Tesla P40 Price & Used Prices: Which Local LLMs Can It Run?

A single Tesla P40 has 24GB of GDDR5 VRAM. At 4-bit quantization and 8K context it fits a dense model of up to about 37B parameters, or about 20B at 8-bit.

Not on a single card. At 4-bit with an 8K context a 70B dense model needs about 43GB, more than the Tesla P40's 24GB. It runs if you split the layers across 2 cards with llama.cpp (48GB total).

Among models released in 2026, the largest dense model that fits entirely in VRAM at 4-bit is Gemma 4 31B, at an estimated 13.3 tokens/s. Among MoE models, Qwen3.6 35B A3B is the fastest at about 86.6 tokens/s. The full list is in the table above.

Estimated from memory bandwidth at 4-bit and 8K context: 8B dense: about 46.2 tokens/s, 14B dense: about 27.4 tokens/s, and 32B dense: about 12.5 tokens/s. MoE models read only the active parameters for each token, so they run much faster. Real-world results are usually 70%–100% of the estimate.

With vLLM tensor parallelism at 4-bit and 32K context, two Tesla P40 cards cannot fit any model of 50B or larger released in 2026. At 4-bit with 32K context the minimum is 8 cards, which fits Qwen3.8 Flash Next and DeepSeek V4 Flash.

As of Sep 29, 2026: eBay median $397 (16 listings) and Xianyu median $179 (13 listings). Figures are medians of live listings, refreshed every 6 hours.

As of Sep 29, 2026: Compshare at $0.06 per hour. All are single-GPU on-demand rates, excluding storage and data transfer.

Taking the eBay median of $397 and electricity at $0.18 per kWh, at 8 hours a day buying pays for itself in about 11.8 years (versus Compshare at $0.06 per hour). If you will use it for longer than 11.8 years, buy; otherwise, rent.

The Arc Pro B60 has 24GB of VRAM and 456 GB/s of bandwidth; the Tesla P40 has 24GB and 347 GB/s. For 32B dense models at 4-bit, the Tesla P40 is estimated at about 12.5 tokens/s and the Arc Pro B60 at about 10.6 tokens/s.