GPUs, Frameworks & Use Cases

Supported GPU models (RTX 4090 to H100), AI/ML frameworks, and real-world use cases from training to rendering.

10 min read

gpupytorchtensorflowai

Supported GPUs, Frameworks & Use Cases

FluxEdge supports a wide range of GPU hardware — from consumer-grade NVIDIA GeForce cards provided by community members to enterprise-grade H100 and Blackwell GPUs from NVIDIA partner Hyperstack. Because FluxEdge is infrastructure-level (Kubernetes + Docker), any framework that runs in a container works out of the box.

Supported GPU Models

Dedicated Machines (Community Providers)

GPU Family	Notable Models	Typical VRAM	Best For
NVIDIA GeForce RTX 40-series	RTX 4090, RTX 4080, RTX 4070 Ti	12-24 GB	AI inference, image generation, rendering
NVIDIA GeForce RTX 30-series	RTX 3090, RTX 3080, RTX 3070	8-24 GB	General GPU compute, inference, gaming
NVIDIA Professional	RTX A6000, A5000, Quadro	16-48 GB	Professional visualization, large model inference
AMD GPUs	Various models	Varies	Compute tasks with ROCm support

Premium Machines (Hyperstack / NVIDIA)

GPU	VRAM	Architecture	Best For
NVIDIA H100	80 GB HBM3	Hopper	Large-scale AI training, enterprise inference
NVIDIA H200	141 GB HBM3e	Hopper	Ultra-large model training, HPC workloads
NVIDIA A100	40/80 GB HBM2e	Ampere	Production ML training and inference
NVIDIA RTX A6000	48 GB GDDR6	Ampere	Professional visualization, model fine-tuning
NVIDIA Blackwell	TBA	Blackwell	Next-gen AI training and inference (planned)

Premium machines require KYC1 verification. These are sourced from NexGen Cloud's Hyperstack platform, an NVIDIA partner, ensuring enterprise-grade reliability and performance.

Supported Frameworks & Tools

FluxEdge is infrastructure-level — any application that runs in a Docker container works. Here are commonly used frameworks and tools confirmed on the platform:

Framework/Tool	Category	Docker Image Example
PyTorch	Deep Learning	pytorch/pytorch:latest
TensorFlow	Deep Learning	tensorflow/tensorflow:latest-gpu
ONNX Runtime	Inference	mcr.microsoft.com/onnxruntime/server
Jupyter Notebook	Interactive Development	Quick Launch template available
Ollama	LLM Inference	Quick Launch template available
Stable Diffusion	Image Generation	Quick Launch template available
NVIDIA NIM	Optimized Inference	NVIDIA NGC catalog
NVIDIA NeMo	AI Agents / RAG	NVIDIA NGC catalog
Hugging Face	Model Hub / Transformers	Custom images from HF Docker
vLLM	High-throughput LLM Serving	vllm/vllm-openai:latest
Blender	3D Rendering	Used in benchmarks; custom images

Use Cases

1
AI/ML Model Training
Train deep learning models with GPU acceleration. Use Dedicated GPUs (RTX 4090) for smaller models or Premium machines (H100, A100) for large-scale training. FluxEdge is ideal for teams that need burst GPU capacity without long-term cloud commitments.
2
AI Inference & LLM Hosting
Run inference servers for production or development. Deploy Ollama, vLLM, NVIDIA NIM, or custom model servers. The zero egress fees make inference serving particularly cost-effective.
3
Image & Video Generation
Run Stable Diffusion, DALL-E alternatives, or video generation pipelines. Quick Launch templates make this a one-click deployment.
4
3D Rendering
Offload Blender, Unreal Engine, or other rendering workloads to GPU machines. Especially useful for animation studios needing burst render capacity.
5
Scientific Computing & HPC
Run computational fluid dynamics, molecular dynamics, bioinformatics, or other GPU-accelerated scientific workloads.
6
Data Processing
Use GPU-accelerated data processing tools (RAPIDS, cuDF) for large-scale data analytics. Dedicated CPU machines are also available for non-GPU data workloads.
7
Development & Prototyping
Spin up Jupyter notebooks with GPU access for quick prototyping and experimentation without setting up local GPU environments.

Crypto mining is also permitted on FluxCore provider machines that opt into it. The Auto-Switch feature seamlessly transitions between mining and rental workloads for maximum utilization.