Benchmark Values

[Kunalkp]

I was wondering why exactly do benchmark values are never the same. They always fluctuate. 

Case 1 : When two different PC's of the same config.

Case 2:  When the benchmark is run on the same PC twice.

Why don't ever the values match ?

[Mira Yurizaki]

18 minutes ago, Kunalkp said:

I was wondering why exactly do benchmark values are never the same. They always fluctuate. 

Case 1 : When two different PC's of the same config.

Case 2:  When the benchmark is run on the same PC twice.

Why don't ever the values match ?

• Environmental factors: a computer may run better when cooler than after 5 rounds of testing when everything is warmer
• Clock drifting: Clock generators aren't exactly accurate because they don't need to be. The only one that may be accurate enough (as in, good enough for a $5 wrist watch) is the system time clock.
  • Environmental factors can also affect clock accuracy.
• Timing of events: You can be at just the right timing to not perfectly sync up. It's like if you commute through back roads with lots of stop lights, sometimes you get all greens while other days you get mostly reds, even though on average over time all the lights run at a certain frequency.

[Kunalkp]

Does Silicon lottery also have any effect ? For intel K version chips; Do that also make the benchmark values vary ?

[dexT]

Some do more than others while some like XTU score the same everytime.

 

If you mess around enough you'll find routines for each benchmark that'll make it score higher, some a few points and some you'll destroy the expected score.

 

The OS you use and how its configured have a huge impact on performance.

[Mira Yurizaki]

17 hours ago, dexT said:

The OS you use and how its configured have a huge impact on performance.

Assuming we're only talking about comparing benchmarks using the same OS, I've found that tuning the OS doesn't really help much with gaming performance vs. the out-of-the-box configuration. By "doesn't really help much", performance is within a margin of error. And that's a good thing.

 

The only thing that would affect performance is the runtime environment, which is the reason why Game Mode on Windows 10 improves UWA more than Win32.

[dexT]

38 minutes ago, M.Yurizaki said:

Assuming we're only talking about comparing benchmarks using the same OS, I've found that tuning the OS doesn't really help much with gaming performance vs. the out-of-the-box configuration. By "doesn't really help much", performance is within a margin of error. And that's a good thing.

 

The only thing that would affect performance is the runtime environment, which is the reason why Game Mode on Windows 10 improves UWA more than Win32.

The desktop theme will impact scores on certain benchmarks not to mention stripping the OS and tweaks I'm not going to share. This is purely for competitive benchmarking and probably has no effect on video games though I don't know. Hardcore benchmarking is a totally different thing than optimizing a gaming PC or workstation.

 

You also tailor what the benchmark likes. GPU-PI for CPU is a good example. It needs XP to score high. It needs Intel OpenCL for 100M and AMD for 1B to score high. 3DMark01 is the same, XP for an extra 2000+ points but with this one the actual order you run the scenes in has a big impact on the overall score.

 

Some OS tweaks are not allowed, and as recently happened, they break a benchmark and it removes point scoring from it until everything is audited.