Hi everyone, thank you for today’s call (2025.02.13) and special thanks to @icarus ! This is the generated summary (short version, please find the long version here ):
The discussion focused on upgrading Internet Computer (ICP) node hardware from Gen1 to Gen3, emphasizing faster AMD EPYC processors, PCIe 5, and improved memory bandwidth for enhanced AI processing. It covered GPU integration, including AMD Instinct and Nvidia options, and the concept of AI subnets for high-performance tasks. The impact of mixing hardware generations on ICP subnets was considered, along with an RFP announcement for AI agent frameworks. Other topics included historical blockchain data access, chain fusion, and community engagement through events and technical discussions. The call provided a deep technical dive into optimizing ICP’s infrastructure for AI advancements.
Links shared during the call:
- RFP for ICP AI Agents: RFP-9: ICP as Infra for AI Agent Frameworks · Issue #62 · dfinity/grant-rfps · GitHub
- Decentralized AI overview | Internet Computer
- Node Provider Machine Hardware Guide - Internet Computer Wiki
- DRAFT Motion Proposal: New Hardware specification and remuneration for IC nodes - #14 by garym
- Gen 1 node example: https://www.supermicro.com/Aplus/system/1U/1023/AS-1023US-TR4.cfm
- Gen 3??? exampe server: XV23-ZX0-AAJ1 (rev. 3.x) | Rack Servers - GIGABYTE Global
- https://www.amd.com/en/products/processors/server/epyc/ai.html
- https://www.amd.com/en/solutions/ai.html
- NVIDIA Data Center GPU Resource Center
- Wormhole™
- d-Matrix Corsair AI Platform | In-Memory Computing for AI
- https://www.esperanto.ai/products/
- GPU Database | TechPowerUp
- CPU Database | TechPowerUp
- Sharing someone’s X thread where they built a server for CPU inferencing DeepSeek 671B model at FP8. 6-8 tokens per second though: Thread by @carrigmat on Thread Reader App – Thread Reader App