Networking-Native RISC-V Processor for Datacenter
summary
A RISC‑V I8500 core is built for line‑rate packet work: RVB23 with SV39/SV48, big‑endian execution, an in‑order three‑issue pipeline and 4‑way SMT that lets one core plus threading double packet throughput versus a Cortex‑A53 in the authors’ Ethernet routing benchmark. DMA delivers headers into local scratchpad with a 3‑cycle header load, an IOCU provides coherent accelerator attachment, and cluster configurations show over 5x scaling moving to six cores with four threads each. Atlas Explorer virtual platform supports the performance modeling.
AI-written summary · about
chapters (7)
- 0:00Introduction to data centers
- 1:57Data processing units
- 2:58MIPS i8500 architecture
- 5:35Hardware multi-threading
- 7:57Performance benchmarks
- 11:09Use cases and data flow
- 16:01Simulation and conclusion
related talks
11mPacket Manipulator Processor A RISC V VLIW Core For Networking Applications
17mDesigning Next Generation 5G Wireless Infrastructure Using RISC-V
18mRoCE BALBOA: Service-Enhanced RDMA Offload Engine for Data Center SmartNICs
31mUnshackling Memory!
15mThe nanoPU: A Nanosecond Network Stack for Datacenters
17mDRack: A CXL-Disaggregated Rack Architecture to Boost Inter-Rack Communication
16mRingLeader: Efficiently Offloading Intra-Server Orchestration to NICs
34mCan the Linux networking stack be used with very high speed applications?
