userbenchmark.com

What is your take on userbenchmark.com?
Is the site a reliable source or is it biased? And how do you know?

It is reliable because it compares benchmarks, every time someone benchmarks their computer on the site it uses that data. They are not transparent on how they calculate stuff, but it would be assumed that they factor comparison parts based off of a threshold of top performers for each part.

well it even says on that photo best bench, worst bench so they are accounting for min max values and then averaging from that probably.

Comparing benchmarks dosn't mean that the data they publish is accurate.
How do you validate the data published isn't biased that was my main tought.

And it would be nice if there was an option to compare not only "main gpu versions" but also variations, for example nvidia founders edition vs rog strix vs evga.

Identity said:

Comparing benchmarks dosn't mean that the data they publish is accurate.
How do you validate the data published isn't biased that was my main tought.

And it would be nice if there was an option to compare not only "main gpu versions" but also variations, for example nvidia founders edition vs rog strix vs evga.

Yes it does because it shows a percentile of expected performance so say 100 people do the benchmark and over 80 of them plot around the same area on the graph then this is a good indication of the parts expected performance which means accurate approximation.
The other 20% is just variance or people that either got bad parts or people that have not maximized their computers and bottlenecks etc.

its not a 1:1 comparison but the data is a pretty good reflection of what to expect when making decisions.

Not really sure why they have it like that but ref cards are at least a basis for expected performance so all aftermarket cards innovate from ref cards which means that you should see performance above the ref cards baseline.

I think if you think about it. The bench scores will account for aftermarket cards anyway so its still a somewhat accurate snapshot of the data because there will be people benching aftermarket on the site. The ref card image just symbolizes the origin where ref cards are the absolute baseline of expected performance and aftermarket cards are generally expected to perform higher than this.

Malneb said:

Yes it does because it shows a percentile of expected performance so say 100 people do the benchmark and over 80 of them plot around the same area on the graph then this is a good indication of the parts expected performance which means accurate approximation.
The other 20% is just variance or people that either got bad parts or people that have not maximized their computers and bottlenecks etc.

its not a 1:1 comparison but the data is a pretty good reflection of what to expect when making decisions.

Not really sure why they have it like that but ref cards are at least a basis for expected performance so all aftermarket cards innovate from ref cards which means that you should see performance above the ref cards baseline.

I think if you think about it. The bench scores will account for aftermarket cards anyway so its still a somewhat accurate snapshot of the data because there will be people benching aftermarket on the site. The ref card image just symbolizes the origin where ref cards are the absolute baseline of expected performance and aftermarket cards are generally expected to perform higher than this.

I was saying, that the people who host userbenchmark.com have the ability to publish fake data if they want to.
They have access to all benchmarks, they can delete them, edit them, create new ones, so no comparing benchmarks dosn't prove that the data is valid.

Well that is slightly pessimist there is no reason to rig the data. You sound set in your ways so the thread is kind of redundant because you already believe what you want to.

Clearly they are comparing parts against each other where it tells you which is the better performer. For what it is doing its a good approximation.

- - - Updated - - -

If they are fixing the data then our guess is as good as yours at that point because they don't allow the public to audit that data on the same terms that they do. Therefore it is assumed that the data that you see on the website is the same data that they crunch behind the scenes.

Outside of that no one would know so i guess yes it is possible but at the same time the numbers line up if you do some test.
GTX 780 is not going to perform the same as a new card and we see these reflections in the data.
The same as an RTX 4090 and RTX 2080 are going to show a clear devide which one is the better card.

I think if they were really trying to fix the data it would be for monetary gain which they do advertise places to source parts on the site with the adds etc. The incentive is there so its possible.

I think that in itself means that the data would be more minute if they were really about sales so it would be per card model not per family if their true intentions were really just trying to promote sales under the guise of benchmarking.

Its probably not that defined... ad revenue would be a way for them to host the site but i doubt it makes them rich. I never click on ads like that because i just source the best price myself many other people would do the same,

i think overall most people are also not going to a benchmarking website to make a purchase they are going to one to benchmark their parts or to compare parts when wanting to upgrade but most people are going to source in their locale. For me if i click on those adds it leads me to American links which is no good to me.

i also think that it would go against them if they were fixing data and its also not that viable to begin with because part prices vary immensely over markets and sellers.
There is no real cut way to say oh this part is hot so pump its data to make it look better so people click on more adds.

Finally it would also be to obvious if one part was considerably better than everything else in its category because computer parts are not progressing with those levels of gains its a very minute small scale that we are actually on with part performance.