Easier CPU/GPU Comparisons On OpenBenchmarking.org, Other New Features

[phoronix]

Phoronix: Easier CPU/GPU Comparisons On OpenBenchmarking.org, Other New Features

With the new OpenBenchmarking.org that's been out in public form since last month and being developed as part of the soon-to-be-released Phoronix Test Suite 10.0, here is the latest feature now enabled in making it much easier for quickly carrying out high-level processor (CPU) and graphics card (GPU) component comparisons along with other improvements...

http://www.phoronix.com/scan.php?page=news_item&px=Easy-CPU-GPU-Linux-Comparisons

[SteveW]

Automatic "value" rankings would be really nice; something like this:

https://www.videocardbenchmark.net/gpu_value.html

[Michael]

Automatic "value" rankings would be really nice; something like this:
https://www.videocardbenchmark.net/gpu_value.html

Automatic value rankings are more difficult in the context of OpenBenchmarking.org given there are hundreds (600+) of very diverse and different test profiles and most people only use a subset of them depending upon what particular workloads are of most interest to them.... So the closest that OB does (and the only thing I can think of in this context) is showing the average percentile of that component based on the percentile rankings for each individual benchmark run. Still not perfect and theoretically possible of gaming the rank in some aspects, but not many alternatives.

e.g. https://openbenchmarking.org/s/AMD%2...950X%2016-Core the 3950X has a 72nd percentile rank overall.

[elatllat]

Using CSS properly to make it legible on mobile would be nice. Google rates the mobile experience at 53%;


https://developers.google.com/speed/...2FProcessor%2F

[Bitter]

This is great start. I'm sure Micheal will improve it over time.

[Teggs]

I appreciate the feature. I expect it will bring both good and bad attention from outside the usual crowd. Attention for Phoronix and the Test Suite is good, but to be honest, superficially similar sites around the web are operator manipulated and magnets for user vitriol. I expect a number of first time viewers will assume this is just another con. I would like to think that users will read and understand the methodology, and the ignorant will be outweighed by people looking for a grounded, facts-based comparison, but... I've been using the internet too long for that.

I'm sure Michael knows what he's doing, and I hope the gamble goes in his favour.

[Michael]

Using CSS properly to make it legible on mobile would be nice. Google rates the mobile experience at 53%;


https://developers.google.com/speed/...2FProcessor%2F

Mobile improvements will come in time albeit not a pressing focus as I imagine most anyone seriously evaluating component upgrades is doing so from a desktop.

[Michael]

I appreciate the feature. I expect it will bring both good and bad attention from outside the usual crowd. Attention for Phoronix and the Test Suite is good, but to be honest, superficially similar sites around the web are operator manipulated and magnets for user vitriol. I expect a number of first time viewers will assume this is just another con. I would like to think that users will read and understand the methodology, and the ignorant will be outweighed by people looking for a grounded, facts-based comparison, but... I've been using the internet too long for that.

I'm sure Michael knows what he's doing, and I hope the gamble goes in his favour.

Yep, zero intervention by me in that data, entirely all based on community-collected data by PTS. The extent of my intervention now is hitting on the power button to the EPYC box that does all the offline data processing every morning (though could always have Phoromatic enable its WoL capabilities too ).

[Neuro-Chef]

Yep, zero intervention by me in that data, entirely all based on community-collected data by PTS.

So every benchmarks helps or is there a certain set that's particularly suitable to provide data for the comparison?

[Michael]

So every benchmarks helps or is there a certain set that's particularly suitable to provide data for the comparison?

Technically every benchmark helps albeit if it's deemed an outlier against the current data it's not factored in at that point in time at least until more data comes in that no longer makes that particular result an outlier, among other similar data safety.