# GPU Projects at AI Tinkerers

> Canonical HTML: https://aitinkerers.org/technologies/gpu
> Markdown URL: https://aitinkerers.org/technologies/gpu.md
> Technology record last updated: 2026-02-25T06:43:59Z
> Generated: 2026-09-22T18:39:34Z

The Graphics Processing Unit (GPU): a massively parallel processor, purpose-built to accelerate computation for graphics rendering and general-purpose workloads (GPGPU).

A GPU is a specialized electronic circuit designed for concurrent, high-speed mathematical calculations: it excels at parallel processing, unlike a CPU’s serial approach. Modern units, like the NVIDIA H100 or AMD Instinct MI300X, feature thousands of cores and utilize high-bandwidth memory (HBM) to manage immense datasets efficiently. Initially focused on accelerating 3D graphics for gaming (e.g., the GeForce 256, 1999), the GPU’s architecture now dominates compute-intensive fields. Key applications include deep learning (AI/ML) model training, complex scientific simulations, and high-resolution video rendering, reducing processing time from hours to minutes.

- Official technology site: https://en.wikipedia.org/wiki/Graphics_processing_unit
- Public AI Tinkerers demos and talks: 12
- Result page: 1 of 1

## Recent Public Talks and Demos

### [Watch 1 hour highly techincal YouTubes in 5 minutes with AI!](https://seattle.aitinkerers.org/talks/rsvp_cjwx88z4AkE)

AG is an agent that watches YouTube podcasts for you so you know which ones to really dig into. With AG, see in 5 minutes a summary of the YouTube, key quotes, see key blackboard / slide / code sections, jump around key passages, and decide if you should spend the full time on the video. Break down highly techincal episodes from Dwarkesh, Lenny, AI Engineer, and more!

- Event context: AI Tinkerers Seattle Summer Bash — 2026-07-29 — Seattle
- Public talk page: https://seattle.aitinkerers.org/talks/rsvp_cjwx88z4AkE

### [stress-testing autonomous research agents](https://zurich.aitinkerers.org/talks/rsvp_N7RI28XXj20)

I built a minimal autoresearch loop for coding agents, and I test the agent-environment behaviour during standard and broken evaluations. I will show how the system is structured as a research sandbox. I will walk through the live workflow, and I will show when and how the different evals I tested changed the agent-environment behaviour.

- Event context: AI Tinkerers Zürich: July Build Night with Swisscom Ventures — 2026-07-01 — Zürich
- Public talk page: https://zurich.aitinkerers.org/talks/rsvp_N7RI28XXj20

### [From Laptop to Supercluster: The New Era of Personal AI Supercomputing](https://paris.aitinkerers.org/talks/rsvp_tpKMaxLqXkQ)

AI development is shifting rapidly from centralized cloud infrastructure to powerful local AI systems that enable developers to build, test, and run large models directly on their desks. Lenovo’s new AI workstation portfolio powered by NVIDIA Blackwell and Grace-Blackwell architectures introduces a new class of systems—from compact personal AI appliances to multi-GPU developer workstations capable of running hundreds-billion-parameter models locally. This talk explores how AI development is moving closer to the developer, enabling faster iteration, lower cloud costs, and new experimentation workflows. We will walk through the new Lenovo AI workstation stack, including the ThinkStation PGX (Grace-Blackwell GB10 superchip) and Blackwell GPU developer workstations, and explain how these systems scale from personal AI experimentation to enterprise-grade model development. Questions the Talk Will Answer What does “personal AI supercomputing” actually mean for developers? How large of an AI model can realistically run locally on a workstation today? When should developers use local AI vs cloud GPU clusters? How do systems like PGX, P3, P5, P7, and PX map to different AI workloads? What does the Grace-Blackwell architecture change in AI workstation design? How can developers prototype and iterate faster using local AI hardware?

- Event context: High-Performance Local AI Development: Kick-off ThinkStation PGX — 2026-03-17 — Paris
- Public talk page: https://paris.aitinkerers.org/talks/rsvp_tpKMaxLqXkQ

### [Not all data is equal before the law](https://hong-kong.aitinkerers.org/talks/rsvp_irrzcSA5FoU)

The premise of my business, Counsel Intelligence, is that legal training data currently available in public sources (regardless of copyright and licensing) is insufficient to train LLMs in become legal experts. This is different to hallucination, since I believe this is a problem that arises from the shallowness of legal data, rather than a probabilistic cause.

- Event context: AI Tinkerers - Hong Kong Meetup (July) - Meetup with Alibaba Cloud — 2025-07-31 — Hong Kong
- Public talk page: https://hong-kong.aitinkerers.org/talks/rsvp_irrzcSA5FoU

### [Dispatcher (Autonomous Drone System)](https://palo-alto.aitinkerers.org/talks/rsvp_-xK6whcJBgQ)

I made an Aerial Intelligence system, and I call it Dispatcher. Jarvis on a drone, skynet, or compute token eater, its basically a drone you don't have to fly because it understands and takes action based on the world around it. Watch it respond to a structure fire (DFR 3.0) https://lnkd.in/g-CFQ_2S Watch me "fly" it while I drive. If you have feedback or just want to "fly" it, just reach out!

- Event context: AI Tinkerers - Palo Alto - November 2024 Meetup — 2024-11-21 — Palo Alto
- Public talk page: https://palo-alto.aitinkerers.org/talks/rsvp_-xK6whcJBgQ

### [Computer Vision Semantic Segmentation](https://bogota.aitinkerers.org/talks/rsvp_hjRtInuCG3s)

El proyecto consiste en entrenar y desplegar un modelo de segmentación semántica utilizando Databricks y TensorFlow, con una base de datos de imágenes tomadas por drones. Se entrenará un modelo como U-Net para segmentar objetos en las imágenes, aprovechando las capacidades distribuidas de Databricks para el procesamiento de datos y el entrenamiento eficiente con GPU. El modelo será desplegado mediante el servicio de *model serving* y endpoints de Databricks, permitiendo realizar inferencias en tiempo real.

- Event context: AI Tinkerers Bogotá No. 5 — 2024-09-26 — Bogotá
- Public talk page: https://bogota.aitinkerers.org/talks/rsvp_hjRtInuCG3s

### [Running llama3 locally without a GPU](https://dubai.aitinkerers.org/talks/rsvp_2u5-OwB5LoU)

I plan to show how my npu laptop can run llama3 locally, and explain limitations and opportunities Structure - brief explanation of what I am going to demo - Demo with llama3 - Demo with a different model, still have to decide which one. To make the demo more engaging I could make a poll in the room to see which model the audience would like to test

- Event context: AI Tinkerers - Dubai Inaugural Meetup (September) — 2024-09-07 — Dubai
- Public talk page: https://dubai.aitinkerers.org/talks/rsvp_2u5-OwB5LoU

### [Korvus - The power of GPU enabled Postgres](https://portland.aitinkerers.org/talks/rsvp_XbyaxO6ZwEM)

This is a brief talk and demo of Korvus, a search SDK that unifies the entire RAG pipeline in a single database query, and a little bit about pgml, an open source postgres extension that brings machine learning into postgres databases.

- Event context: AI Tinkerers Portland Inaugural Meetup - July — 2024-07-24 — Portland
- Public talk page: https://portland.aitinkerers.org/talks/rsvp_XbyaxO6ZwEM

### [Training Diffusion Transformers for Style Transfer (for the GPU poor)](https://la.aitinkerers.org/talks/rsvp_z4k8pNn_y6w)

We trained diffusion transformers to generate images from text in the style of the Simpsons and Legos on a single commodity GPU.

- Event context: May 21st - LA AI Tinkerers Meetup & Demos — 2024-05-22 — Los Angeles
- Public talk page: https://la.aitinkerers.org/talks/rsvp_z4k8pNn_y6w

### [streamdiffusion in flair ai realtime feature](https://sf.aitinkerers.org/talks/rsvp_iMQqQpJR_RU)

accelerate image generation by pipelining your tensors with streamdiffusion! we optimize our real-time product photoshoot preview by basically storing intermediate tensors in the diffusion process so that we effectively are batching our inputs even when only a single user is on a gpu machine at a time.

- Event context: AI Tinkerers - San Francisco - April 2024 Meetup — 2024-04-30 — San Francisco
- Public talk page: https://sf.aitinkerers.org/talks/rsvp_iMQqQpJR_RU

### [An open-source engine for provisioning AI infrastructure](https://berlin.aitinkerers.org/talks/rsvp_2jMdD8uK-3s)

Enterprises are increasingly focused on training and deploying Generation AI (Gen AI) models within their infrastructure. However, the conventional Kubernetes stack, while widely used, is not specifically optimized for the unique demands of the AI era, presenting several limitations. In response, dstack emerges as a new generation of infrastructure automation tailored to meet the challenges and opportunities of the AI era.

- Event context: AI Tinkerers Berlin - March 21 — 2024-03-21 — Berlin
- Public talk page: https://berlin.aitinkerers.org/talks/rsvp_2jMdD8uK-3s

### [An open-source toolkit to run GPU workloads on any cloud](https://munich.aitinkerers.org/talks/rsvp_gNGUWrRVWTs)

Developing, training, and deploying LLMs (Large Language Models) can be a challenging task. Managing the infrastructure can be a lot of pain, even when you're using a major cloud provider. It can be even more challenging if you're using alternative GPU providers. However, dstack, an open-source orchestration engine, can significantly simplify running GPU workloads across any cloud GPU provider.

- Event context: AI Tinkerers Munich - January 18 — 2024-01-18 — Munich
- Public talk page: https://munich.aitinkerers.org/talks/rsvp_gNGUWrRVWTs

## Related Technologies

- [Codex](https://aitinkerers.org/technologies/codex) ([Markdown](https://aitinkerers.org/technologies/codex.md)) — 44 public demos
- [dstack](https://aitinkerers.org/technologies/dstack) ([Markdown](https://aitinkerers.org/technologies/dstack.md)) — 2 public demos
- [NVIDIA](https://aitinkerers.org/technologies/nvidia) ([Markdown](https://aitinkerers.org/technologies/nvidia.md)) — 7 public demos
- [OpenAI](https://aitinkerers.org/technologies/openai) ([Markdown](https://aitinkerers.org/technologies/openai.md)) — 112 public demos
- [AGENTS](https://aitinkerers.org/technologies/agents) ([Markdown](https://aitinkerers.org/technologies/agents.md)) — 20 public demos
- [AI infrastructure](https://aitinkerers.org/technologies/ai-infrastructure) ([Markdown](https://aitinkerers.org/technologies/ai-infrastructure.md)) — 2 public demos
- [AMD EPYC](https://aitinkerers.org/technologies/amd-epyc) ([Markdown](https://aitinkerers.org/technologies/amd-epyc.md)) — 2 public demos
- [Anthropic](https://aitinkerers.org/technologies/anthropic) ([Markdown](https://aitinkerers.org/technologies/anthropic.md)) — 36 public demos
- [BERT](https://aitinkerers.org/technologies/bert) ([Markdown](https://aitinkerers.org/technologies/bert.md)) — 179 public demos
- [BLOOM](https://aitinkerers.org/technologies/bloom) ([Markdown](https://aitinkerers.org/technologies/bloom.md)) — 115 public demos
- [Claude](https://aitinkerers.org/technologies/claude) ([Markdown](https://aitinkerers.org/technologies/claude.md)) — 174 public demos
- [Cloud computing](https://aitinkerers.org/technologies/cloud-computing) ([Markdown](https://aitinkerers.org/technologies/cloud-computing.md)) — 3 public demos
- [Codebase](https://aitinkerers.org/technologies/codebase) ([Markdown](https://aitinkerers.org/technologies/codebase.md)) — 2 public demos
- [DALL-E 2](https://aitinkerers.org/technologies/dall-e-2) ([Markdown](https://aitinkerers.org/technologies/dall-e-2.md)) — 7 public demos
- [DALL-E 3](https://aitinkerers.org/technologies/dall-e-3) ([Markdown](https://aitinkerers.org/technologies/dall-e-3.md)) — 12 public demos
- [Databricks](https://aitinkerers.org/technologies/databricks) ([Markdown](https://aitinkerers.org/technologies/databricks.md)) — 5 public demos
- [Databricks Model Serving](https://aitinkerers.org/technologies/databricks-model-serving) ([Markdown](https://aitinkerers.org/technologies/databricks-model-serving.md)) — 1 public demo
- [DFR 3](https://aitinkerers.org/technologies/dfr-3) ([Markdown](https://aitinkerers.org/technologies/dfr-3.md)) — 1 public demo