TurboGPT: train 22KiB transformer in 13s
Get picks like this daily. Today's launch numbers, the top launches, AI and tech news, and a daily opportunity pick, straight to your inbox.
Want to reach these readers? Sponsor an issue.
Inspired by minGPT for home experiments.Requires CUDA 13.4; build script is Windows only.
HN · launched Sep 29, 2026 · 56 upvotes (top 13%) · 0 views
Niche: transformer model development tools · 17 launches, 0 in the last 7 days · Explore this niche →
Category and tags
- Horizontal
- Model & infra
- Developer
- AI-native
- Hobby / open-source project
Visit website ↗ Source listing ↗
More like this: all 17 launches in transformer model development tools →
Could you build this?
Partial While basic transformer models are well documented, writing high-performance, low-level custom CUDA kernels that train a micro-transformer in seconds requires specialized GPU programming expertise.
What it would actually take: The project is implemented in C++/CUDA and compiled with MSVC/NVCC for Windows, optimizing memory layout, kernel fusions, and matrix multiplications directly without heavy PyTorch runtime overhead. The hard part is low-level GPU memory alignment, warp-level primitives, and numerical stability in tight CUDA kernels to achieve 13-second training times. It requires an engineer experienced in low-level CUDA optimization and computer architecture.
Discussion
8 comments analyzed.
Competitors mentioned: Andrej Karpathy's GPT tutorial
Concerns raised: Reinventing the wheel without learning value, Low-effort AI/vibe-coded project, Labeling model size by disk instead of parameter count, Missing hn1g.txt file
Competitors
Other products that read as similar to this one : 250 launches clear the similarity bar, closest 8 shown.
Attention rank: #42 of 251 (itself plus its competitors, highest first, normalized so YC and Product Hunt are compared fairly).
Launched 335 days after the earliest competitor.
- TRiP · hn · 2026-04-30 · 38 upvotes · similarity 0.48
- I built GPT from scratch to understand how it works · hn · 2026-01-14 · 7 upvotes · similarity 0.46
- TinyGPU · github · 2026-09-19 · 22 upvotes · similarity 0.45
- tiny-transformer-from-scratch · github · 2026-11-02 · 99 upvotes · similarity 0.44
- Llama 3.1 70B on a single RTX 3090 via NVMe-to-GPU bypassing the CPU · hn · 2026-02-21 · 395 upvotes · similarity 0.43
- RUDA · github · 2026-09-09 · 14 upvotes · similarity 0.42
- NamiGPT · github · 2026-10-28 · 161 upvotes · similarity 0.41
- Making a GBA game with GPT-6 Astra · hn · 2026-09-08 · 16 upvotes · similarity 0.41
The niche, labels, vibe code rating and competitor comparison on this page are analysis by Nicheloom. If you use them, please credit nicheloom.com. How you may use our data.