the tiny corp
@__tinygrad__
We make tinygrad; sell tinybox for the GPU middle class.
Our mission is to commoditize the petaflop.
211 Following 84.9K Followers
@johnrobinsn This is the plan. I suspect a lot of the funding and effort for this can be traced back to Anthropic and EA aligned orgs too. They see you having sovereign compute as a major threat to their plans for world domination.
Show more
To the massive dismay of those looking to control compute, it's way too late. CPUs and GPUs are everywhere. Models will get smarter, algorithms will get better, and bottlenecks we used to believe in will be overcome. You will lose and the Internet will remember what you did.
Show more
The future depends on the direction this trend takes.
Meet the new tinybox. Configurable. Starting at $7,000.
We're testing a prototype
By the end of week 3, your compiler will be able to run uir programs like this. Quiz question: what's the PyTorch equivalent?
Compilers for Machine Learning
Carnegie Mellon University
Spring 2027 Semester
In person, Pittsburgh
10-411/711, MWF 10-10:50
139-seat classroom reserved
Lectures not posted online
Don't miss registration!
Show more
I'm not sure why every SoC with an NPU isn't shipping tinygrad, it's way better than the vendor libraries. Add support yourself, or contract us for a fast port. SNPE and RKNN shouldn't be things.
We will also ship a *.uir syntax highlighting plugin for VSCode. Before tinygrad is 1.0, we need a stable on disk UOp representation.
tinygrad's uir (UOp IR) is now rendered in VIZ=1 on master. You can click the nodes in the graph or the variables in the uir view and they will sync.
There will be a 10 question quiz posted on the class website. If you don't honestly get at least 7/10, you will fail the course.
As ML hardware becomes more heterogeneous, the case for tinygrad will be more and more obvious.
TinyStories-15M on a $60 Google Coral: all 6 transformer layers on the chip, 176 tok/s at batch 1.
An AI agent got the chip and zero docs. It wrote the driver, decoded the instruction set and built a @__tinygrad__ backend. tinygrad is small enough for an agent to understand.
Show more
Anyone at Huawei want to give us two Ascend 950 boxes in exchange for a tinygrad port?
Despite record usage of the models, ZAI stock is down over 2x from its peak. The Chinese markets have a real understanding of the economic impact of AI, the US ones do not. It's very important that the US government doesn't bail any of this out.
Show more
@chamath The cyber incidents weren't enough for them to get the regulatory capture they wanted. They need to make a bio incident next. I hope our government is on top of this, their ideology has no limits to what they will do for "good"
Show more
Overall, a very tasteful chip launch by OpenAI. It takes the best of both GPUs and TPUs. I wish they were for sale, but we'll all get the Chinese knockoffs in 3 years. I can't wait until China starts spamming fab capacity.
Show more
We merged some high performance kernels for RDNA3 Qwen 3.8 27B. Written in the tinygrad kernel language, which are the same UOps with memory movement specified. 46 tok/s on 7900XTX, no spec decode.
Show more
I can't believe Qualcomm's market cap is only 168.79B. Does anyone know if a leveraged buyout is possible? With new leadership that actually wants to sell chips it could easily 5x, we'll get a Qualcomm chip in every robot. And save money by firing the whole 80s era sales org.
Show more
@IanCutress Posting FLOPS, GB, GB/s, and W is not a meme dump, it's what every single respectable chip maker does. Compare the recent Cerebras press to the Etched one. Etched posted a video where they put an empty wooden box in a truck and said don't worry our lead investor thinks it's good.
Show more
Chestnut purchased, I'll shoot at the 3090
