ROCm 10.2 continues our work to make AI development easier on hardware you control—from workstations and local servers to the edge. This release brings:
• Validated vLLM and SGLang builds, with AITER included in vLLM images for Radeon and Ryzen AI
• Local scaling across up to eight GPUs in one system
• Faster storage-to-GPU transfers with hipFile and PyTorch
• Deployment planning with ROCm InferaSim
Swipe through the highlights, then read the blog: https://t.co/7cNmr9rgc9
More possibilities for local AI, across connected @Windows PCs.
AMD is collaborating with @Microsoft on the RDMA over USB4 developer preview for Windows, helping developers explore larger models across connected PCs.
"At AMD, we believe clustering is a critical part in a developer's experience to unlocking new possibilities for local AI and high-performance compute. We're excited to be working alongside Microsoft to support their new RDMA over USB4 driver to enhance the experience." - Andrej Zdravkovic, Senior Vice President and Chief Software Officer, AMD
Learn more: https://t.co/6ZLLwE066k
Unsloth Desktop now supports NPUs! NPUs use 2-4x less energy per token for local models vs GPUs!
We collabed with AMD to support the current Strix, Krackan and Gorgon devices, and future AMD devices with NPUs will also be supported out of the box!
Blog: https://t.co/ZpaBQWxlKQ
Unsloth Desktop. Meet the NPU.
AMD XDNA 2 NPUs are the world's first NPUs supported by @UnslothAI Desktop. Connect Hermes Agent, OpenClaw, Cline, VS Code and more through the Unsloth API.
100+ agentic apps. Now on your NPU.
https://t.co/MO83Qpi8uO
Choose your path at AMD AI DevDay in Paris on December 12, co-hosted by @joinstationf. Get hands-on building with agents and local models, or dive into AMD GPU architecture, programming, and optimization in the ROCm certification track. Register now for free! https://t.co/aQSABXU6Rw
Microsoft Ignite is a chance to get hands-on with your next AI project.
AMD learning labs help developers explore everything from model routing in the cloud to agents running on your PC. Here are some of the labs you can find at #MSIgnite:
→ Token Efficiency with vLLM Semantic Router
Build a single endpoint serving multiple models on AMD Instinct GPUs, then route each request to an appropriate model to improve performance, manage token costs and keep sensitive workloads on trusted local models.
→ Build an OpenClaw Multi-Agent Team Using Microsoft Foundry and AMD GPUs
Create and coordinate a multi-agent workflow using cloud endpoints and Microsoft Foundry on AMD GPU infrastructure.
→ Reinforcement Learning and Local Models on AMD Computers with Unsloth
Learn how to fine-tune and optimize open models locally on AMD systems for practical developer workflows.
→ Build On-Device Local Agents on AMD Ryzen AI PCs
Develop local agentic skills that run close to user data, highlighting responsive and privacy-aware AI experiences on AMD Ryzen AI PCs.
Explore the labs and more ways to build with AMD at @Microsoft Ignite:
https://t.co/OnW2bE8SDK
Can @PyTorch + ROCm run an LLM? Start with the stack: model, framework, ROCm software, driver, OS and supported AMD hardware.
This AMD AI Playbook takes you from setup to GPT-OSS-20B inference and a document summarizer.
Explore it and others: https://t.co/sAkJdO9uEJ
Build bigger with AMD Ryzen AI Halo + Strata.
Run larger, higher-quality Qwen and DeepSeek models locally on Windows and Linux, with a cloud-like experience that brings the latest community AI models directly to your PC. https://t.co/BFm9gGiRgA
Microsoft Ignite is a chance to get hands-on with your next AI project.
AMD learning labs help developers explore everything from model routing in the cloud to agents running on your PC. Here are some of the labs you can find at #MSIgnite:
→ Token Efficiency with vLLM Semantic Router
Build a single endpoint serving multiple models on AMD Instinct GPUs, then route each request to an appropriate model to improve performance, manage token costs and keep sensitive workloads on trusted local models.
→ Build an OpenClaw Multi-Agent Team Using Microsoft Foundry and AMD GPUs
Create and coordinate a multi-agent workflow using cloud endpoints and Microsoft Foundry on AMD GPU infrastructure.
→ Reinforcement Learning and Local Models on AMD Computers with Unsloth
Learn how to fine-tune and optimize open models locally on AMD systems for practical developer workflows.
→ Build On-Device Local Agents on AMD Ryzen AI PCs
Develop local agentic skills that run close to user data, highlighting responsive and privacy-aware AI experiences on AMD Ryzen AI PCs.
Explore the labs and more ways to build with AMD at @Microsoft Ignite:
https://t.co/OnW2bE8SDK
We’re officially 2 stops into the AI DevDay World Tour!
The journey is just getting started, and we’re excited to bring hands-on AI learning, building, and innovation to more cities around the world.
Want to join us at an upcoming stop? Paris, cohosted with @joinstationf, and Hong Kong registration is open!
Explore the other tour locations here: https://t.co/5VHh8daqjx
Hope to see you there for a hands-on workshop!
Complimentary Token Lab points are valid for 30 days, or until you use them up. How long they last depends on the model and your token usage—check the model catalogue for each model’s point cost per 1M tokens.
The rate limit is 20 requests per minute; requests above that will be rate limited. For a weekend hack, you’ll have the full weekend within the validity window, but how far the points go depends on what you build and which model you use.
Your next AI project starts with an API key.
Explore DeepSeek V4, Qwen3.8, Gemma 4, and more through Token Lab, powered by AMD GPUs.
Join the AMD AI Developer Program for Free API access, activate your complimentary Token Lab points, and start building.
Try Token Lab: https://t.co/XF8W2Bxsuj
Meet the AMD Developer Assistant 🤖
Need quick documentation, code examples, or troubleshooting help? Get instant answers pulled directly from curated AMD resources, including:
• AMD ROCm platform docs & blogs
• AMD Ryzen AI developer docs
• GPUOpen SDKs, manuals & tools
• AMD AI Playbooks... and more!
2 ways to use it:
• Use /ask in our Discord server: https://t.co/OiGVk8N1q8
• Chat with the bot in our forums: https://t.co/DTP3j31jgP
Jump in and try it out!
Build and run demanding AI workloads locally with AMD Ryzen AI Halo. More room for evolving models and agentic workflows, without tying your development to the cloud. Learn more about AMD Ryzen AI Halo processors: https://t.co/7Wory9tePq
Project Zenith is here!
Built with @Microsoft, it's now available on our new Ryzen AI Halo, powered by Gorgon Halo.
Less time setting up. More time building, experimenting, and bringing ideas to life.
Coming soon to more @AMD powered devices.
More freedom to build. More possibilities to unlock.
Building your AI skills should lead somewhere.
With the AMD AI Developer Program, earn points for AMD AI Academy courses, ROCm certification, hackathons and more. Level up and unlock AMD merch!
Join the AMD AI Developer Program to learn more: https://t.co/53qQxmQU9Y
#AMDevs
Less setup. More building.
@Microsoft's Project Zenith is now available on AMD Ryzen AI Halo, bringing a ready-to-code Windows 11 experience to local AI development.
Preconfigured dev tools. Guided AMD AI Playbooks. Your models, frameworks, and workflows.
Start building: https://t.co/xBDQljxwfP