AGAnchorGPU
NVIDIAIn stock

Rent L40S GPU

A dedicated NVIDIA L40S with 48 GB GDDR6 ECC, 16 vCPU, 128 GB system RAM, and 2 TB local NVMe per card.

Node pricing

One price for the whole period.

The product model allocates each GPU to one account for the complete term. Adding cards to a node applies the catalog volume discount to every card.

NodeTotal VRAM7 days30 daysDiscount
1× L40S48 GB$148$5200%
2× L40S96 GB$281$9885%
4× L40S192 GB$545$1,9148%
8× L40S384 GB$1,066$3,74410%

Regional inventory

Three regions in the catalog.

DallasTexas · US-TX-1
260cards in catalog
AshburnVirginia · US-VA-1
110cards in catalog
PortlandOregon · US-OR-1
60cards in catalog

Ready-to-run images

Start with your stack installed.

Ubuntu 24.04CUDA readyPyTorch 2.8CUDA 12.8vLLMOpenAI-compatible APIComfyUIFlux & SDXLOllamaLocal model serverJAXCUDA 12Debian 13Minimal

Workload fit

When to consider L40S.

Good starting point

A 48 GB option to compare for video pipelines and inference when a 24 GB working set is too restrictive.

Your catalog configuration pairs 48 GB GPU memory with 128 GB host RAM and 2 TB local NVMe per card.

Before choosing the term

Separate model compute, memory pressure and video processing in your measurements. A combined pipeline can be limited outside the GPU kernels.

Use $148 for a 7-day pilot, or $520 for a 30-day allocation after validating the workload. Options are additional.

Optional resources

Add only what the node needs.

Options are billed for the same fixed term as the GPU node.

Extra NVMe storage1 TB / period

$9 / week $29 / month

Dedicated public IPv4address / period

$3 / week $10 / month

Extra bandwidth10 TB / period

$12 / week $39 / month

Private colocation VLANnode / period

$25 / week $85 / month