Running on desktop test machine HASWELL: Windows 10 Home 64-bit 3.0 GHz Intel 8-Core i7-5960X 32GB of RAM 3840x2160 pixels display ASUS NVIDIA GeForce GTX 980 Ti The public Zen Garden link has a timedemo benchmark mode, which renders 3200 frames and measures the overall duration it took. The benchmark can run either in WebGL disabled or enabled mode. The WebGL disabled mode no-ops all WebGL calls away to focus to stress WebAssembly execution performance. Links to the benchmarks: WebGL disabled: https://s3.amazonaws.com/mozilla-games/ZenGarden/EpicZenGarden.html?playback&novsync&noaudio&fakegl WebGL enabled: https://s3.amazonaws.com/mozilla-games/ZenGarden/EpicZenGarden.html?playback&novsync&noaudio At the end of the run, a list of measured benchmark values is displayed. The interesting results data for a rendering benchmark is the "totalRenderTime" field. Results: WebGL disabled: - Chrome 59.0.3042.4 (Official Build) canary (64-bit): 14834 msecs - FF Nightly 55.0a1 (2017-03-16) (64-bit): 12325 msecs i.e. Firefox wins in WebAssembly execution over Chrome by running through the demo in -16.9% less time. WebGL enabled: - Chrome 59.0.3042.4 (Official Build) canary (64-bit): 19716 msecs - FF Nightly 55.0a1 (2017-03-16) (64-bit): 26090 msecs i.e. Firefox takes +32.3% more time compared to Chrome when WebGL execution is included. For Chrome, enabling WebGL in the demo contributed a +32.9% time increase. For Firefox, enabling WebGL increased the execution time by +111.7%. There is an offline downloadable version of the benchmark at https://s3.amazonaws.com/mozilla-games/ZenGarden/2017-03-16-ZenGarden.zip. (The contents of that zip are also included in the emunittest suite)
Reporting some data from a partner who was testing this performance disparity on latest Firefox and Chrome (Firefox 54 and Chrome 59 I believe): i7-4790K & GTX 970, Chrome: 135fps i7-4790K & GTX 970, Firefox: 105fps i7-3610QM & GT 650M, Chrome: 53fps i7-3610QM & GT 650M, Firefox: 35fps Celeron N3150, ChromeOS: 22fps
You need to log in before you can comment on or make changes to this bug.