Speedometer 3.0 Benchmark comparison between compiler variants

[Girofox]

So i tried the 4 variants for Windows with interesting results. Beware this is still with the thinLTO flag!

Setup

• i7-4510u + Intel HD Graphics 4400

Test method

• HWiNFO64 + polling rate 0.5s + 200 samples for average
• Browser window exactly half screen size docked to left + focused
• IA core power (W) to measure

Baseline

• around 0.5 W idle IA core power usage
• undervolt of 75 mV

Results

AVX-Builds (Windows 11, x64)

AVX:	6.75 ± 0.22,  Geomean 148.51 ± (3.3%) 4.97 ms
	CPU: 9.54 W (IA Cores, avg)

AVX2:	6.73 ± 0.13,  Geomean 148.58 ± (1.9%) 2.88 ms, CPU: 9.50 W (IA Cores, avg)
	CPU: 9.50 W (IA Cores, avg)

SSE-Builds (Windows 11, x64)

SSE3:	6.71 ± 0.26,  Geomean 149.44 ± (4.0%) 6.03 ms
	CPU: 9.64 W (IA Cores, avg)

SSE4:	6.74 ± 0.19,  Geomean 148.54 ± (2.9%) 4.36 ms
	CPU: 9.60 W (IA Cores, avg)

Conclusion

• The bare scores are almost the same and within uncertainty range. But the Geomean is a tiny bit higher on the SSE3 variant.
• Average power usage is higher on the SSE builds especially SSE4 and lower on AVX buids (did expect opposite)
• The differences are very small, maybe Javascript performance (TodoMVC) is not affected by build
• Power usage is higher than in Chromium builds for my setup