Robinhood Chain / Pons / NVDA
gpuendpoints.com
$GPU trades on Pons against NVDA. Fees from that pay for GPU time. Hold 10,000 $GPU and the machines are free.
Treasury
0 NVDA
What is sitting aside for GPU time.
Access
10,000 $GPU
Minimum bag to start a machine.
GPUs
9
16–80 GB VRAM
Pair
NVDA
Buys, sells, and fees are in NVDA.
GPUs
Full listThree classes. Consumer cards (4090, 4080, 3090) for images and small models. Professional (A6000, A5000) when you need 24–48 GB without a datacenter bill. A100 / H100 / L40 for large training and inference. Starts last one hour, then the node stops.
Applications
All appsPrebuilt images so you are not assembling CUDA yourself. Pick an app in the market, then a GPU that fits it. Same 10k $GPU gate. Same one-hour cap.
Image generation
Stable Diffusion
Automatic1111 WebUI for text-to-image.
Image generation
ComfyUI
Node graph for Stable Diffusion pipelines.
Language models
Text Generation WebUI
Oobabooga UI for local LLMs.
Development
JupyterLab
Notebooks with PyTorch and TensorFlow.
Training
PyTorch Training
PyTorch 2.1, CUDA 11.8, JupyterLab.
Language models
vLLM
OpenAI-compatible LLM HTTP API.
Cluster
active 0load 0%
Online
waiting
Last change
RTX 4090
Feed