Google says Chrome 91 has 23% faster performance thanks to Sparkplug, a new non-optimizing compiler for its V8 JavaScript engine
Google says its new JavaScript compiler saves 17 years of Chrome users' CPU time each day. It should also benefit all Chromium browsers that use it.
Context & Ripple Effects
Chrome 91 extends a recurring performance push that included faster page loads in Chrome 85 and earlier work on load times and memory use. Sparkplug shifts that effort into V8's compilation path, with Google framing the gain in aggregate CPU time rather than a synthetic score.
That framing matters because Google had already retired the Octane JavaScript benchmark, arguing that benchmark-targeted tuning could diverge from real-world performance. The reported benefit is also relevant beyond Chrome because Chromium browsers using V8 can inherit it.
First-order effects
- Chrome 91 users receive faster JavaScript execution, while Google reduces the CPU time needed across Chrome's installed base.
- Chromium-based browsers that use V8 gain access to Sparkplug rather than having to build an equivalent compiler independently.
Second-order effects
- Browser vendors built on Chromium have a lower-cost path to performance improvements, concentrating the practical benefit of V8 investment across that ecosystem.
- Google can make performance claims around everyday CPU use instead of relying on a benchmark it had criticized as misaligned with real-world optimization.
Third-order effects
- If V8 improvements continue to flow through Chromium, browser performance differentiation will depend increasingly on engine and platform choices rather than each Chromium browser's independent optimization work.
- The shift toward workload-oriented performance claims makes benchmarks a less complete market interface, even as later browser comparisons such as Chrome 99's Speedometer result continue to shape competitive messaging.
The trend: Browser performance is becoming a shared Chromium-platform advantage, with JavaScript-engine efficiency emphasized over benchmark-specific tuning.