SiFive Performance

The SiFive Performance™ P270 is an 8-stage, dual-issue, highly efficient in-order pipeline compatible with the RISC-V RV64GBCV ISA. With full support for the RISC V Vector Extension 1.0, and combined with the popular SiFive Recode utility, which translates existing SIMD software from popular legacy architectures to RISC-V Vector assembly code, the P270 is an ideal replacement for dated SIMD architectures.

SiFive Performance Family

The SiFive Performance family of RISC-V processors is designed for maximum throughput, while preserving power efficiency for workloads as varied as branch-intensive operating systems, and multimedia processing. Ranging from 8-stage, dual-issue, in-order architectures equipped with 256-bit vector engines, to advanced three and four-issue out-of-order processors, with and without vector compute capabilities, with best-in-class industry benchmark performance, the SiFive Performance Family represents the future of compute.

SiFive Performance
P270 Key Features

  • 256-bit vector length processor
    - Variable length operations, up to 256-bits of data per cycle, with dynamic vector length configuration
    - Ideal balance of control and data parallel compute
  • Performance benchmarks
    - 5.75 CoreMarks/MHz
    - 3.25 DMIPS/MHz
    - 4.6 SpecINT2k6/GHz
  • Scalar processing built from U7 series core
    - Multi-layer Caching support for optimum data movement
    - Stride Prefetcher
    - Virtual memory support, up to 48-bit addressing
  • High performance, flexible connectivity to SoC peripherals
  • Implements RISC-V Vectors v1.0
  • Dual issue scalar unit runs concurrently with vector unit
  • Key vector unit attributes
    - VLEN = 256. DLEN = 128 (datapath width). ELEN = 64 (datatypes)
    - Separate memory and ALU pipelines for concurrent operation
    - Vector operations, decoded and queued in Vector Unit for parallel operation of Scalar and Vector units
  • Vector ALU
    - 128b ALU can perform 2x64b, 4x32b, 8x16b, 16x8b ops/cycle
    - Integer and Floating point data types supported
  • Vector Loads/Stores are 128b/cycle
    - L2 cache treated as primary memory
    - Load from L1 cache, initiates L2 cache load in parallel, minimizing L1 cache miss impact
  • Multi-core, multi-cluster processor configuration, up to 8 cores

Contact Sales