NVIDIA A40 cloud GPU rental
Rent an NVIDIA A40 cloud GPU from $0.638 per hour with 48 GB of GDDR6 at 696 GB/s of memory bandwidth. FastInfra bills pods per minute with no commitment, so a one-hour experiment costs $0.638 and you stop paying the moment you delete the pod. Run 1 to 10 GPUs per pod, connect over SSH or JupyterLab, and see the exact price before you deploy.
Prices synced November 3, 2026 · billed per minute
A40 specs and pricing
The NVIDIA A40 is a datacenter Ampere GPU with 48 GB of GDDR6 at 696 GB/s of memory bandwidth. On FastInfra it is a good fit for 30B–70B inference, full fine-tuning of mid-size models, video generation. As a rule of thumb, one A40 holds the weights of a model of about 19B parameters at 16-bit precision, or about 76B with 4-bit quantization, after leaving room for the KV cache.
The A40 runs on the Secure tier in vetted datacenters at $0.638 per GPU-hour.
A40 price per hour by GPU count
Live FastInfra prices per pod-hour. Pods bill per minute, so you pay only for the minutes a pod exists.
| Pod size | Total VRAM | Secure / hr | Month (730 h, lowest) |
|---|---|---|---|
| 1× A40 | 48 GB | $0.638 | $466 |
| 2× A40 | 96 GB | $1.28 | $931 |
| 4× A40 | 192 GB | $2.55 | $1,863 |
| 8× A40 | 384 GB | $5.10 | $3,726 |
What an A40 costs over time
One GPU at the lowest on-demand rate ($0.638/hr).
| Duration | On-demand |
|---|---|
| 1 hour | $0.638 |
| 8-hour working day | $5.10 |
| 24 hours | $15.31 |
| 1 week (168 hours) | $107 |
| 1 month (730 hours) | $466 |
Which LLMs fit on the A40?
GPUs needed to hold the model weights with about 20% of VRAM left for the KV cache. Rule of thumb only: long contexts and big batches need more.
| Model size | Examples | FP16 / BF16 | 8-bit | 4-bit |
|---|---|---|---|---|
| 7–8B | Llama 3.1 8B, Qwen3 8B, Mistral 7B | 1 GPU | 1 GPU | 1 GPU |
| 12–14B | Gemma 3 12B, Qwen3 14B, Phi-4 | 1 GPU | 1 GPU | 1 GPU |
| 27–32B | Gemma 3 27B, Qwen3 32B | 2 GPUs | 1 GPU | 1 GPU |
| 70B | Llama 3.3 70B | 4 GPUs | 2 GPUs | 1 GPU |
| 120B | GPT-OSS 120B | 8 GPUs | 4 GPUs | 2 GPUs |
| 235B | Qwen3 235B-A22B | Needs more GPUs | 8 GPUs | 4 GPUs |
| 405B | Llama 3.1 405B | Needs more GPUs | Needs more GPUs | 8 GPUs |
| 671B | DeepSeek V3 / R1 | Needs more GPUs | Needs more GPUs | Needs more GPUs |
Best for
30B–70B inference, full fine-tuning of mid-size models, video generation.
About the Ampere architecture
Ampere brought third-generation Tensor Cores with BF16 and TF32 support, which modern training code relies on, and the A100 added Multi-Instance GPU partitioning. Ampere cards remain a dependable, lower-cost option for fine-tuning, inference and research, though they lack the FP8 support of Ada, Hopper and Blackwell.
Only need inference?
If you want to call an open model such as Llama, Qwen, DeepSeek or GPT-OSS rather than manage a server, the pay-per-token inference API is often cheaper than keeping a GPU busy around the clock. It is OpenAI-compatible and uses the same account and wallet as your pods. Browse every model with live prices.
A40 vs similar GPUs
Live lowest on-demand price per GPU-hour, and price per GB of VRAM.
| GPU | VRAM | Architecture | From / hr | Per GB-hr |
|---|---|---|---|---|
| A40 (this page) | 48 GB | Ampere | $0.638 | $0.013 |
| NVIDIA RTX A6000 | 48 GB | Ampere | $0.357 | $0.007 |
| NVIDIA L40 | 48 GB | Ada Lovelace | $0.746 | $0.016 |
| NVIDIA RTX 6000 Ada | 48 GB | Ada Lovelace | $0.80 | $0.017 |
| NVIDIA L40S | 48 GB | Ada Lovelace | $0.854 | $0.018 |
| NVIDIA RTX PRO 5000 | 48 GB | Blackwell | $0.886 | $0.018 |
| NVIDIA A100 SXM 40GB | 40 GB | Ampere | $1.08 | $0.027 |
A40 rental — frequently asked questions
How much does it cost to rent an NVIDIA A40 per hour?
$0.638 per GPU-hour on demand. Billing is per minute while the pod exists, so 24/7 use for a 730-hour month costs about $466 per GPU at the lowest on-demand rate. Prices sync from live capacity and are shown before you deploy.
How much VRAM does the NVIDIA A40 have?
48 GB of GDDR6 at 696 GB/s of memory bandwidth. The chip is rated at up to 300 W. A pod with 10 GPUs gives 480 GB of combined VRAM for tensor- or pipeline-parallel workloads.
What size LLM can I run on an NVIDIA A40?
Roughly 19B parameters at FP16 or BF16, 38B at 8-bit and 76B at 4-bit on a single GPU, reserving about 20% of VRAM for the KV cache and activations. Mixture-of-experts models need memory for all experts, not only the active ones. Long contexts and large batches need more headroom.
Is the NVIDIA A40 billed per hour or per minute?
Per minute. The hourly rate is divided by 60 and charged for every minute the pod exists. Attached volumes bill per GB-month and continue while a pod is stopped; delete the pod to stop all charges. There is no minimum term or reservation fee.
How do I rent an NVIDIA A40 on FastInfra?
Create an account, add prepaid credit, open the GPU configurator and choose the A40, the tier, the GPU count, a container template and storage. The configurator shows the hourly and monthly price, then deploys a pod you reach over SSH or JupyterLab within minutes.
Can I rent multiple A40 GPUs in one pod?
Yes, up to 10 per pod. The price scales linearly: 8 GPUs cost $5.10 per hour on Secure.
Is there a cheaper GPU than the A40 with the same VRAM?
Yes. The NVIDIA RTX A6000 has 48 GB and starts at $0.357 per hour. It differs in architecture (Ampere), memory bandwidth and multi-GPU interconnect, so compare throughput for your workload before switching.
Deploy an A40 in minutes
Price the exact pod, storage and template in the configurator, add credit, and connect over SSH or JupyterLab. Billed per minute, no commitment.