Breaking the limits of x86
summary
Builds a cache-coherent fabric that maps DRAM from many motherboards into one global address space using a custom silicon 6-port chip and a kernel driver that extends APIC interrupt space into a 32-bit Linux IRQ space. Ships a global hardware timer, an ARM-based manager appliance for PXE/TFTP, and a JavaScript wiring tool for topology and cabling. Demonstrations include 34 TFLOPS from seven AMD Opteron nodes and a 5,000-core deployment achieving >10 TB/s STREAM bandwidth; firmware and drivers are open source on GitHub.
AI-written summary · about
chapters (6)
- 0:00Start
- 2:18How is this done?
- 3:23Scalable routing
- 6:09Interrupt delivery teardown
- 14:42Configuration
- 21:29Questions and links
related talks
31mUnshackling Memory!
26mBuilding Cache Coherent, Heterogeneous-ISA Processors Using P-Mesh
45mBeyond Machine Boundaries: Experience with Memory Disaggregation
22mAn Open and Coherent Memory Centric Architecture Enabled by RISC V
30mIntegrating Gen-Z in Server-Class RISC-V Processors
11mXerxes: Extensive Exploration of Scalable Hardware Systems with CXL-Based Simulation...
23mTileLink: A Free And Open Source, High Performance Scalable Cache Coherent Fabric Designed...
12mDirect Access, High-Performance Memory Disaggregation with DirectCXL
