How reliable are synthetic benchmarks ACTUALLY?

[Tristan_86]

I have lots of people telling me synthetic benchmarks such as Geekbench and AnTuTu aren’t reliable and are inaccurate.

Is this true? Is there a legitimate reason for this? Why would they exist then?

Also, can you compare Android to iOS? What about phones to PCs?

One guy told me there are way too many different factors to actually compare them and claims Android CPUs are faster than the A11 and synthetic benchmarks don’t mean anything. Is that being biased? Or is it legitimately true?

My iPhone 8 Plus scores higher than my desktop with an i3 6100 on Geekbench. I’m kind of skeptical about that. Is my phone really more powerful? If not, what’s the reason for it?

Thanks!

 

[Gam3r01]

Synthetic benchmarks tell you exactly one thing. How good something is at running that singular benchmark.
Synthetic tests do not equal real world performance in any way, shape, or form.

If you have two sets of the same hardware running the same test, then it has some validity, in showing if they are performing on par, or if something is wrong.

You cant really use a synthetic benchmark to compare Andoid to IOS, and you most certainly cant use them to compare to PCs.

Use your PC to render a video, now use your iPhone to render the exact same video (hint, you cant) and compare that. A real world test. That is the only valid comparison of performance.

 

[InvalidError]

Synthetic benchmarks are a weighed averages of various workloads and only give you a general idea of performance.

Real application and game benchmarks are only valid for those specific applications and games. Different applications and game stress components differently, so you can't use any single real-world benchmark to judge any CPU, GPU, RAM, SoC, etc.'s overall performance either. You have to run a bunch of different benchmarks and look at trends that emerge from there to get an overall feel.

If you want to infer performance data for an application or game that you cannot find many benchmarks for, then you'll have to look at what results are available to see what other application or game shows similar performance characteristics and attempt educated guesses from there.

If you want reliable benchmark results for a specific task, then you'll have to find benchmarks for that specific task and the hardware combination you are interested in. Anything else requires approximation using knowledge of trends in similar workloads.