tinygrad
To commoditize the petaflop and enable AI for everyone by building a simple, open-source neural network framework and affordable high-performance hardware.
At a Glance
- AI Researchers
- Deep Learning Engineers
- Small-to-medium AI labs
- Hobbyists ('GPU middle class')
AI Tools by tinygrad
(1)
tinygrad
Minimalist Deep Learning Framework
Discussions
No discussions yet
Be the first to start a discussion about tinygrad
Latest News
Tiny Corp releases first eGPU drivers for Apple Silicon Macs, exclusive to tinygrad.
Tiny Corp starts accepting pre-orders for the $10 million Exabox AI cluster.
Tiny Corp announces shipping-container scale AI hardware for large-scale training.
Tinygrad adds support for Intel Level Zero backends.
Products & Services
A simple and powerful open-source neural network framework that breaks down complex networks into 3 OpTypes.
A workstation with 4x AMD Radeon 9070XT GPUs, 64GB GPU RAM, and 32-core AMD EPYC CPU. Priced at $12,000.
A workstation with 4x RTX PRO 6000 Blackwell GPUs, 384GB GPU RAM, and 32-core AMD GENOA CPU. Priced at $65,000.
A large-scale containerized AI cluster with 720x RDNA5 GPUs, targeting ~1 Exaflop of performance. Priced at approximately $10 million.
Market Position
Positioned as a transparent, open-source alternative to the NVIDIA/CUDA ecosystem and heavy frameworks like PyTorch. Focuses on hardware interoperability and lowering the cost of compute.
Leadership
Founders
George Hotz
Founder of comma.ai; known as 'geohot', the first person to jailbreak the iPhone and reverse engineer the PlayStation 3. Worked at Google Project Zero, SpaceX, and Facebook.
Executive Team
George Hotz
CEO & Founder
Prolific hacker and founder of comma.ai. Leading the development of tinygrad and the tiny corp's hardware strategy.
Board of Directors
Founding Story
Started as a toy project by George Hotz in Oct 2020 to teach himself neural networks. It evolved into a mission to disrupt the complex and proprietary AI hardware/software stacks by offering a simpler, hardware-agnostic alternative.
Business Model
Revenue Model
Direct sales of high-performance AI hardware (workstations and clusters). The software framework (tinygrad) is open-source.
Pricing Tiers
4x AMD GPUs, 64GB GPU RAM
4x NVIDIA GPUs, 384GB GPU RAM
Containerized cluster, 720x GPUs
Target Markets
- AI Researchers
- Deep Learning Engineers
- Small-to-medium AI labs
- Hobbyists ('GPU middle class')
- LLM inference and training
- AI workstation computing for independent researchers
- Large-scale GPU cluster management
- Edge AI development
- Independent AI researchers and open-source contributors