> For the complete documentation index, see [llms.txt](https://docs.stg.ai.neevcloud.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.stg.ai.neevcloud.com/vm-service/overview.md).

# Overview

NeevCloud Virtual Machines give you on-demand compute infrastructure you can provision in minutes. You choose the hardware configuration, attach a pre-built AI template or bare OS, and connect via SSH, no setup overhead, no long-term commitment.

NeevCloud VMs are purpose-built for AI/ML workloads, with NVIDIA GPU-powered instances and single-region availability in Indore, India.

## What You Can Build on NeevCloud VMs

NeevCloud VMs are purpose-built for AI/ML infrastructure. Here is what you can do with them:

* Train deep learning models, LLMs, vision models, speech models, and recommender systems on NVIDIA GPU hardware
* Run LLM inference at scale using vLLM or TGI behind an OpenAI-compatible API endpoint
* Fine-tune foundation models (LLaMA 3, Mistral, Gemma, Falcon) with LoRA, QLoRA, or full fine-tuning
* Build and serve computer vision pipelines for object detection, segmentation, and real-time video analytics
* Run generative AI workloads, diffusion models, code generation, and multimodal systems
* Preprocess and engineer features from large datasets before training
* Orchestrate ML pipelines, experiment tracking, and model evaluation using CPU VMs

## VM Types

NeevCloud offers two VM types. Your choice depends on whether your workload needs GPU acceleration.

|                | CPU VM                                                                                | GPU VM                                                                 |
| -------------- | ------------------------------------------------------------------------------------- | ---------------------------------------------------------------------- |
| Purpose        | General-purpose and AI pipeline compute                                               | AI/ML training, inference, and HPC                                     |
| Hardware       | Virtual CPU cores + RAM, no GPU                                                       | NVIDIA GPU (T4 / A30 / RTX 5090 / A100 / H100 / H200) + vCPU + RAM     |
| Primary AI Use | Data pipelines, preprocessing, model evaluation, orchestration, lightweight inference | Model training, fine-tuning, LLM inference, diffusion, computer vision |
| Region         | as-south-1 (Indore, India)                                                            | as-south-1 (Indore, India)                                             |

| Info: If your workload does not call a CUDA kernel, meaning it does not use PyTorch, TensorFlow, or any GPU-accelerated library at runtime, a CPU VM is the right and more cost-effective choice. |
| ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |

## CPU VM Use Cases (AI/ML Focused)

CPU VMs support AI workflows that focus on data processing, orchestration, and lightweight execution where GPU acceleration is not required.

* **Data Preprocessing & Feature Engineering** - Clean, transform, and prepare datasets for model training. Run pandas, Spark, or Dask pipelines on large tabular and text corpora before they reach the GPU.
* **Model Evaluation & Testing** - Run validation loops, benchmarking scripts, and accuracy checks on already-trained model checkpoints. Evaluate on held-out test sets without occupying a GPU.
* **Lightweight AI Inference** - Serve small or optimized ML models like XGBoost, LightGBM, and classical scikit-learn models. Also suitable for quantized GGUF models served via llama.cpp on CPU.
* **AI Pipeline Orchestration** - Manage ML workflows, job scheduling, and automation using tools like Apache Airflow, Prefect, or Kubeflow Pipelines. The orchestration layer does not need GPU compute.
* **AI Development Environments** - Host JupyterLab notebooks and VSCode Server environments for dataset exploration, debugging, and rapid prototyping before committing to a full training run on a GPU VM.
* **Distributed Training Coordination** - Handle parameter servers, coordinator nodes, and orchestration roles in large-scale distributed training setups where the CPU node manages communication between GPU workers.

## GPU VM Use Cases

GPU VMs are designed for high-performance computing workloads that require parallel processing and CUDA acceleration.

* **AI Model Training** - Train deep learning models from scratch or from checkpoints including LLMs, vision transformers, diffusion models, and recommender systems. Multi-GPU configurations are supported for large runs.
* **AI Model Inference** - Deliver fast, high-throughput predictions for large-scale AI applications. Serve models with vLLM, TGI, Triton Inference Server, or TorchServe for production-grade throughput.
* **Generative AI Applications** - Power LLMs, Stable Diffusion, SDXL, and other content generation models for text, image, video, and code. Run real-time generation APIs that need low latency and high VRAM.
* **Computer Vision Processing** - Enable object detection (YOLO, DETR), semantic segmentation, facial recognition, and real-time video analytics pipelines at scale.
* **Scientific Computing & Simulations** - Accelerate complex simulations in physics, climate modeling, and quantitative finance using GPU-optimized numerical libraries like CuPy, JAX, and RAPIDS.

## Deployment Region

All NeevCloud VM instances are currently available in the following region:

| Region ID  | Location      | Notes                                                                                  |
| ---------- | ------------- | -------------------------------------------------------------------------------------- |
| as-south-1 | Indore, India | Primary region for all GPU and CPU VM instances. Low-latency access from across India. |