NVIDIA Quadro P5000 16GB - Gaming / CUDA / Local AI GPU

NVIDIA Quadro P5000 professional workstation GPU with 16GB GDDR5X VRAM. 2 for sale! Condition: USED. GPU only. See photos for exact cosmetic condition. The P5000 is still an interesting budget card because you get 16GB VRAM, Pascal CUDA support and roughly GTX 1080-class raw compute in a professional blower-style GPU. SPECS • 16GB GDDR5X VRAM • 2,560 CUDA cores • ~8.9 TFLOPS FP32 • 288 GB/s memory bandwidth • PCIe 3.0 x16 • 4x DisplayPort 1.4 + DVI-D • 180W • CUDA Compute Capability 6.1 AI / LOCAL LLM 16GB makes this a nice middle ground for local AI. Great for: • llama.cpp / GGUF • Ollama with compatible CUDA builds • Local coding/chat models • 7B / 8B models with plenty of VRAM • 12B / 14B quantized models • Some larger quantized and MoE models depending on architecture • Partial GPU offloading of models larger than VRAM • Whisper • Embeddings • Stable Diffusion • CUDA development Community llama.cpp testing has shown a Quadro P5000 reaching around 36 tok/s on a 7B Q4 model, so these Pascal cards can still provide very usable local-LLM performance. Actual tokens/sec varies dramatically by model, quantization, context, backend and settings. No Tensor cores, so this isn’t meant to beat a newer RTX GPU. The appeal is getting 16GB of CUDA-capable VRAM relatively cheaply. GAMING Performance is around the GTX 1080 era depending on the title and drivers. Typical expectations: • 1080p esports: 100-200+ FPS possible • 1080p High: roughly 60-100 FPS • 1440p High: roughly 45-75 FPS • 4K Medium: roughly 30-50 FPS in appropriate titles No hardware ray tracing or DLSS. This is probably the best all-around card of the group if you want gaming + 16GB VRAM + local AI capability. $325 OBO each

Condition

Used (normal wear)

Type

Graphic/Video card

Brand

Nvidia