The Future Technologies Initiative (FTI) at OCP has been a great resource to incubate emerging technologies, architectures and directions. Recent years, one such initiative has been the AI HW/SW Co-Design effort focusing on both the emerging technology used for AI, as well as the ecosystem required for efficient resource management, both the discovery as well as the composition and management to meet the service level objectives of the AI workloads. This year has focused on a Framework for a polymorphic architecture, enabling composition of an AI ecosystem of heterogeneous accelerators.
At OCP Global Summit next week, the framework of the AI HW/SW Co-Design working group will present, on Wednesday October 16th, a summary session in the morning and a poster session in the afternoon. Please join the AI co-design workstream update led by:
- Weifeng Zhang, Corporate VP, Lenovo Research
- Jack Harwood, Sr. Distinguished Engineer, Marvell
In addition to the work completed over the past year, the Future Technologies Symposium, on October 16th, 12:30-5pm, will bring together industry leading technologists and researchers highlighting the directions of the AI ecosystem, specifically programming models, communication, data services as well as both hardware for AI and AI for HW design. Leading the Symposium is a set of presentations demonstrating how to “Leverage the Full Ecosystem to Enable AI”. For this we will hear from:
- Amir Yazdanbakhsh, Research Scientist, Google DeepMind: “From Compilers to Code Whisperers: Can Generative AI Solve the Optimization Puzzle”
- Tushar Krishna, Professor Georgia Tech: “Towards a Standardized Representation for Deep Learning Collective Algorithms”
- CJ Newburn, Distinguished Engineer Nvidia: “Making a GPU into a data access engine”
- Mike Schulte, Sr Fellow AMD: “CPU/GPU HW Optimizations for AI”
Following the presentations there will be two panels, further diving into the Communication and Hardware topics. The first Panel will be hosted by Danny Moore, Director at AMD, covering “Connecting Intelligence: Interconnects and AI”. The Symposium will be joined by industry leaders:
- Gilad Shainer: SVP of Networking at Nvidia
- Jai Kumar: Distinguished Engineer at Broadcom
- Kurtis Bowman: Director AMD, Consortium Lead for UAlink
- Taeksang Song: VP at Samsung
The panel will offer candid opinions on the directions of communication technology, protocols and directions for the future of the AI ecosystem. You will hear how communication may change, address the issues in the AI ecosystem and data center, and proposals for future innovation, standardization and collaboration.
The second panel will provide a focus on “AI for AI System Design” and will be moderated by Dr. Shobha Vasudevan, an AI/ML Research Scientist at Google, and Professor at UIUC. Joining her on the panel to discuss how AI can be used to design hardware today, as well as tackling the complexities of future designs, will be:
- Durgesh Srivastava: CTO at MIPS
- Ann Wu: Founder of Silimate
- Hans Bouwmeester: COO at Primis AI
- Geetha Rangarajan: Synopsys
With the increasing dimensions to AI chip design, adding chiplets, 3D, thermal, etc. hardware design is quickly becoming an increasingly complex system. This panel will offer insight into the direction of the tools available for designing the complex AI chips as well as where the problem space could lead in design and how these tools may head to address the problem space.
These insightful talks and panels will provide a full stack lens into the work required for building AI ecosystems as well as optimizing the ecosystem for the emerging AI workloads. With this, we can leverage these insights to set our sights on work to be tackled in the OCP AI workstreams and the collaboration required to meet the goals created for future work and deliverables.
You can check out the full schedule and register here: https://bit.ly/3xO1X76