Tesla P40 for local LLMs — what fits in 24 GB & the real cost
Rent a Tesla P40 from $0.13/hr ($0 markup, as of 2026-07-02). 323 of the 445 models in our catalog fit its 24 GB — plus the honest own-vs-rent-vs-API math, electricity and amortization included.
Three honest ways to run AI on a Tesla P40
Own it: buy the card and pay electricity + amortization — local is real money, not "$0".
Rent it: from $0.13/GPU-hr on your own vendor account (as of 2026-07-02, $0 markup — we never resell compute).
Skip it: run a comparable open model through your own API key, paying per million tokens (prices as of 2026-07-06).
Rent a Tesla P40 — real vendor rates, $0 markup
Vast.ai: $0.13/GPU-hr — vendor's own price as of 2026-07-02; you pay them directly.
What fits in 24 GB
323 of the 445 models in the Spanvero catalog fit on one Tesla P40 at a sensible quant (context capped at 16k for the estimate). Most capable first:
Laguna XS.2 (poolside, 33.4B) — ~24 GB at Q4_K_M; own-key API alternative ≈ $0.15/1M tokens.
sarvam 30b (sarvamai, 32.2B) — ~22 GB at Q4_K_M; own-key API alternative ≈ $0.36/1M tokens (size estimate).
llm jp 4 32b a3b thinking (llm-jp, 32.1B) — ~23 GB at Q4_K_M; own-key API alternative ≈ $0.36/1M tokens (size estimate).
NVIDIA Nemotron 3 Nano 30B A3B BF16 (nvidia, 31.6B) — ~22 GB at Q4_K_M; own-key API alternative ≈ $0.35/1M tokens (size estimate).
Nemotron Cascade 2 30B A3B (nvidia, 31.6B) — ~22 GB at Q4_K_M; own-key API alternative ≈ $0.35/1M tokens (size estimate).
Qwen3 30B A3B (Qwen, 30.5B) — ~22 GB at Q4_K_M; own-key API alternative ≈ $0.31/1M tokens.
Qwen3 Coder 30B A3B Instruct (Qwen, 30.5B) — ~22 GB at Q4_K_M; own-key API alternative ≈ $0.17/1M tokens.
Qwen3 30B A3B Instruct 2507 (Qwen, 30.5B) — ~22 GB at Q4_K_M; own-key API alternative ≈ $0.12/1M tokens.
Qwen3 30B A3B Thinking 2507 (Qwen, 30.5B) — ~22 GB at Q4_K_M; own-key API alternative ≈ $0.85/1M tokens.
Tongyi DeepResearch 30B A3B (Alibaba-NLP, 30.5B) — ~22 GB at Q4_K_M; own-key API alternative ≈ $0.34/1M tokens (size estimate).
Qwen3 30B A3B Base (Qwen, 30.5B) — ~22 GB at Q4_K_M; own-key API alternative ≈ $0.34/1M tokens (size estimate).
lynx instruct 30b (bineric, 30.5B) — ~22 GB at Q4_K_M; own-key API alternative ≈ $0.34/1M tokens (size estimate).
North Mini Code 1.0 (CohereLabs, 30.5B) — ~22 GB at Q4_K_M; own-key API alternative ≈ $0.34/1M tokens (size estimate).
granite 4.1 30b (ibm-granite, 28.9B) — ~24 GB at Q4_K_M; own-key API alternative ≈ $0.33/1M tokens (size estimate).
Gemma 2 27B Instruct (Google, 27B) — ~22 GB at Q4_K_M; own-key API alternative ≈ $0.65/1M tokens.
Used NVIDIA Tesla P40 24GB (used market): street price $320.00 (as of 2026-06-30; GPUDojo used-price roundup + eBay listings, Jun 2026 (avg ~$404, typical working-pull listings $300–$320; passive server card — needs ducted cooling, no FP16 fast path)) → $0.2922/day amortized over 3 years.
Electricity ≈ $0.7265/1M tokens at 250 W and $0.19/kWh (EIA Electric Power Monthly Table 5.6.A — U.S. residential average, Apr 2026, as of 2026-07-02), assuming ~18 tok/s (8B-class model at Q4).
vs renting: owning wins above ~3.5 active hours/day at the cheapest logged rate ($0.13/hr).
vs API: Llama 3.1 8B Instruct at $0.03/1M (as of 2026-07-06) is cheaper than the electricity alone — owning never breaks even at any volume.
As an Amazon Associate we earn from qualifying purchases. Each link opens an Amazon search for the category — not an endorsement of any specific product — and Cynosure LLC may earn a commission at no extra cost to you.
A short email of real AI price moves, straight from the daily log — no hype. We're collecting the list now; the first issue goes out when it opens. Unsubscribe with one click.