New OpenBenchmarking.org Features Enhance Discovering Popular + Reliable Tests

[phoronix]

Phoronix: New OpenBenchmarking.org Features Enhance Discovering Popular + Reliable Tests

As part of the new OpenBenchmarking.org being developed as part of Phoronix Test Suite 10.0 due out next quarter, some new features were deployed live on OpenBenchmarking.org this weekend...

http://www.phoronix.com/scan.php?page=news_item&px=OB-New-Test-Metrics

[schmidtbag]

Really cool changes. I can see this being super helpful to people trying to test a new system.

[Michael]

Really cool changes. I can see this being super helpful to people trying to test a new system.

Thanks, that's among the goals.

[lumks]

Would be nice to have a feature where you could compare hardware, like in CPUa vs CPUb. I have much more trust in OBorg compared to userbenchmark for example.

[Michael]

Would be nice to have a feature where you could compare hardware, like in CPUa vs CPUb. I have much more trust in OBorg compared to userbenchmark for example.

I have a feature hopefully ready in a week or two to hopefully address that. These features today are stepping stones.... Because of having nearly 600 different possible benchmarks and all the combinations of Linux software that can impact performance, it's more difficult to provide an automated side-by-side for any given CPus that is very reliable. But the work on percentile ranking, being able to automatically determine the most reliable tests in terms of minimal result noise and popular tests (the data being exposed with today's update) is part of that underlying work for being able to help categorize automatically what tests should be shown to users when comparing CPUs where there is sufficient data available, etc.

[zboszor]

Somewhere in this forum it was asked that the frame time graphs should use logarithmic scale because the linear scale squashes the interesting interval if there are a few samples off the scale. Using logarithmic scale would provide good resolution for the interesting interval and would also show the anomaly samples.

[Michael]

Would be nice to have a feature where you could compare hardware, like in CPUa vs CPUb. I have much more trust in OBorg compared to userbenchmark for example.

This is still W.I.P. and actively working on it in real-time this weekend to polish it up, but is this what you are after? https://openbenchmarking.org/vs/Processor/