GPU by Hand ✍️ Together
Keynote Lecture Sponsored by Together AI
“Dad, why didn’t you buy a ton of NVIDIA stock and become a millionaire? Didn’t you work on AI?”
My son hit me with that one recently. Ouch! He asked me the same question about Bitcoins last year. 🤣
No, I didn’t retire by owning $NVDA. But I do own something that I chose: The knowledge of how GPUs actually work — threads, warps, blocks, CUDA.
Next week I am going to share that knowledge in my keynote lecture.
Title: GPU by hand ✍️ Together
Date: 10/30
Time: 10am (Pacific Time)
Register: https://luma.com/khuz5upe
The leap from Hopper to Blackwell isn’t just a story of “faster GPUs” — it’s a fundamental rethinking of how every layer of the stack is organized to unlock the next frontier of AI scale. And as typical in an AI by Hand lecture, instead of focusing on benchmarks from the outside, we will dive inside the black box — unpacking the math behind these advances and making the underlying mechanisms accessible to you.
Outline
Primer: Grid -> Block -> Warp -> Thread
Quantization: FP8 -> MXFP4
Scaling: Layer -> Micro-Tensor
Packaging: Single-Die -> Dual-Die
Communication: In-GPU -> In-switch reduction
In the end…
do I want to be remembered as a lucky dad,
or a devoted dad who helped him understand the world he’s growing into?
This keynote lecture is sponsored by Together AI.
RSVPd, excited for the session.
It’s lame that you need a work email to register…