Can Tesla P4 8GB run DeepSeek-V4-Pro?

NVIDIA · 8 GB GDDR5 · 192 GB/s bandwidth · DeepSeek · 861.6B (active 49B) · MoE · model ctx up to 1000K

CUDA

No — DeepSeek-V4-Pro at Q4 needs ≈ 475.49 GB even at 4K context, beyond the Tesla P4 8GB's 8 GB GDDR5. It would only run with heavy CPU/disk offload. FP16 needs ≈ 1725.42 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	475.49 GB	No	—
Q4	8K default	476.11 GB	No	—
Q4	32K	479.78 GB	No	—
FP16	4K	1724.81 GB	No	—
FP16	8K default	1725.42 GB	No	—
FP16	32K	1729.1 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	Tesla P4 8GB VRAM
Q4	473.88 GB	1.23 GB	1 GB	476.11 GB	8 GB
FP16	1723.2 GB	1.23 GB	1 GB	1725.42 GB	8 GB

Other GPUs that run DeepSeek-V4-Pro

• DGX Spark (GH200 480GB) — ≈6.5 tok/s Q4 est.

All GPUs for DeepSeek-V4-Pro →

Other models for the Tesla P4 8GB

• MiniCPM5-1B — ≈113.1 tok/s Q4 est.
• Qwen3.5-2B — ≈61.1 tok/s Q4 est.
• DeepSeek-R1-0528-Qwen3-8B-layer-mix-bpw-3.8-mlx — ≈53.1 tok/s Q4 est.
• Llama-3.2-3B-Instruct — ≈38.2 tok/s Q4 est.
• Qwen3-4B-Instruct-2507 — ≈30.5 tok/s Q4 est.
• NVIDIA-Nemotron-3-Nano-4B-BF16 — ≈30.5 tok/s Q4 est.
• Qwen3-4B-Thinking-2507 — ≈30.5 tok/s Q4 est.
• gemma-4-E2B-it — ≈24.0 tok/s Q4 est.

All models on the Tesla P4 8GB →