Top Builders

Explore the top contributors showcasing the highest number of app submissions within our community.

AMD

Advanced Micro Devices (AMD) is a global semiconductor company that designs CPUs, GPUs, and accelerators for data centers, PCs, and embedded systems. Founded in 1969, AMD has built a significant AI infrastructure position through its AMD Instinct GPU line and the open-source ROCm software stack, which together serve as an alternative to proprietary GPU ecosystems for large-scale AI development.

General
CompanyAdvanced Micro Devices, Inc.
Founded1969
HeadquartersSanta Clara, California, USA
Websiteamd.com
DocumentationROCm Docs
GitHubgithub.com/ROCm
Developer HubAMD ROCm Developer Hub
TypeSemiconductor / AI Infrastructure

Start building with AMD products

AMD provides cloud-based GPU access, open-source software tooling, and developer resources for building AI applications at scale. Whether you are training a custom model, running large-scale inference, or benchmarking AI workloads, AMD's infrastructure stack gives you the compute and software you need without proprietary lock-in. Explore what the community has built on AMD by checking out AMD Use Cases and Applications.

Core Products

AMD Instinct GPU Accelerators

The AMD Instinct series are data center GPUs built for AI training and inference at scale. The MI300X is based on the CDNA 3 architecture and supports up to 192GB of HBM3 memory, making it well-suited for large language model inference where memory capacity is a bottleneck. The MI325X extends this to 288GB of HBM3E memory. Seven of the ten largest model builders and AI companies, including Meta, OpenAI, Microsoft, and xAI, run production workloads on Instinct GPUs.

ROCm (Radeon Open Compute)

ROCm is AMD's open-source software platform for GPU-accelerated computing. It supports HIP, OpenCL, and OpenMP programming interfaces and integrates with major ML frameworks including PyTorch, TensorFlow, and JAX. ROCm 7 is the current version, engineered for generative AI and HPC workloads with expanded hardware compatibility and new development tools.

For framework support, installation guides, and libraries, see our ROCm tech page.

HIP SDK

The AMD HIP (Heterogeneous-compute Interface for Portability) SDK allows developers to write GPU-accelerated code that runs on AMD hardware. HIP code is also designed to be portable to CUDA, lowering the barrier for developers migrating workloads from other GPU platforms.

AMD Developer Cloud

AMD provides a cloud environment where developers can access AMD Instinct GPU hardware for testing and benchmarking, along with free credits, training materials, and community support.

For setup details, credit access, and tutorials, see our AMD Developer Cloud tech page.

Developer Resources

AMD's open-source developer ecosystem is built around ROCm, with documentation, libraries, and tooling available for AI and HPC workloads on AMD hardware.

Key Features

Open-source software stack ROCm is fully open-source under the MIT and Apache 2.0 licenses, giving developers full visibility into the toolchain and the ability to contribute upstream.

Large memory capacity The MI300X provides up to 192GB of HBM3 memory per GPU, enabling inference of very large models (70B+ parameter) on a single accelerator without model parallelism.

Framework compatibility ROCm supports PyTorch, TensorFlow, JAX, and ONNX Runtime, allowing most standard AI training and inference pipelines to run without significant modification.

HIP portability HIP code compiles for both AMD and NVIDIA hardware, reducing the cost of maintaining GPU-specific codebases across infrastructure environments.

Use Cases

Large language model inference The high HBM capacity of AMD Instinct GPUs makes them a practical choice for serving large models where VRAM is the primary constraint.

AI model training Teams training custom models at scale use AMD Instinct GPUs through cloud providers and on-premise clusters as a cost-competitive alternative to other data center GPU options.

HPC workloads ROCm's support for scientific computing libraries makes AMD hardware a common choice for high-performance computing in research and enterprise environments.

Hackathon and prototyping AMD provides cloud access and credits for developers building AI prototypes, making it possible to test workloads on AMD hardware without upfront hardware costs. Explore upcoming AI hackathons that use AMD infrastructure.

Edit on GitHub

amd AI Technologies Hackathon projects

Discover innovative solutions crafted with amd AI Technologies, developed by our community members during our engaging hackathons.

GarudaLinux

GarudaLinux

Garuda Linux is a rolling-release Linux distribution built on top of Arch Linux, designed to be more user-friendly and visually polished than vanilla Arch while retaining access to the massive Arch User Repository (AUR) for software installation. It ships with a customized KDE Plasma desktop by default (though other desktop environments like GNOME, Xfce, i3, Sway, and others are also available as separate editions), featuring a distinctive, futuristic aesthetic with translucent windows, custom themes, and vibrant color schemes.Key technical features include: BTRFS filesystem by default, paired with Snapper for automatic system snapshots — meaning if an update breaks your system, you can boot into a previous working snapshot straight from the bootloader (via Garuda's "Assistant" tool).

Simple Request Router

Simple Request Router

This project delivers an intelligent routing agent engineered to balance strict enterprise accuracy requirements against cloud API token consumption. Built to comply with an evaluation environment restricted to a 4 GB RAM and 2 vCPU footprint, the system relies on an intentional two-step isolation architecture that separates intent classification from task execution. At the core of the local optimization strategy is an ultra-lightweight, 4-bit quantized Gemma 4 (Effective 2B) parameter model running locally via llama-cpp-python. When an evaluation payload enters the pipeline, the local model acts strictly as a zero-token-cost binary classification gate. It evaluates incoming prompts across eight distinct capability domains. Direct, low-complexity natural language tasks - such as sentiment analysis, named entity recognition, basic definitions, and text summarization - are processed and answered entirely in-house, incurring a total external token cost of zero. Conversely, when the local gate detects complex algorithmic patterns, multi-step mathematical reasoning, code debugging specs, or deductive logic puzzles, it flags the task as COMPLEX and routes it to the premium Fireworks AI cloud infrastructure. By isolating the routing logic and utilizing defensive system constraints like strict stop-token structures and max-token caps, this architecture minimizes proxy token overhead while ensuring maximum accuracy for more demanding requests.

Simple Request Router

Simple Request Router

This project delivers an intelligent routing agent engineered to balance strict enterprise accuracy requirements against cloud API token consumption. Built to comply with an evaluation environment restricted to a 4 GB RAM and 2 vCPU footprint, the system relies on an intentional two-step isolation architecture that separates intent classification from task execution. At the core of the local optimization strategy is an ultra-lightweight, 4-bit quantized Gemma 4 (Effective 2B) parameter model running locally via llama-cpp-python. When an evaluation payload enters the pipeline, the local model acts strictly as a zero-token-cost binary classification gate. It evaluates incoming prompts across eight distinct capability domains. Direct, low-complexity natural language tasks - such as sentiment analysis, named entity recognition, basic definitions, and text summarization - are processed and answered entirely in-house, incurring a total external token cost of zero. Conversely, when the local gate detects complex algorithmic patterns, multi-step mathematical reasoning, code debugging specs, or deductive logic puzzles, it flags the task as COMPLEX and routes it to the premium Fireworks AI cloud infrastructure. By isolating the routing logic and utilizing defensive system constraints like strict stop-token structures and max-token caps, this architecture minimizes proxy token overhead while ensuring maximum accuracy for more demanding requests.

FusionFinance

FusionFinance

In this project, I used both LLMs and ML trading algorithms. The reason for this is that while some pure-LLM frameworks can achieve a high level of return, such as the TradingAgents framework, this is highly volatile. Given that the LLMs were predicting periods of time that were outside of their knowledge cutoff, the Sharpe ratio was only around 0.27, much lower than the S&P 500. On the other hand, ML algorithms, which are widely used by quant firms, seem more reliable, but they simply cannot make qualitative inferences, such as looking for signals like sentiment. So, I brought the two together in layers, ultimately beating the S&P 500 (in a stable and reproducible manner).

Simple Video captioner

Simple Video captioner

A two-stage video captioning agent, built reliability-first. Stage one samples keyframes from each clip and sends them to a vision-language model (Kimi K2.6 on Fireworks), which returns a dense, factual description — this is what the accuracy score is made of. Stage two turns that description into the four requested styles as a single JSON object — this is what the style score is made of. Splitting them means a caption can only be as accurate as what was actually seen, and the pressure to be funny never contaminates the grounding. Everything underneath is about not failing. A complete, valid results.json is written before the first network call, then each clip is atomically replaced as it finishes, so a crash or a hang still leaves a scoreable file behind. The agent enforces its own 540-second deadline and exits cleanly rather than waiting to be killed. Each stage walks a fallback ladder — API, then local model, then template — so no single failure is fatal. In testing, a real local-model outage still produced twelve valid captions with 292 seconds to spare. The local model is the part I most wanted to build. I distilled the teacher's style behaviour into Gemma 4 with LoRA on teacher-labelled examples, merged the adapter, and quantized it to a Q4_K_M GGUF — 5.3 GB, 2.86x smaller than bf16 — served by llama-server on CPU inside the container, no GPU required. It styles a clip in about 30 seconds on two cores. It ships as the fallback rather than the primary, because I measured instead of assuming. A blind, order-swapped LLM judge (n=19, minimum detectable effect +/-0.44) found the LoRA bought style (+0.61) but cost accuracy (-0.29), and the teacher still beat the student 10-3 on swap-surviving verdicts. The same judge found Q4_K_M statistically indistinguishable from bf16 — 4-bit quantization was free. So the distilled model earns its place by surviving an outage, not by going first.

SOAP Copilot: AI Clinical Scribe on AMD

SOAP Copilot: AI Clinical Scribe on AMD

Physicians spend roughly two hours a day on clinical documentation, time pulled directly away from patient care, and a well-documented driver of burnout that costs the US healthcare system an estimated $8.3B a year in lost productivity. SOAP Copilot is a multi-agent AI system that turns a raw doctor-patient conversation into complete, structured clinical documentation in under 30 seconds. Three specialized agents, all built on Llama 3.3 70B Instruct, handle the pipeline in sequence: a SOAP Generator produces a structured Subjective/Objective/Assessment/Plan note, an ICD Coder extracts diagnosis codes with confidence scores, and a Summary Writer rewrites the note as a plain-language patient summary. The entire stack runs on open-source models, so no PHI ever needs to leave the user's own infrastructure. The project was originally prototyped and LoRA fine-tuned on a single AMD MI300X (192GB HBM3) via AMD Developer Cloud, running vLLM on ROCm 7.2. That memory headroom is what makes serving a 70B-parameter model in BF16 precision practical on one chip, something no consumer GPU and few cloud GPUs can do. For this submission, production inference is served through Fireworks AI, which itself runs on AMD Instinct hardware, so the project demonstrates the AMD AI stack end to end: self-managed fine-tuning through to hosted inference. Roadmap items include a distilled specialist model for lower-cost deployment, retrieval-augmented generation over clinical guidelines using the Qdrant vector database already wired into the architecture, and an on-premise, HIPAA-compliant deployment path for health systems that cannot send patient data to any external API.