RISC-V is Slow

(This blog post was discussed on YCombinator's Hacker News here.)

After watching ARM displace Intel from most of my personal and server computing, I've been following RISC-V. It's an exciting open-source ISA with real potential, but RISC-V is currently quite slow compared to modern CPUs in 2024, even slower than Raspberry Pi boards.

How Slow is RISC-V?#

You can start with these Geekbench performance scores for various RISC-V processors: https://browser.geekbench.com/search?q=RISC-V.

I haven't seen any single threaded scores above 150 and no multi-threaded scores higher than 1500. The best result I've seen is the Sophgo Mango's score of 147 single threaded and 1300 multi-threaded. And the average score for RISC-V CPUs in Geekbench is around 80 single threaded and 400 multi-threaded.

In comparison, Apple's M4 series is seeing scores around 3800 single-threaded and 15000 multi-threaded. A more fair comparison would be the Raspberry Pi series of CPUs, but even then it isn't close. The Raspberry Pi 5 scores around 1000 single-threaded and 2200 multi-threaded, while the Raspberry Pi 4 scores 300 single-threaded and 739 multi-threaded.

So in that context, the fastest RISC-V CPU I could find is 25x slower than a top-of-the-line Apple M-series chip and ~7x slower than a Raspberry Pi 5 and 2x slower than a Raspberry Pi 4 for single-threaded performance. Those that say that RISC-V is a viable replacement for x86 or ARM in the near term are kidding themselves.

Why Is RISC-V So Slow?#

RISC-V implementations in the wild lack out-of-order execution and advanced branch prediction — the two features that give modern high-performance cores most of their speed. Most commercial RISC-V chips execute instructions in the order they arrive. Closing that gap requires substantial microarchitectural investment, not just faster clock speeds.

RISC-V is Great for Embedded Systems#

RISC-V is gaining ground in embedded applications where cost and power efficiency matter more than raw speed. Espressif's SoC lineup, which has largely moved to RISC-V, is a good example: these chips power IoT devices where low power consumption counts for more than clock performance.

Current Path to High Performance#

Two projects are pushing RISC-V toward higher performance. The Berkeley SonicBOOM project produced an experimental out-of-order core with more advanced architectural features, though it hasn't reached mainstream use. Tenstorrent, founded by chip designer Jim Keller (AMD's Zen architecture), is building high-performance RISC-V processors. Neither has appeared in mainstream benchmarks yet, but Tenstorrent's track record makes the effort credible.

ARM's Path as a Historical Reference#

ARM started in low-power embedded systems with similar performance constraints. Through years of architectural improvements, ARM expanded into mobile, then servers, then high-performance desktops. That trajectory contributed to Intel's market plateau. But that evolution took decades of sustained investment.

If you're building something that needs raw performance, RISC-V isn't ready. The current generation of commercially available RISC-V processors remains well behind modern x86 and ARM in computational power, with research and startup work still years from mainstream availability.