Rent NVIDIA A40 GPU by the hour
The NVIDIA A40 is a datacenter GPU built on the Ampere architecture with compute capability 8.6 and 45 GB of VRAM. On ONIE, it delivers about 29.9 TFLOPS of single-precision performance and about 536 GB/s of memory bandwidth within a 300 W power envelope. You can deploy it with per-hour billing, full SSH access, and ready templates including PyTorch/CUDA, vLLM, Ollama, and ComfyUI.
Specifications
VRAM per GPU
Architecture
Compute capability
FP32 compute
Memory bandwidth
Power limit
Values reported by the infrastructure provider for machines currently listed; real performance varies by machine.
Available configurations
Live stock and prices, updated continuously. The final price is confirmed before you order.
Loading live prices…
What the A40 is good for
Large Language Model Inference
Run models that require up to 45 GB of memory using pre-configured vLLM or Ollama templates.
Deep Learning Fine-Tuning
Fine-tune neural networks in PyTorch/CUDA with ample VRAM for larger batch sizes and weights.
Generative Image Workflows
Execute complex diffusion pipelines and hold multiple checkpoints in memory simultaneously using ComfyUI.
Datacenter Compute Tasks
Utilize about 29.9 TFLOPS of single-precision compute and about 536 GB/s of memory bandwidth for simulation and rendering workloads.
Is it the right GPU for you?
Choose the A40 when your workload requires 45 GB of VRAM on a datacenter Ampere architecture, which exceeds standard consumer memory capacities. Opt for a smaller consumer card if your models fit into less memory, or move to a higher-bandwidth datacenter GPU if your training jobs are heavily bottlenecked by memory throughput.
Compare with similar GPUs
Frequently asked questions
Is the NVIDIA A40 suitable for LLM inference and fine-tuning?
Yes, the 45 GB VRAM capacity makes the A40 well-suited for hosting medium-to-large language models and performing parameter-efficient fine-tuning. You can launch vLLM, Ollama, or PyTorch/CUDA templates directly on ONIE to start working immediately.
How does the A40 compare to smaller consumer GPUs or larger datacenter GPUs?
The A40 provides 45 GB of VRAM, giving you more memory headroom than standard consumer GPUs for memory-bound workloads. However, its memory bandwidth of about 536 GB/s is lower than that of higher-tier datacenter GPUs that use HBM memory.
What software environments are available for the A40 on ONIE?
The A40 supports CUDA compute capability 8.6 and can be deployed with ready templates for PyTorch/CUDA, vLLM, Ollama, and ComfyUI. Full SSH access is included so you can customize your software stack as needed.
How does billing work for A40 instances?
ONIE uses per-hour billing for all A40 rentals. You only pay for the time your instance is active, making it practical for both short experiments and longer production runs.
Rent the A40 by the hour
Create an account, choose a configuration and connect over SSH. Billing stops when you delete the GPU.
Rent A40