# Choosing your Rental

### **Understanding Your Compute Needs**

Before renting a GPU, consider the **type of workload** you’re running. Different tasks have **different GPU requirements**, and selecting the wrong hardware can lead to **wasted resources or suboptimal performance**.

#### **Common GPU Use Cases:**

| **Workload Type**                                             | **Recommended GPU Type**     | **Key Factors**                                        |
| ------------------------------------------------------------- | ---------------------------- | ------------------------------------------------------ |
| **AI Training (Deep Learning, LLMs)**                         | **A100, H100, RTX 4090**     | High VRAM, Tensor Cores, Multi-GPU support             |
| **AI Inference (Model Deployment, Fine-Tuning)**              | **RTX 3090, RTX 4090, A100** | Lower VRAM needs, but high precision compute required  |
| **3D Rendering (Blender, Unreal, AI-Generated Art)**          | **RTX 3090, RTX 4090, A100** | High CUDA cores, VRAM, and Tensor RT support           |
| **Data Science (Simulations, High-Performance Computing)**    | **RTX 4090, A100, 3090**     | High FP32/FP64 performance, large dataset processing   |
| **Video Processing (Encoding, AI Upscaling)**                 | **RTX 3090, 4090, A6000**    | Fast memory bandwidth, CUDA acceleration               |
| **Blockchain & Cryptographic Tasks (ZK-Proofs, Computation)** | **RTX 4090, A100**           | High core count, memory bandwidth, parallel processing |

Each of these tasks requires different **levels of compute power, memory, and bandwidth**, so **choosing the right GPU can significantly impact execution time and costs**.

***

### **Browsing GPUs in the Marketplace**

Once you understand your workload, you can **browse available GPUs** in Nebula AI’s marketplace. The platform provides **detailed specifications** for each listed GPU, including:

* **GPU Model & Generation** – Determines **compute power & efficiency** (e.g., RTX 4090 vs. RTX 3090).
* **VRAM Size** – Important for **deep learning, rendering, and large dataset processing**.
* **Performance Benchmarks** – FLOPs, Tensor core efficiency, and past rental performance.
* A**vailability & Location** – Some GPUs may be located in **specific geographic regions** (lower latency).
* **Rental Price & Type** – Choose between **Spot vs. On-Demand pricing models**.

Clicking on a listing provides **a deeper breakdown**, helping you make **a data-driven decision**.

***

### **Comparing Spot vs. On-Demand Rentals**

Nebula AI provides **two rental options**:

#### **On-Demand Rentals (Guaranteed, Higher Cost)**

* **Guaranteed access** to the GPU for the selected duration.
* **Fixed pricing**, ensuring uninterrupted compute time.
* Best for **long AI training, stable inference workloads, and critical projects**.
* More expensive than **Spot Rentals** but ensures **job completion** without interruptions.

#### **Spot Rentals (Lower Cost, Flexible Availability)**

* Rent at **discounted prices**, but can be **outbid** by other users.
* Ideal for **non-time-sensitive tasks**, such as **batch processing or exploratory workloads**.
* **Can be interrupted** if another user **bids a higher price**, requiring workload resumption.

{% hint style="info" %}
If your workload requires **consistent uptime**, go for **On-Demand Rentals**. If you’re flexible and want to **minimize costs**, Spot Rentals can **save up to 40%**.
{% endhint %}

***

### **Evaluating GPU Host Ratings & Reliability**

Since Nebula AI is a **decentralized GPU marketplace**, different GPU providers (hosts) have varying levels of **uptime and reliability**. Before renting, check:

* **Host Uptime %** – Indicates how consistently the provider keeps their GPU available.
* **Rental History** – Shows previous rental completions and renter satisfaction.
* **User Ratings** – High ratings suggest a **trusted host with stable performance**.
* **Connection Speed** – Some hosts have **faster internet speeds**, which is important for **real-time AI inference**.

Prioritizing **high-uptime hosts** ensures you **avoid disruptions** and maintain **stable compute performance**.

***

### **Choosing the Right GPU for Long-Term vs. Short-Term Rentals**

* **Short-Term Rentals (<24 hours)** – Best for **quick tests, benchmarking, and exploratory workloads**.
* **Medium-Term Rentals (1-7 days)** – Ideal for **training small AI models, rendering, and iterative deep learning**.
* **Long-Term Rentals (>7 days)** – Used for **extended AI training, large-scale simulations, and continuous workloads**.

For **longer rental durations**, consider negotiating **bulk pricing** with hosts or using **future automated pricing discounts**.

***

### **Selecting Additional Features (If Needed)**

Some GPU listings may provide **extra services** to improve the rental experience:

* **Pre-installed AI/ML Environments** – Comes with **PyTorch, TensorFlow, Jupyter Notebooks** pre-configured.
* **Remote Storage Options** – Some hosts offer **persistent storage** for long-term workloads (upcoming feature).
* **High-Speed Networking** – Low-latency connections for real-time applications.
* **Custom Software Configurations** – Tailored setups for **specific AI models or research use cases**.

If you need **plug-and-play GPU access**, prioritize listings that include **pre-configured software environments**.

***

### **Finalizing Your Selection & Renting the GPU**

Once you’ve identified the **best GPU for your task**, proceed with the rental process:

**Confirm Pricing & Duration** – Ensure you’ve selected **Spot or On-Demand** based on your needs.

**Review Host Reliability** – Check uptime, network speeds, and past renter feedback.

**Check GPU Specs One Last Time** – Make sure **VRAM, CUDA cores, and compute power** match your workload.

**Click "Rent Now" & Confirm Payment** – Approve the **on-chain transaction** in your connected wallet.

**Deploy Workloads Immediately** – Access the GPU via **SSH, Jupyter Notebook, or API**.