How to choose the right configuration for your tasks?

Choosing the right configuration depends on your workload: model training, rendering, inference, or gaming. Key parameters include GPU video memory (VRAM), CPU core count, system RAM, and type of disk storage.

For AI Model Training best suited GPUs:

• Tesla V100 (32 GB)
• A100 (80 GB)
• H100 (80 GB)
• H200 (141 GB)

These accelerators deliver top performance in floating-point operations and support large-scale models.

RAM Recommendations:

1. System RAM should be at least equal to the total VRAM of all GPUs in the system.
2. For example: 1× A100 (80 GB VRAM) → minimum 80 GB RAM
3. For 4× A100 → minimum 320 GB RAM

This ensures efficient data loading, intermediate result storage, and large batch processing.

For rendering & High-End Gaming we recommend the following GPUs (RTX series):

• RTX 2080 Ti (11 GB) – suitable for light rendering or entry-level gaming  
• RTX 3090 / RTX 4090 / A5000 (24 GB each) – ideal for professional rendering and high-fidelity gaming  
• RTX 5090 (32 GB) – flagship option for 8K rendering, real-time ray tracing, and complex scenes  

System Recommendations:

1. CPU: Minimum 8 cores  
2. RAM: At least 24 GB per GPU, depending on scene complexity or game requirements  
3. Storage: SSD (≥160 GB) for fast texture and scene loading  

Efficient GPU options for Model Inference are:

• Tesla T4 (16 GB) – cost-effective for small and medium models  
• Tesla A10 (24 GB) – versatile for LLMs, diffusion models, and multimodal tasks  
• Tesla A2 (16 GB) – optimized for lightweight and edge inference  
• RTX 3080 (10 GB) – excellent price-to-performance for real-time inference  

RAM Recommendations:

1. T4 / A2: Minimum 16 GB RAM
2. A10 / RTX 3080: 24–32 GB RAM, especially when handling long contexts or concurrent requests  

Streaming & Lightweight Workloads:

• RTX 2080 Ti or T4 are well-suited  
• RAM: 8–16 GB is sufficient  
• Storage: 100–200 GB HDD or SSD (disk I/O is not a bottleneck here)

Storage Recommendations:

For disk-intensive workloads (e.g., large datasets, frequent I/O):  

• Use SSD-based volumes (`Volume` instance + SSD-backed Volume)  
• Or choose non-replicated local SSDs (`Local` instances) for maximum I/O performance  

For less demanding tasks:  Use the more cost-effective HDD-backed volumes.

⚠️ Important: All configurations are scalable: you can add GPUs, increase RAM, or expand storage as your workload grows.