the tiny corp

@__tinygrad__

We make tinygrad; sell tinybox for the GPU middle class. Our mission is to commoditize the petaflop.

following
followers

Vera Rubin's claimed speed advantage over RTX 6000 and B300 racks

A user says their calculations make buying either rack hard to justify and has requested prices from several sellers.

3 Sources, 15h ago, updated 15h ago

First tinybox red flash goes up for $19,000

Tinygrad’s October 16 announcement invited buyers to get one.

7 Sources, 3d ago, updated 3d ago

The case against prioritizing inference speed over AI accelerator fundamentals

A post calls accelerator companies’ focus on inference and tokens per second at the expense of fundamentals a serious mistake.

1 Source, 5d ago, first seen 5d ago

10-411's week-five goal: An MNIST trainer and Qwen3 runner built using your compiler

The tinygrad account says weeks six through 10 will focus on GPU usage and performance.

1 Source, 5d ago, first seen 5d ago

Meta introduces Muse Wonder and Muse Wonder Full AI models

Meta says Muse Wonder launched through its Model API and Muse Code, while Full was available to Muse Code subscribers.

152 Sources, 15h ago, updated 15h ago

FirstFT: Mark Zuckerberg says Meta has reached AI ‘frontier’

Want to deploy exaboxes?

A post asks whether readers want to deploy exaboxes while quoting another post, without adding details.

1 Source, 11d ago, first seen 11d ago

AMD reportedly gave Tinygrad a $2 million contract

Tinygrad says AMD also sent it boxes and calls its software “quite usable” now.

3 Sources, 11d ago, first seen 11d ago

New tinybox with a 4090, 7900 XTX and 9070 XT is now available

The tinygrad account says it used the G5/192/4TB machine for tinygrad development all summer.

3 Sources, 12d ago, updated 12d ago

Running UIR programs by the end of week three

Tinygrad asks readers to identify the PyTorch equivalent of the UIR example it linked.

1 Source, 13d ago, first seen 13d ago

Compilers for Machine Learning planned at Carnegie Mellon for spring 2027

The tinygrad account says the Pittsburgh class will meet three mornings a week and lectures won’t be posted online.

1 Source, 14d ago, first seen 14d ago

Tinygrad says it finished AMD MLPerf contract with all four targets hit

Tinygrad said in X posts that it finished an AMD MLPerf contract with all four targets hit, criticized Tenstorrent, Qualcomm and Intel, and separately argued for a far smaller datacenter hardware design.

2 Sources, 14d ago, updated 14d ago

OpenAI releases results from its internal AI math model on GitHub

OpenAI says it drew on an independent mathematics advisory group’s advice about how to release the results.

435 Sources, 9d ago, updated 9d ago

How does compiler performance factor into the 'Compilers for Machine Learning' grading breakdown?

The poster says compiler performance is graded on a curve, with the bottom performer getting 0 points and the top getting 15.

2 Sources, 15d ago, first seen 15d ago

tinygrad plans a VS Code syntax-highlighting plugin for *.uir files

tinygrad says it needs a stable on-disk UOp representation before version 1.0.

1 Source, 15d ago, first seen 15d ago

tinygrad’s UOp IR now renders in VIZ=1 on master

Tinygrad says clicking a graph node or a variable in the UOp IR view syncs the two views.

1 Source, 15d ago, first seen 15d ago

The case for tinygrad as ML hardware becomes more heterogeneous

The tinygrad account predicts that increasingly varied machine-learning hardware will make its case more obvious.

2 Sources, 17d ago, first seen 17d ago

A text-based representation for tinygrad's UOps in student compilers

The tinygrad account says students need a UOps format they can parse and output, and asks for feedback.

2 Sources, 18d ago, first seen 18d ago

Robot RAM needs for action and knowledge

A reply speculates that robots will need little RAM, have limited knowledge and look things up online when needed.

2 Sources, 19d ago, first seen 19d ago

University host sought for a winter-semester machine-learning compilers course

The tinygrad account asks universities to host George for a 12-plus-week, hands-on course in which students would build a compiler.

1 Source, 19d ago, first seen 19d ago

Qwen3.8-27B reportedly runs at about 90 tokens per second on a four-year-old AMD GPU over USB

tinygrad says its small size and pure-Python code make it particularly easy to build on.

1 Source, 21d ago, first seen 21d ago

Featured in

Vera Rubin's claimed speed advantage over RTX 6000 and B300 racks

First tinybox red flash goes up for $19,000

The case against prioritizing inference speed over AI accelerator fundamentals

10-411's week-five goal: An MNIST trainer and Qwen3 runner built using your compiler

Meta introduces Muse Wonder and Muse Wonder Full AI models

Want to deploy exaboxes?

AMD reportedly gave Tinygrad a $2 million contract

New tinybox with a 4090, 7900 XTX and 9070 XT is now available

Running UIR programs by the end of week three

Compilers for Machine Learning planned at Carnegie Mellon for spring 2027

Tinygrad says it finished AMD MLPerf contract with all four targets hit

OpenAI releases results from its internal AI math model on GitHub

How does compiler performance factor into the 'Compilers for Machine Learning' grading breakdown?

tinygrad plans a VS Code syntax-highlighting plugin for *.uir files

tinygrad’s UOp IR now renders in VIZ=1 on master

The case for tinygrad as ML hardware becomes more heterogeneous

A text-based representation for tinygrad's UOps in student compilers

Robot RAM needs for action and knowledge

University host sought for a winter-semester machine-learning compilers course

Qwen3.8-27B reportedly runs at about 90 tokens per second on a four-year-old AMD GPU over USB

Featured in

Vera Rubin's claimed speed advantage over RTX 6000 and B300 racks

First tinybox red flash goes up for $19,000

The case against prioritizing inference speed over AI accelerator fundamentals

10-411's week-five goal: An MNIST trainer and Qwen3 runner built using your compiler

Meta introduces Muse Wonder and Muse Wonder Full AI models

Want to deploy exaboxes?

AMD reportedly gave Tinygrad a $2 million contract

New tinybox with a 4090, 7900 XTX and 9070 XT is now available

Running UIR programs by the end of week three

Compilers for Machine Learning planned at Carnegie Mellon for spring 2027

Tinygrad says it finished AMD MLPerf contract with all four targets hit

OpenAI releases results from its internal AI math model on GitHub

How does compiler performance factor into the 'Compilers for Machine Learning' grading breakdown?

tinygrad plans a VS Code syntax-highlighting plugin for *.uir files

tinygrad’s UOp IR now renders in VIZ=1 on master

The case for tinygrad as ML hardware becomes more heterogeneous

A text-based representation for tinygrad's UOps in student compilers

Robot RAM needs for action and knowledge

University host sought for a winter-semester machine-learning compilers course

Qwen3.8-27B reportedly runs at about 90 tokens per second on a four-year-old AMD GPU over USB