Quadro P2000 holds 113 of the models in our catalogue and is, in practice, a Q4_K_M card — the largest it takes is ChatGLM2 6B at Q4_K_M.
Neighbours in memory rather than in price: memory decides whether a card can do the job at all, so two cards of the same size at different prices is the comparison you are making. Ordered by memory, then bandwidth — not by the value column, which is worked out from bandwidth and price alone and therefore rewards a cheap card whatever its software stack does to that bandwidth in practice. Read it as one input, not as a ranking.
NVIDIA Quadro P2000 — 5 GB VRAM.
- Brand
- NVIDIA
- VRAM
- 5 GB GDDR5
- Bandwidth
- 140.2 GB/s
- FP16 Compute
- 6 TFLOPS
- FP32 Compute
- 3 TFLOPS
- CUDA Cores
- 1,024
- TDP
- 75 W
- Architecture
- Pascal
With 5 GB VRAM and 140.2 GB/s bandwidth, this GPU handles models up to 5.06B parameters.
Speed ≈ bandwidth / model_size × efficiency. A 7B model at Q4 runs at ~16 tok/s.
| Model | Size | VRAM Q4 | tok/s | Avg |
|---|---|---|---|---|
| K2 Horizon 3.7B | 5.06B | 3.6 GB | 22 | 37.3 |
| InternLM2 5B | 4.5B | 3.2 GB | 25 | 47.6 |
| Qwen3-VL 4B Instruct | 4.44B | 3.2 GB | 25 | 26.6 |
| MedGemma 1.5 4B | 4.3B | 3.1 GB | 26 | 5.9 |
| Gemma 3 4B | 4.3B | 3.1 GB | 26 | 109.7 |
| Ministral 3 3B Reasoning | 4.25B | 3.1 GB | 26 | 60.1 |
| Qwen 1.5 4B | 4B | 2.9 GB | 28 | 12.6 |
| Qwen3 4B | 4B | 2.9 GB | 28 | 40.7 |
| Qwen3-4B Instruct 2507 | 4B | 2.9 GB | 28 | 37.2 |
| Qwen3-Embedding 4B | 4B | 2.9 GB | 28 | 68.5 |
| Granite Vision 4.1 4B | 4B | 2.9 GB | 28 | — |
| Nemotron 3 Nano 4B | 3.97B | 2.9 GB | 28 | 60.1 |
| Ministral 3 3B | 3.85B | 2.8 GB | 29 | 18.3 |
| Phi-3.5 Mini 3.8B | 3.82B | 2.8 GB | 29 | 46.6 |
| phi-3-mini-4k 3.8B | 3.8B | 2.8 GB | 30 | 30.5 |
| Phi-4-mini 3.8B | 3.8B | 2.8 GB | 30 | 49.0 |
| Qwen2.5-VL-3B | 3.8B | 2.8 GB | 30 | 30.1 |
| Jina Embeddings v4 | 3.8B | 2.8 GB | 30 | 56.2 |
| Granite 4.2 3B | 3.66B | 2.7 GB | 31 | 38.0 |
| Cogito 3B | 3.61B | 2.7 GB | 31 | 22.1 |