NVIDIA L4 rental: early access
Ada Lovelace · PCIe (being confirmed)
We’re collecting early-access requests. We’re placing our capacity in Uzbekistan. We don’t publish prices yet. We’ll confirm availability in our reply.
L4
Request L4We’ll reply with terms for your workload. Nothing to pay before the contract.
Estimate usage- Memory per GPU
- 24 GB GDDR6
- Memory bandwidth
- 0.3 TB/s
- GPU interconnect
- PCIe 4, no NVLink
NVIDIA L4 specifications
- Architecture
- Ada Lovelace
- Announced
- 2023
- Memory
- 24 GB
- Memory type
- GDDR6
- Memory bandwidth
- 0.3 TB/s
- FP32
- 30.3 TFLOPS
- FP16/BF16, dense
- 121 TFLOPS
- FP8, dense
- 242 TFLOPS
- FP4, dense
- none
- FP64
- not in NVIDIA’s card specs
- RT cores
- yes
- NVENC/NVDEC
- NVENC 2, NVDEC 4
- MIG
- none
- GPU interconnect
- PCIe 4, no NVLink
- Power (TDP/TGP)
- 72 W
- Form factor
- PCIe (being confirmed)
Memory is specified per GPU. Multiple GPUs have separate memory: using a larger model across them requires compatible software and model partitioning. NVLink does not make them one GPU.
The L4 form factor is still being confirmed. Specs from NVIDIA’s official materials, checked 26 September 2026.
The L4 by workload
- Inference: LLMs up to 14B in FP8 on one GPU
- Fine-tuning: QLoRA up to 14B, LoRA up to 8B on one GPU
- Training from scratch: models up to 1B on one GPU
- Rendering: scenes up to 16 GB
- FP64 computing: not suited to this GPU
An estimate from open formulas, not a benchmark or an invoice.
Servers with the NVIDIA L4
Server configuration (CPU, RAM, disk, network) is being confirmed.
Sizing assumption: up to 8 GPUs per server. We’ll confirm the actual configuration in our reply.
More than 8 GPUs? Discuss a clusterSources
- NVIDIA L4 (opens in a new tab)
- NVIDIA L4 GPU Accelerator Product Brief (opens in a new tab)
- NVIDIA Ada GPU Architecture (v2.02) (opens in a new tab)
- NVIDIA Video Encode and Decode Support Matrix (opens in a new tab)
- NVIDIA Multi-Instance GPU User Guide (opens in a new tab)
- NVIDIA Newsroom, March 21, 2023 (opens in a new tab)
Similar GPUs
- NVIDIA RTX A5000
24 GB GDDR6 · 0.77 TB/s · NVLink bridge
L4 leads on 2 of 4 metrics, RTX A5000 on 1. L4 has more FP32 (30.3 vs 27.8 TFLOPS) and FP16/BF16 (121 vs 111.1 TFLOPS). But RTX A5000 has more memory bandwidth (0.77 vs 0.3 TB/s). Equal: memory size.
- NVIDIA L40S
48 GB GDDR6 · 0.86 TB/s · PCIe 4
L40S leads on all 5 metrics. L40S has more FP8 (733 vs 242 TFLOPS) and FP32 (91.6 vs 30.3 TFLOPS).
Derived: half the with-sparsity FP16 figure NVIDIA gives for the RTX A5000 (222.2 TFLOPS); NVIDIA gives no dense one.
We’ll reply with terms for your workload. Nothing to pay before the contract.