SiFive Performance P650 RISC-V core to outperform Arm Cortex-A77 performance per mm2

About six months have passed since the SiFive announcement of the Performance P550 “fastest 64-bit RISC-V processor” ever, and the company has now introduced an even faster RISC-V core with the Performance P650 that’s expected to match Cortex-A77 performance.

Building upon the Performance P550 design, the SiFive Performance P650 is scalable to sixteen cores using a coherent multicore complex, and delivers a 40% performance increase per clock cycle based on SiFive engineering estimated performance in SPECInt2006/GHz, thanks to an expansion of the processor’s instruction-issue width. The company compares P650 to the Arm family by saying it “maintains a significant performance-per-area advantage compared to the Arm Cortex-A77”.

SiFive Performance P650

SiFive Performance P650 key features:

  • 64-bit RISC-V (RV64GCB) core Sv39/Sv48 Virtual Memory Support
  • Multi-core, multi-cluster processor configurations with up to 16 cores
  • Performance > 11 SpecINT2006/GHz
  • Thirteen stage, four-issue, out-of-order pipeline tuned for scalable performance
  • Private L2 Caches and Streaming Prefetcher for improved memory performance
  • Cache stashing to L3 for tightly coupled accelerators
  • SECDED ECC with Error Reporting
  • Hypervisor Extension and System Level Virtualization IP
  • SiFive WorldGuard System Security

Other architecture enhancements over the previous generation include a higher maximum clock frequency (Liliputing says up to 3.5 GHz), platform-level memory management, interrupt control units, and support for the new RISC-V hypervisor extension for virtualization.

SiFive will offer an “Architecture Preview” to select lead customers in Q1 2022 for evaluation of the SiFive Performance P650 with a development kit including RTL evaluation, test bench RTL,  a software development kit, an FPGA bitstream, and documentation.

General availability for the Performance P650 is expected in Q2 2022. The cores should be used in SoCs targeting a variety of markets from data center to edge, automotive, compute, mobile, and more. More details may be found on the product page.

Via SiFive press release.

Share this:
FacebookTwitterHacker NewsSlashdotRedditLinkedInPinterestFlipboardMeWeLineEmailShare

Support CNX Software! Donate via cryptocurrencies, become a Patron on Patreon, or purchase goods on Amazon or Aliexpress

ROCK Pi 4C Plus

20 Replies to “SiFive Performance P650 RISC-V core to outperform Arm Cortex-A77 performance per mm2”

  1. Has any of the previous SiFive OoO CPU ever been independently benchmarked? I mean is there any easily available board running U84 or U87?

    1. btw, U84 is P550 and U87 is P650. At least what I’ve seen so far. I’m looking forward to intel horse creek platform in ’22.

  2. Note that performance per area does not interest end users, it’s only for silicon vendors. What matters most is performance or efficiency. Here it would be nice to know how the raw performance compares to A77. Especially with 16 cores, where the A77 has LSE atomic extensions while (last time I checked) RISC-V still uses the non-scalable LL/SC model.

    1. Yes that would be most interesting! Given a comparable efficency, like an a77, more performance per area is better.

      It’s getting interesting, riscv is now just two generations behind leading arm perf.

      1. > riscv is now just two generations behind leading arm perf

        Based on what? Do you know more than those obscure SpecINT2006/GHz ratings? Doesn’t it sound a little absurd to publish SpecINT2006 numbers years after ‘SPEC CPU 2017’ made their earlier 2006 benchmark suite officially retired?

        Well, it saves them the burden to let results being reviewed and published by SPEC. But other than that?

        1. I guess 2006 is just more easy to be run in emulated environment. I mean on model core loaded into big fpga clocked around 100 MHz or so. Also I don’t mind as long as there are others 2006 numbers for comparison.

    2. “performance per area does not interest end users”

      I am probably weirder than a typical end user, but better area efficiency could mean cheaper chips because of the smaller die size, or an increase in core count compared to other ARM SoCs. Let’s get some cheap 16-core single board computers already.

  3. Dual core u54 is equal to dual core cortex a7. That is true . u74 is equal to cortex a53. Long journey..,.. and they very expensive. I thin google wont port android and android services to risc v. I will be suicede and mean independent for huaweii and other china companies. Usa wants arm to belong nvidia and monopoly. Risc v is in very difficult position. Intel wants to buy it… And maybe kill….

    1. USA does not want Nvidia / Arm deal, they fear conflict of interest. Intel will not buy RISC V. More companies are entering the SoC market and many will get burned.

      1. The NVIDIA/Arm deal felt like a non-starter from the very beginning because NVIDIA competitors would also become their customers. Europe and China governments also have problems with it.
        Nobody is going to buy “RISC-V”. I think the deal was about Intel purchasing SiFive.

        1. Yes SiFive turned them down. Qualcomm are buying Nuvia

          ” NUVIA comprises a proven world-class CPU and technology design team, with industry-leading expertise in high performance processors, Systems on a Chip (SoC) and power management for compute-intensive devices and applications. “

  4. Based on SiFive data P550 -> P650 perf. increase is 8.9 -> a bit more 11 spec2006/ghz. Does not look to me like claimed 40% perf. increase but looks more in line with increasing width of SoC from 3 to 4.

    1. Maybe an other approach is to let others make/break their teeth on it and make your opinion after some time. I’m personally seeing that it seems to be progressing. 2 years ago I wouldn’t have bet much on SBCs with that, now we’re starting to see them appear. We’ll finally have a chance to measure its real value.

      1. It needs development still. The instructions are there, next is building cost effective powerful SoC, and that needs time to mature IMO.

Leave a Reply

Your email address will not be published. Required fields are marked *

Khadas VIM4 SBC
Khadas VIM4 SBC