# NVIDIA H100 Projects at AI Tinkerers

> Canonical HTML: https://aitinkerers.org/technologies/nvidia-h100
> Markdown URL: https://aitinkerers.org/technologies/nvidia-h100.md
> Technology record last updated: 2026-03-03T04:02:07Z
> Generated: 2026-09-22T23:45:34Z

The NVIDIA H100 Tensor Core GPU: Your order-of-magnitude leap for exascale AI and HPC workloads.

This is the Hopper architecture powerhouse, designed for massive scale. It features fourth-generation Tensor Cores and the dedicated Transformer Engine with FP8 precision, delivering up to 30X faster LLM inference over the previous generation. With up to 80GB of ultra-fast HBM3 memory and 900 GB/s NVLink interconnect, the H100 provides unmatched throughput for data center needs. We're talking secure, scalable compute: from small, partitioned Multi-Instance GPU (MIG) jobs to trillion-parameter models. This is the platform that accelerates your time-to-solution.

- Official technology site: https://www.nvidia.com/h100/
- Public AI Tinkerers demos and talks: 4
- Result page: 1 of 1

## Recent Public Talks and Demos

### [Real-Time Snooker Table Understanding with PyTorch, MobileNet-SSD, and Jetson Devices](https://hong-kong.aitinkerers.org/talks/rsvp_R8OKuhCvMWw)

This talk walks through how I built a full computer-vision pipeline that detects snooker balls, cues, cushions, pockets, and ball trajectories using PyTorch and MobileNet-SSD, deployed on Jetson Nano devices. The goal of the system was to convert raw video into structured table state: precise x/y coordinates of every ball, detection of ball disappearance, and automatic scoring. The project was developed in 2022 during COVID and required solving several real-world challenges: wide-angle camera distortion, severe dataset imbalance across ball colors, inconsistent bounding-box scales, and motion blur from fast-moving cues and balls. Models were trained on H100 GPUs and optimized to run efficiently on edge hardware (Jetson Nano) for low-latency inference. I will cover the dataset strategy, model architecture choices, augmentation methods, SSD tuning specifics, and the geometric logic that converts detections into scoring and gameplay interpretation.

- Event context: AI Tinkerers Hong Kong: Deep Dive Series — Technical Lessons from Shipping AI/ML/CV — 2025-12-18 — Hong Kong
- Public talk page: https://hong-kong.aitinkerers.org/talks/rsvp_R8OKuhCvMWw

### [How to Train an LLM-RecSys Hybrid for Steerable Recs](https://seattle.aitinkerers.org/talks/rsvp_9U-5XPnMAUg)

I'll be demoing how I finetuned Qwen3-8B to understand product IDs. The result is a language model that can converse in both English and item IDs, not with retrieval or other tools, but as a single, “bilingual” model where items (i.e., semantic IDs) are part of its vocabulary. Like a recommender model, it can recommend items given historical interactions. But the big surprise—and capability unlock—was when I found that I could simply chat with the model to steer its recommendations, and it could reason about its choices, offer explanations, and creatively name product bundles.

- Event context: AI Tinkerers Seattle September Meetup — September 30, 2025 — 2025-10-01 — Seattle
- Public talk page: https://seattle.aitinkerers.org/talks/rsvp_9U-5XPnMAUg

### [Secure AI Inference with Hardware Attestation](https://sf.aitinkerers.org/talks/rsvp_14eplt0Cd3w)

I'll demonstrate a working prototype of running an open-source LLM inside an AWS Nitro Enclave, showing how to achieve verifiable privacy for AI inference. The live demo will show: -Real-time hardware attestation of the Nitro Enclave -A working chat interface connecting to the enclave -(potentially) Privacy-preserving aggregated analytics using DAP (Distributed Aggregation Protocol), which is a multiparty computation protocol

- Event context: AI Tinkerers - San Francisco - November 2024 Meetup - Securing Agentic Systems — 2024-11-22 — San Francisco
- Public talk page: https://sf.aitinkerers.org/talks/rsvp_14eplt0Cd3w

### [Speedrunning NanoGPT training runs](https://portland.aitinkerers.org/talks/rsvp_rFDus1GRZYc)

NanoGPT is a small scale (124m parameter) transformer architecture initially made by Andrei Karpathy for learning how to build/train LLMs. There is an active, ongoing effort to integrate various architectural changes and optimizer improvements into NanoGPT in order to break training speed records. These speedruns are tracked by Keller Jordan (who started the speedrunning effort) here: https://github.com/KellerJordan/modded-nanogpt/tree/master. As of last week, with PyTorch 2.5 and the new Muon optimizer, NanoGPT speedruns have reached parity with LLM.c in training speeds for 120m scale models. My current experiment/attempt at speedrunning is taking Keller Jordan's speedrun and integrating differential attention layers (https://arxiv.org/abs/2410.05258) into it. DIfferential attention splits the attention heads into excitatory and inhibitory attention heads, allowing the attention layer to focus more clearly on context which it deems relevant. The paper showed a 25-30% improvement in loss for models of the same parameter count, so I am hoping that combining it with the Muon optimizer will break the latest speedrunning record. My fork is here: https://github.com/RyanPersson/modded-nanogpt/tree/differential-flash-attention (Differential-flash-attention branch, master is in sync with Keller Jordan's repo.) I've been training/testing on Lambda Cloud H100 instances. I got my varient running on a single H100 node last night, but it still segfaults on an 8xH100 DGX cluster. I am hoping to get that resolved and get it working by the meetup next Tuesday, but figured I would go ahead and put in a request to present because it'd be fun to talk about either way.

- Event context: AI Tinkerers Portland @Nedspace - October 29th 2024, 6pm — 2024-10-30 — Portland
- Public talk page: https://portland.aitinkerers.org/talks/rsvp_rFDus1GRZYc

## Related Technologies

- [PyTorch](https://aitinkerers.org/technologies/pytorch) ([Markdown](https://aitinkerers.org/technologies/pytorch.md)) — 273 public demos
- [AMD SEV-SNP](https://aitinkerers.org/technologies/amd-sev-snp) ([Markdown](https://aitinkerers.org/technologies/amd-sev-snp.md)) — 1 public demo
- [Apple Private Cloud Compute](https://aitinkerers.org/technologies/apple-private-cloud-compute) ([Markdown](https://aitinkerers.org/technologies/apple-private-cloud-compute.md)) — 1 public demo
- [AWS Nitro Enclaves](https://aitinkerers.org/technologies/aws-nitro-enclaves) ([Markdown](https://aitinkerers.org/technologies/aws-nitro-enclaves.md)) — 3 public demos
- [Distributed Aggregation Protocol](https://aitinkerers.org/technologies/distributed-aggregation-protocol) ([Markdown](https://aitinkerers.org/technologies/distributed-aggregation-protocol.md)) — 1 public demo
- [Gemini](https://aitinkerers.org/technologies/gemini) ([Markdown](https://aitinkerers.org/technologies/gemini.md)) — 188 public demos
- [GPT-4](https://aitinkerers.org/technologies/gpt-4) ([Markdown](https://aitinkerers.org/technologies/gpt-4.md)) — 529 public demos
- [Jetson Nano](https://aitinkerers.org/technologies/jetson-nano) ([Markdown](https://aitinkerers.org/technologies/jetson-nano.md)) — 2 public demos
- [LangChain](https://aitinkerers.org/technologies/langchain) ([Markdown](https://aitinkerers.org/technologies/langchain.md)) — 445 public demos
- [LLM](https://aitinkerers.org/technologies/llm) ([Markdown](https://aitinkerers.org/technologies/llm.md)) — 123 public demos
- [MobileNet-SSD](https://aitinkerers.org/technologies/mobilenet-ssd) ([Markdown](https://aitinkerers.org/technologies/mobilenet-ssd.md)) — 1 public demo
- [Muon optimizer](https://aitinkerers.org/technologies/muon-optimizer) ([Markdown](https://aitinkerers.org/technologies/muon-optimizer.md)) — 1 public demo
- [NanoGPT](https://aitinkerers.org/technologies/nanogpt) ([Markdown](https://aitinkerers.org/technologies/nanogpt.md)) — 4 public demos
- [Python](https://aitinkerers.org/technologies/python) ([Markdown](https://aitinkerers.org/technologies/python.md)) — 662 public demos