NVIDIA · Ada Lovelace · 2023
NVIDIA L4
72 watts. The efficiency play for video, small-model inference, and high-density serving.
24 GB
GDDR6
0.3 TB/s
Mem bandwidth
121 TF
FP16 dense
242 TF
FP8 dense
72 W
TDP
$0.32 /hr
From · Vast.ai
Single-slot, no external power connector, clouds deploy it densely and price it low. Think of it as the successor to the T4.
Where to rent a L4
On-demand prices per GPU-hour; marketplace rates refresh daily, list prices reviewed 2026-09. Reserved and spot run 30-70% lower. Full breakdown with the used market and rent-vs-buy math: L4 pricing.
| Provider | $/GPU-hr | Notes |
|---|---|---|
| Vast.ai | $0.32 | Verified-listing rate, refreshed daily |
| RunPod | $0.49 | neocloud |
| GCP | $0.71 | g2-standard-4 |
| OVHcloud | $0.75 | hyperscaler |
| Modal | $0.80 | Serverless, per-second billing |
| AWS | $0.98 | g6.xlarge |
Specifications in context
Bars scaled against the best value in the whole catalog (B200 / MI325X era).
| Architecture | Ada Lovelace (2023) |
|---|---|
| Memory | 24 GB GDDR6, 0.3 TB/s |
| Interconnect | PCIe Gen4 |
| Form factor | PCIe (low-profile) |
| Partitioning | No MIG (time-slicing / vGPU only) |
What fits on one L4
Weights + ~2 GB runtime overhead against 22 GB usable VRAM. Longer contexts and bigger batches need more, check the VRAM calculator.
| Model | Params | Highest precision that fits |
|---|---|---|
| Llama 3.2 1B | 1.24B | FP16 |
| Llama 3.2 3B | 3.21B | FP16 |
| Llama 3.1 8B | 8.03B | FP16 |
| Qwen2.5 7B | 7.62B | FP16 |
| Qwen2.5 14B | 14.8B | 8-bit |
| Mistral 7B | 7.25B | FP16 |
| Gemma 2 9B | 9.24B | 8-bit |
| Gemma 2 27B | 27.2B | 4-bit |
| Phi-4 14B | 14.7B | 8-bit |
| gpt-oss-20b | 20.9B | 4-bit |
Best for
- 7B-class quantized models
- Whisper / embedding / rerank serving
- Video transcode + AI pipelines