The Open Ecosystem for the Agentic Age
summary
AMD unveils an open AI stack centered on Rockom 7 that adds distributed inference, training with data/pipeline/tensor parallelism, and MI350/MI355 series support; Rockom runs about 1.88M Hugging Face models out of the box and ships feature updates every two weeks. The stack ties Pensando DPUs, EPYC CPUs, and Instinct GPUs into an enterprise path with Kubernetes GPU operator, Triton and JAX performance work, and claims MI355X 1.3x speed vs B200 on DeepSeek R1 FB8 — distributed decode/prefill pools reduce inference bottlenecks.
AI-written summary · about
chapters (11)
- 0:00Open ecosystems and Agentic AI
- 1:27Open ecosystems drive AI innovation
- 2:11Agentic AI rise and impact
- 3:14Agentic AI systems under the hood
- 4:37AMD AI platform for Agentic AI
- 5:41AMD open ecosystem investments
- 7:28AMD Rockom AI stack deep dive
- 9:13Rockom training and open-source leadership
- 10:38Distributed inference and local development
- 12:41Enterprise adoption with Rockom
- 14:04Future of AI and collaboration
related talks
23mKeynote
11mKeynote: Open Source Stack Empowers the AI Era
38mIntroducing Open Platform for Enterprise AI
03mKeynote: AMD & PyTorch: A Powerful Combination for Generative AI
29mOpen Source Sovereign AI: Building Enterprise GenAI Platforms for Control & Comp...
15mKeynote: Fujitsu's Open Source Vision and Initiatives
13mEmpowering Developers through Open-Source AI
10mCharacterizing AI inferencing stacks