12GB VRAM
Q4	376.48 GB	0.93 GB	1 GB	378.4 GB	12 GB
FP16	1369.0 GB	0.93 GB	1 GB	1370.92 GB	12 GB

Other GPUs that run DeepSeek-R1

• DGX Spark (GH200 480GB) — ≈8.6 tok/s Q4 est.

All GPUs for DeepSeek-R1 →

Other models for the Titan X Pascal 12GB

• MiniCPM5-1B — ≈282.8 tok/s Q4 est.
• Qwen3.5-2B — ≈152.7 tok/s Q4 est.
• DeepSeek-R1-0528-Qwen3-8B-layer-mix-bpw-3.8-mlx — ≈132.8 tok/s Q4 est.
• Llama-3.2-3B-Instruct — ≈95.5 tok/s Q4 est.
• Qwen3-4B-Instruct-2507 — ≈76.4 tok/s Q4 est.
• NVIDIA-Nemotron-3-Nano-4B-BF16 — ≈76.4 tok/s Q4 est.
• Qwen3-4B-Thinking-2507 — ≈76.4 tok/s Q4 est.
• gemma-4-E2B-it — ≈59.9 tok/s Q4 est.

All models on the Titan X Pascal 12GB →