Home › GPUs & Macs › Rent or buy
Rent or buy a GPU for local LLMs
RunPod's hourly price for a card against what the card cost at launch: how many rented hours it takes to spend the price of the card. RunPod prices fetched from runpod.io/pricing on 2026-10-22.
Break-even hours
| Card | Launch MSRP | Community $/h | Break-even h | Years at 4 h/day | Secure $/h | Break-even h | Years at 4 h/day | Buy |
|---|---|---|---|---|---|---|---|---|
| NVIDIA RTX 5090 | $1,999 | $0.69 | 2,897 | 2.0 | $1.19 | 1,680 | 1.2 | Amazon |
| NVIDIA RTX 3090 | $1,499 | $0.22 | 6,814 | 4.7 | $0.50 | 2,998 | 2.1 | Amazon |
| NVIDIA RTX 4090 | $1,599 | $0.34 | 4,703 | 3.2 | $0.89 | 1,797 | 1.2 | Amazon |
break-even hours = launch MSRP / RunPod price per hour years at 4 h/day = break-even hours / (4 × 365)
Left out on purpose: electricity, the rest of the PC, resale value of a card you own, and RunPod storage and network charges. Buying gets cheaper if you would resell the card or already have the PC; renting gets cheaper if you use it a few hours a month. MSRP is the launch price from the dated Wikipedia revisions on the value page, not today's used price (used prices pending). Only cards that RunPod rents and that we have a launch price for are in this table.
Rent a GPU on RunPodAffiliate link: if you sign up, RigBench may earn a referral credit. It never affects rankings or numbers.
Every GPU on RunPod's price list
Speed is the fastest measured Llama 2 7B Q4_0 generation run we have for that card (same test as the value page). Cost per million tokens assumes one request at a time at that speed on the cheaper of the two prices; batching many requests costs far less per token.
| GPU | VRAM GB | Community $/h | Secure $/h | tok/s | $ per 1M tokens |
|---|---|---|---|---|---|
| B300 | – | $6.94 | $8.99 | – | – |
| H200 | 141 | $3.59 | $5.29 | – | – |
| B200 | 180 | $5.98 | $7.99 | – | – |
| RTX Pro 6000 | 96 | $1.69 | $2.49 | – | – |
| H100 NVL | 94 | $2.59 | $3.19 | – | – |
| H100 PCIe | 80 | $1.99 | $2.89 | – | – |
| H100 SXM | 80 | $2.69 | $3.99 | 281 measured | $2.66 |
| A100 PCIe | 80 | $1.19 | $1.79 | 201 measured | $1.65 |
| A100 SXM | 80 | $1.39 | $1.79 | – | – |
| Pro 6000 MIG 48GB | 48 | $1.00 | $1.29 | – | – |
| L40S | 48 | $0.79 | $1.09 | – | – |
| RTX 6000 Ada | 48 | $0.74 | $0.99 | 179 measured | $1.15 |
| A40 | 48 | $0.35 | $0.59 | 126 measured | $0.77 |
| L40 | 48 | $0.69 | $0.82 | 154 measured | $1.25 |
| RTX A6000 | 48 | $0.33 | $0.59 | 145 measured | $0.63 |
| RTX 5090 | 32 | $0.69 | $1.19 | 300 measured | $0.64 |
| Pro 6000 MIG 24GB | 24 | $0.50 | $0.69 | – | – |
| L4 | 24 | $0.44 | $0.59 | – | – |
| RTX 3090 | 24 | $0.22 | $0.50 | 172 measured | $0.36 |
| RTX 4090 | 24 | $0.34 | $0.89 | 190 measured | $0.50 |
| RTX A5000 | 24 | $0.16 | $0.27 | 140 measured | $0.32 |
$ per 1M tokens = price per hour / (measured tok/s × 3600) × 1,000,000
"–" means RunPod shows no price for that tier, or we have no measured Llama 2 7B Q4_0 run on that GPU. Prices are on-demand pod prices as shown on the page at 2026-10-22T22:51:18+00:00 (UTC) and change often; check RunPod before you decide.
As an Amazon Associate I earn from qualifying purchases. Affiliate policy.