GPU MODELS / L20
L20 cloud GPU pricing
42 listings across 3 providers. Cheapest published rate: $0.575/GPU/hr at AutoDL. Sorted by per-GPU hourly price. Every row links to the provider's own published price.
WHAT IS THE L20
The L20 is the China-market companion to the H20: a 48 GB Ada datacenter card with the L40S's memory system (48 GB, 864 GB/s) but export-compliant compute, sold only in China, Hong Kong, and Macau. Where the H20 covers Hopper-class inference, the L20 is the cheaper Ada-class workhorse - Alibaba and other China clouds list it as the default mid-tier inference instance.
WHAT IT IS GOOD AT
WHO SHOULD NOT RENT THE L20
| Provider | Config | VRAM | Location | $/hr | $/GPU/hr | Source |
|---|---|---|---|---|---|---|
 AutoDL | L20 × 1 | 48 GB | China | $0.575 | $0.575 | receipt |
| L20 × 1 | 48 GB | China (Shanghai) | $2.230 | $2.230 | receipt | |
| L20 × 1 | 48 GB | China (Hangzhou) | $2.230 | $2.230 | receipt | |
| L20 × 1 | 48 GB | China (Shenzhen) | $2.230 | $2.230 | receipt | |
| L20 × 1 | 48 GB | China (Beijing) | $2.230 | $2.230 | receipt | |
| L20 × 8 | 48 GB | China (Shanghai) | $17.849 | $2.231 | receipt | |
| L20 × 8 | 48 GB | China (Hangzhou) | $17.849 | $2.231 | receipt | |
| L20 × 8 | 48 GB | China (Shenzhen) | $17.849 | $2.231 | receipt | |
| L20 × 8 | 48 GB | China (Beijing) | $17.849 | $2.231 | receipt | |
| L20 × 8 | 48 GB | US East (N. Virginia) | $17.979 | $2.247 | receipt | |
| L20 × 1 | 48 GB | US East (N. Virginia) | $2.257 | $2.257 | receipt | |
| L20 × 1 | 48 GB | China | $2.336 | $2.336 | receipt | |
| L20 × 1 | 48 GB | China (Shanghai) | $2.450 | $2.450 | receipt | |
| L20 × 1 | 48 GB | China (Hangzhou) | $2.450 | $2.450 | receipt | |
| L20 × 1 | 48 GB | China (Shenzhen) | $2.450 | $2.450 | receipt | |
| L20 × 1 | 48 GB | China (Beijing) | $2.450 | $2.450 | receipt | |
| L20 × 1 | 48 GB | US East (N. Virginia) | $2.472 | $2.472 | receipt | |
| L20 × 1 | 48 GB | US West (Silicon Valley) | $2.527 | $2.527 | receipt | |
| L20 × 8 | 48 GB | US West (Silicon Valley) | $20.236 | $2.530 | receipt | |
| L20 × 8 | 48 GB | Thailand (Bangkok) | $20.358 | $2.545 | receipt | |
| L20 × 1 | 48 GB | Thailand (Bangkok) | $2.545 | $2.545 | receipt | |
| L20 × 8 | 48 GB | South Korea (Seoul) | $20.507 | $2.563 | receipt | |
| L20 × 8 | 48 GB | Malaysia (Kuala Lumpur) | $21.001 | $2.625 | receipt | |
| L20 × 1 | 48 GB | Malaysia (Kuala Lumpur) | $2.630 | $2.630 | receipt | |
| L20 × 1 | 48 GB | Hong Kong | $2.651 | $2.651 | receipt | |
| L20 × 8 | 48 GB | Hong Kong | $21.284 | $2.660 | receipt | |
| L20 × 8 | 48 GB | Singapore | $21.310 | $2.664 | receipt | |
| L20 × 1 | 48 GB | Singapore | $2.672 | $2.672 | receipt | |
| L20 × 1 | 48 GB | US West (Silicon Valley) | $2.783 | $2.783 | receipt | |
| L20 × 1 | 48 GB | Thailand (Bangkok) | $2.800 | $2.800 | receipt | |
| L20 × 1 | 48 GB | Malaysia (Kuala Lumpur) | $2.888 | $2.888 | receipt | |
| L20 × 1 | 48 GB | Hong Kong | $2.927 | $2.927 | receipt | |
| L20 × 1 | 48 GB | Singapore | $2.930 | $2.930 | receipt | |
| L20 × 8 | 48 GB | Indonesia (Jakarta) | $23.975 | $2.997 | receipt | |
| L20 × 1 | 48 GB | Indonesia (Jakarta) | $3.003 | $3.003 | receipt | |
| L20 × 8 | 48 GB | Japan (Tokyo) | $24.115 | $3.014 | receipt | |
| L20 × 1 | 48 GB | Japan (Tokyo) | $3.024 | $3.024 | receipt | |
| L20 × 8 | 48 GB | Germany (Frankfurt) | $25.268 | $3.159 | receipt | |
| L20 × 1 | 48 GB | Germany (Frankfurt) | $3.169 | $3.169 | receipt | |
| L20 × 1 | 48 GB | Indonesia (Jakarta) | $3.297 | $3.297 | receipt | |
| L20 × 1 | 48 GB | Japan (Tokyo) | $3.316 | $3.316 | receipt | |
| L20 × 1 | 48 GB | Germany (Frankfurt) | $3.475 | $3.475 | receipt |
THE L20 MARKET, BY THE NUMBERS
3 providers publish 42 L20 configurations on the panel today. Published per-GPU hourly rates run from $0.575 (AutoDL) to $3.47 (Alibaba Cloud), a 6x spread between the cheapest and the most expensive published rate for the same chip. The median listing sits at $2.56/GPU/hr. Rows are published catalog rates; stock flags appear per row where the provider reports them. All rates are on-demand - committed-use and reserved pricing is excluded.
Across every L20 row on the panel, the daily floor held flat at $0.575/GPU/hr over 8 days of tracking. The median continuously-tracked listing held flat at $2.56. Snapshots run daily; the full history is public in the repo.
At the current floor, one L20 running around the clock costs about $420 a month (730 hours of on-demand arithmetic). Node shape matters: the single-GPU floor is $0.575/GPU/hr against $2.231/GPU/hr on an 8-GPU node (+288.0%). Per-GPU floors by config size: 1x $0.58 · 8x $2.23. Compare the alternatives below before committing - the same budget often buys more than one chip class.
L20 VS THE ALTERNATIVES
Compare live rates: L20 vs L40S · L20 vs H20
L20 PRICING FAQ
See also alternatives to the L20 · cheapest GPU-hour per model · inference chips · training chips · by location