GPU cloud / L4
L4 cloud pricing
24 GB VRAM per GPU. Compare 3 sourced offers. Use measured tokens/second from your own model and setup to calculate per-million-token costs.
| GPU | Provider | VRAM | GPU count | USD/hour | $/GB/hr | Billing | Type | Verified | Source |
|---|---|---|---|---|---|---|---|---|---|
| L4 | runpod | 24 GB | 1 | $0.49 | $0.020 | per-second | on-demand | 2026-09-26 | Official price ↗ |
| L4 | modal | 24 GB | 1 | $0.80GPU only | $0.033 | per-second | on-demand | 2026-09-26 | Official price ↗ |
| L4 | ovhcloud | 24 GB | 1 | $1.00 | $0.042 | hourly | on-demand | 2026-09-26 | Official price ↗ |
What fits in VRAM? (estimate)
4-bit weights with 20% headroom; real KV cache, context length and framework overhead vary. Multi-GPU memory is not automatically pooled.
- 7B: ~4.2 GB — fits by this estimate
- 13B: ~7.8 GB — fits by this estimate
- 30B: ~18.0 GB — fits by this estimate
- 70B: ~42.0 GB — exceeds single-card VRAM
Adjust precision and headroom →
Inference throughput and cost per 1M tokens
No comparable public throughput benchmark has been verified for this card. Enter your own measured tokens/second; the site does not assume a speed.
Calculate cost per million tokens →
GPU comparisons
Prices can change. Each offer links its official source and verification date. Affiliate links, if added later, will be marked sponsored.