Skip to content
Run LLM Locally

GPU · NVIDIA

Tesla P40 for local LLMs: which models run?

With 24 GB of VRAM and 346 GB/s of bandwidth, the Tesla P40 runs models up to about Qwen3 32B entirely on the GPU.

  • VRAM24 GB GDDR5
  • Bandwidth346 GB/s
  • Board power250 W
  • ArchitecturePascal
ShareFacebookXWhatsAppLinkedInEmail

Specifications

VRAM
24 GB
Memory type
GDDR5
Memory bus
384-bit
Bandwidth
346 GB/s
Board power
250 W
Backends
CUDA
Interface
PCIe 3.0 x16
Released
September 2016
Segment
Data center
FP16 compute
Not available

Official specData updated 24 days agoSource: NVIDIA specifications

What the Tesla P40 runs

Popular models on Tesla P40 + 64 GB RAM · 8K context
Qwen3 30B-A3BQ4_K_M · 8K~19.7 GB · 66–98 tok/s estimateExcellent
Qwen3 Coder 30B-A3B InstructQ4_K_M · 8K~19.7 GB · 66–98 tok/s estimateExcellent
gpt-oss-20bMXFP4 · 8K~13 GB · 44–120 tok/s estimateExcellent
Qwen3 14BQ8_0 · 8K~17.5 GB · 16–21 tok/s estimateExcellent
DeepSeek R1 Distill Qwen 14BQ8_0 · 8K~17.8 GB · 16–20 tok/s estimateExcellent
Qwen2.5 14B InstructQ8_0 · 8K~17.8 GB · 16–20 tok/s estimateExcellent
Phi-4 14BQ8_0 · 8K~17.6 GB · 16–20 tok/s estimateExcellent
Mistral Nemo 12B InstructQ8_0 · 8K~14.9 GB · 19–24 tok/s estimateExcellent
Gemma 3 12B ITQ8_0 · 8K~14.1 GB · 20–26 tok/s estimateExcellent
Gemma 2 9B ITQ8_0 · 8K~12.7 GB · 23–31 tok/s estimateExcellent
Qwen3 8BQ8_0 · 8K~10.5 GB · 27–36 tok/s estimateExcellent
Llama 3.1 8B InstructQ8_0 · 8K~10.2 GB · 28–37 tok/s estimateExcellent
DeepSeek R1 Distill Llama 8BQ8_0 · 8K~10.2 GB · 28–37 tok/s estimateExcellent
Qwen2.5 7B InstructQ8_0 · 8K~9.2 GB · 30–40 tok/s estimateExcellent
Mistral 7B Instruct v0.3Q8_0 · 8K~9.2 GB · 31–40 tok/s estimateExcellent
Gemma 3 4B ITQ8_0 · 8K~5.4 GB · 55–76 tok/s estimateExcellent
Qwen3 4BQ8_0 · 8K~6.2 GB · 49–67 tok/s estimateExcellent
Phi-4 Mini Instruct 3.8BQ8_0 · 8K~6 GB · 52–71 tok/s estimateExcellent
Llama 3.2 3B InstructQ8_0 · 8K~5.1 GB · 62–84 tok/s estimateExcellent
Qwen3 1.7BQ8_0 · 8K~3.7 GB · 97–137 tok/s estimateExcellent
Llama 3.2 1B InstructQ8_0 · 8K~2.5 GB · 156–218 tok/s estimateExcellent
Qwen3 0.6BQ8_0 · 8K~2.5 GB · 172–263 tok/s estimateExcellent
Mixtral 8x7B InstructQ3_K_M · 8K~21.6 GB · 33–44 tok/s estimateCompatible
Qwen3 32BQ4_K_M · 8K~22.3 GB · 12–16 tok/s estimateCompatible
DeepSeek R1 Distill Qwen 32BQ4_K_M · 8K~22.4 GB · 12–16 tok/s estimateCompatible
Qwen2.5 32B InstructQ4_K_M · 8K~22.4 GB · 12–16 tok/s estimateCompatible
Qwen2.5 Coder 32B InstructQ4_K_M · 8K~22.4 GB · 12–16 tok/s estimateCompatible
QwQ 32BQ4_K_M · 8K~22.4 GB · 12–16 tok/s estimateCompatible
Gemma 3 27B ITQ5_K_M · 8K~21 GB · 13–17 tok/s estimateCompatible
Gemma 2 27B ITQ5_K_M · 8K~22.3 GB · 13–17 tok/s estimateCompatible
Mistral Small 3.2 24B InstructQ6_K · 8K~21 GB · 13–17 tok/s estimateCompatible
gpt-oss-120bMXFP4 · 8K~63.2 GB · 9.5–19 tok/s estimatePossible but slow
Llama 4 Scout 17B-16E InstructQ3_K_M · 8K~55.7 GB · 3.5–6.5 tok/s estimatePossible but slow
Qwen2.5 72B InstructQ3_K_M · 8K~38.2 GB · 1.2–2.5 tok/s estimatePossible but slow
Llama 3.3 70B InstructQ3_K_M · 8K~37 GB · 1.3–2.5 tok/s estimatePossible but slow
DeepSeek R1 Distill Llama 70BQ3_K_M · 8K~37 GB · 1.3–2.5 tok/s estimatePossible but slow
DeepSeek R1 (671B)Q8_0 · 8K~699 GB · No dataIncompatible
Qwen3 235B-A22BQ8_0 · 8K~247 GB · No dataIncompatible

For each model, the highest-precision quantization that runs entirely in fast memory; otherwise the one with the best result. Speeds are ranges; "estimate" unless a benchmark exists for this exact hardware.

EstimateData updated 24 days agoSource: Run LLM Locally engine

Limits in plain words

24 GB of VRAM holds the weights and KV cache of models up to about Qwen3 32B at Q4_K_M. Bigger models spill into system RAM and slow down sharply: generation speed is then limited by RAM bandwidth, not by the GPU.

Frequently asked questions

Which LLMs can the Tesla P40 run?
Entirely on the GPU: Mixtral 8x7B Instruct (Q3_K_M), Qwen3 32B (Q4_K_M), DeepSeek R1 Distill Qwen 32B (Q4_K_M), Qwen2.5 32B Instruct (Q4_K_M), Qwen2.5 Coder 32B Instruct (Q4_K_M), QwQ 32B (Q4_K_M), Qwen3 30B-A3B (Q4_K_M), Qwen3 Coder 30B-A3B Instruct (Q4_K_M). Larger models need CPU/GPU offload.
Can the Tesla P40 run a 70B model?
Not recommended: a 70B model at Q4_K_M needs about 45.1 GB; the Tesla P40 has 24 GB.
How is the speed estimated?
Token generation reads every active weight once per token, so it is limited by memory bandwidth (346 GB/s here). We apply an efficiency range per backend and label the result as an estimate until real benchmarks exist.