NVIDIA cloud GPUs from ₹27.74/hour
RTX 3090, L4, A100, RTX 4090. Per-minute billing in INR via UPI — no FX fees, no foreign card needed. Spin up an instance in 60 seconds, run your ML/Stable Diffusion/LLM workload, stop when done. Competitive pricing with the major US-based providers, billed in rupees with native UPI checkout.
Why pay in USD to RunPod when you can pay in rupees via UPI to an Indian provider?
Cloud GPU rental for ML and AI work has been dominated by US-based providers (RunPod, Lambda Labs, AWS, Google Cloud, Azure) charging in USD. For Indian developers and ML practitioners, that creates three layers of unnecessary friction:
- USD billing means every minute of GPU time gets converted at your bank's FX rate plus a 2-5% foreign transaction fee. On a 10-hour Stable Diffusion training run that's ₹50-150 in invisible currency overhead.
- Foreign card requirement blocks students and freelancers without international cards from using cloud GPUs at all.
- No UPI checkout means signup friction even when you have a card — manual address entry, foreign verification flows, billing surprises.
AIC Cloud's GPU service runs on the same NVIDIA hardware as the major Western providers — same RTX 3090, same A100 — and pricing per-hour is broadly comparable. The differentiation is what surrounds the price: INR billing (no FX, no foreign card fee), native UPI checkout, real human WhatsApp support, and per-minute billing without hourly minimums. Same GPUs, equivalent pricing, dramatically less friction.
Cloud GPU pricing — AIC vs the alternatives
Per-hour rates for the most common ML GPU. Real rupee cost including FX layer where applicable.
| Provider | RTX 3090 (24 GB) | RTX 4090 (24 GB) | A100 (80 GB) | Billing currency |
|---|---|---|---|---|
| AIC Cloud | ₹27.74/hr (~$0.33) | ₹138/hr (~$1.64)* | ₹163/hr (~$1.94) | INR via UPI ✅ |
| RunPod | $0.34/hr (~₹29) | $0.69/hr (~₹58) | $1.89/hr (~₹161) | USD, intl card |
| Lambda Labs | Not offered | Not offered | $1.29/hr (limited availability) | USD, intl card |
| AWS p3 / p4 / g5 | $1.20/hr (g5.xlarge ≈ A10G) | Not offered | $3.06/hr (p4d, per A100) | USD, complex pricing |
Competitor rates from public pricing pages as of 2026; USD figures converted at ~₹85/USD.
Common workloads on AIC Cloud GPU
🎨 Stable Diffusion / ComfyUI
RTX 3090 generates a 1024×1024 SDXL image in ~6-8 seconds at 30 steps. Run ComfyUI, AUTOMATIC1111, Forge, or InvokeAI. 24 GB VRAM handles ControlNet + LoRA + multiple models simultaneously.
~₹14 for a 30-minute generation session.
🧠 LLM fine-tuning (QLoRA)
Fine-tune Llama 3.1 8B with QLoRA on a single RTX 3090. Mistral Small 22B fits on RTX 4090. Full fine-tuning of 70B-class models needs A100 80 GB or multi-GPU setups.
~₹100-300 for a typical Llama 3.1 8B QLoRA run.
⚡ Local LLM inference (Ollama / vLLM)
Serve Llama 3.1 8B at 65-80 tokens/sec on RTX 3090. Mistral 22B at ~22 t/s. Build private chatbots, code assistants, or local AI agents without per-token API costs.
~₹665/day for 24/7 hosted Llama 8B (auto-stop drops this dramatically).
🎬 3D rendering (Blender)
Cycles rendering on RTX 3090 is roughly 5-10× faster than CPU-only. RTX 4090 nearly doubles that throughput. Perfect for architectural viz, product renders, animation frames.
~₹277 for a 10-hour render queue.
🔬 ML experimentation / Kaggle
Iterate on model architectures, hyperparameter tuning, dataset preprocessing. Per-minute billing means stopping a run mid-experiment costs nothing extra — perfect for the trial-and-error loop of ML research.
Total control over what you pay for.
🎮 Cloud gaming (Parsec)
Run AAA games on a cloud GPU and stream via Parsec / Moonlight to your laptop. RTX 4090 plays anything at max settings. Niche use case but works well for low-end laptop users.
~₹30/hour of actual play time.
Get started in 60 seconds
- 1
Top up your wallet via UPI
₹500 covers ~18 hours of RTX 3090 time. PhonePe, GPay, Paytm, net banking, or any Indian debit/credit card via Razorpay.
- 2
Pick GPU + template
Dashboard → Cloud GPU → pick RTX 3090 / 4090 / A100 → choose a pre-built template (PyTorch, Ollama, ComfyUI, Stable Diffusion WebUI) or vanilla Ubuntu + CUDA.
- 3
SSH in, run your workload
Get SSH credentials from dashboard. NVIDIA drivers + CUDA are pre-installed. Verify with
nvidia-smi. Run your training script, ComfyUI server, or Ollama instance. - 4
Stop when done
Dashboard → Stop. Billing pauses immediately (per-minute granularity). Resume anytime — your data stays on the persistent volume.
Cloud GPU — frequently asked questions
What's the cheapest cloud GPU in India for AI / ML work?
AIC Cloud's RTX 3090 at ₹27.74/hour (~$0.33/hr) is the entry GPU tier with INR/UPI billing. RTX 3090 has 24 GB VRAM which fits Llama 3.1 8B, Stable Diffusion XL, Mistral Small, and most agent workloads comfortably. For more headroom, L4 (22 GB) at ₹44.21/hr or A100 (80 GB) at ~₹163/hr are also available with per-minute billing.
How does AIC Cloud GPU pricing compare to RunPod, Lambda Labs, AWS?
AIC Cloud RTX 3090: ₹27.74/hr (~$0.33). RunPod RTX 3090 Community: $0.34/hr (~₹29). Lambda Labs RTX 3090: ~$0.50/hr. Pricing is broadly comparable to RunPod. The differentiation isn't price alone — it's INR billing (no FX, no foreign card fee), native UPI checkout, real human WhatsApp support, and per-minute billing. AWS, GCP, Azure equivalent GPU instances are 5-10× more expensive.
Is the GPU billed hourly or per-minute?
Per-minute. Run your training job for 47 minutes, pay for 47 minutes — not an hour. This matters for iterative work (Stable Diffusion experiments, LLM evaluations) where you start and stop frequently.
Can I run Stable Diffusion XL on RTX 3090?
Yes, comfortably. SDXL needs ~10-12 GB VRAM at FP16; RTX 3090 has 24 GB so you have headroom for ControlNet, LoRA training, multiple models loaded. Generation speed ~6-8 seconds for a 1024×1024 image at 30 steps.
Which GPU should I pick for LLM fine-tuning?
For Llama 3.1 8B QLoRA: RTX 3090 (24 GB) is fine. For Llama 3.1 70B QLoRA: you need A100 (80 GB) or multi-GPU. For Mistral 22B QLoRA: RTX 4090 or RTX 3090 with offload. Quantised QLoRA dramatically reduces VRAM needs.
Does AIC Cloud accept UPI for GPU rental?
Yes. GPU service runs on the same wallet/billing as the rest of the platform — top up via UPI (PhonePe, GPay, Paytm), net banking, or Razorpay. GPU minutes deduct in real time.
How fast does a GPU instance spin up?
Typically 30-60 seconds from clicking 'Deploy' to SSH access on a fully provisioned Ubuntu 22.04 image with CUDA drivers installed. Pre-built templates for PyTorch, ComfyUI, Ollama available.
What happens if I forget to stop the GPU?
It keeps billing. Set a wallet balance alert in your dashboard, or use auto-stop after N minutes of GPU-idle. Per-minute billing limits damage — a forgotten 24-hour RTX 3090 is ~₹665.
Can I use the GPU for crypto mining?
No. Crypto mining is against ToS across virtually every cloud GPU provider — generates almost no margin, creates wear/heat issues. Mining workloads are auto-detected and instances terminated.
What about A100 / H100 availability?
A100 80 GB available (~₹163/hr typically). H100 availability is limited globally in 2026. For most ML work outside trillion-parameter training, RTX 4090 or A100 are the sweet spot.
Start training, generating, fine-tuning
₹27.74/hour for RTX 3090. Per-minute billing. INR + UPI checkout. No contracts. Spin up in 60 seconds.
Free setup help · No setup fee · WhatsApp support included