Home › Budget builds › under $1500
Local LLM build under $1500 — what actually fits
The 24GB-class question: used 4090 vs renting an A100.
What $1500 maps to in hardware
Typical accelerators in this band: RTX 3060 12GB, RTX 4060 Ti 16GB, RTX 5080 16GB, RTX 3090 24GB, RTX 4090 24GB (launch-MSRP anchors, all Wikipedia, retrieved 2026-10-20: RTX 3090 $1,499 (launched 2020-09-24) · RTX 4090 $1,599 · RTX 5090 $1,999; MSRP ≠ today’s used price — that is why the box above is empty). Street prices change weekly: confirm on eBay/Amazon yourself; we link, never guess.
Models that fit this class at q4_k_m (8k context, GiB)
| model | need | smallest fit |
|---|---|---|
| Ornith 1.5 35B-A3B (MoE) | 22.9 | RTX 3090 24GB |
| Qwen3.6 35B-A3B (MoE) | 22.9 | RTX 3090 24GB |
| Nemotron 3 Nano 30B-A3B (MoE) | 20.1 | RTX 3090 24GB |
| Gemma 4 31B | 21.9 | RTX 3090 24GB |
| GLM-4.7 Flash | 20.3 | RTX 3090 24GB |
| Xing 4.0 29B-A4B (MoE) | 20.2 | RTX 3090 24GB |
| Qwen3 30B-A3B (MoE) | 20.2 | RTX 3090 24GB |
| Muse Glimmer 30B | 19.2 | RTX 3090 24GB |
| Qwen3.6 27B | 18.3 | RTX 3090 24GB |
| Qwen3.8 27B | 18.3 | RTX 3090 24GB |
| Gemma 4 26B-A4B (MoE) | 17.0 | RTX 3090 24GB |
| gpt-oss-20b | 13.7 | RTX 4060 Ti 16GB |
| Gemma 4 12B | 8.6 | RTX 3060 12GB |
| Ornith 1.5 9B | 6.8 | RTX 3060 12GB |
| Qwen3.5 9B | 6.8 | RTX 3060 12GB |
Or skip hardware: what $1500 of cloud time buys
| instance | $/h | hours for $1500 |
|---|---|---|
| RunPod RTX 5090 32GB | $0.99 | 1515 |
| RunPod A100 80GB | $1.59 | 943 |
| RunPod H100 SXM | $3.49 | 429 |
modelvram.com public table (self-stated check 2026-09-29, retrieved 2026-10-20). Partial re-verify 2026-10-20: Lambda A100 40GB $1.99/h CONFIRMED on lambda.ai/pricing (fetched directly); Lambda H100 rows ambiguous on-page ($3.99 vs $4.09 configs) → not adopted; RunPod & Vast pricing pages are JS-rendered, no extractable static figures — verify rates visually before spending.
FAQ
What can I run on a local LLM build under $1500?
Roughly the 12–24GB VRAM class. That comfortably runs most 7–32B models at q4_K_M–q6_K with 8k context; see the fitted list above. Buy prices are user-entered — we don't guess market prices.
Would $1500 be better spent renting cloud GPUs?
$1500 buys about 1515 hours of RTX 5090 (RunPod, $0.99/h) or 943 hours of A100 80GB (RunPod, $1.59/h) at the competitor-published rates we cite (check date 2026-09-29, retrieved 2026-10-20). If you need the model more than ~a few hundred hours per month, buying tends to win — use the break-even calculator.
Is a used GPU a good bet?
See used GPU break-even: MSRP anchors are sourced; street prices must be entered by you.