Discussion Userbenchmark is unreliable?

[jacob249358]

I've always used user benchmark when comparing CPU and GPU. Just not AMD vs intel cause it's widely known its way off. But recently I've noticed some more odd results.
https://gpu.userbenchmark.com/Compare/Nvidia-RTX-3080-Ti-vs-Nvidia-RTX-3090/4115vs4081
examples of CPU brand misrepresentation https://cpu.userbenchmark.com/Compare/Intel-Core-i3-8350K-vs-AMD-Ryzen-9-3950X/3935vs4057

Please click the green human to continue - UserBenchmark

https://gpu.userbenchmark.com/Compare/Nvidia-RTX-3060-vs-AMD-RX-6600/4105vsm1680901 (6600 is always within 10% of the 3060 and beats it in some games)
at this point is user benchmark unreliable? Is there any kind of verification when people run it? why are some of these results so messed up when its real people doing this and not fictional benchmarks?
Id love to hear peoples thoughts on this

 

[kanewolf]

I've always used user benchmark when comparing CPU and GPU. Just not AMD vs intel cause it's widely known its way off. But recently I've noticed some more odd results.
https://gpu.userbenchmark.com/Compare/Nvidia-RTX-3080-Ti-vs-Nvidia-RTX-3090/4115vs4081
examples of CPU brand misrepresentation https://cpu.userbenchmark.com/Compare/Intel-Core-i3-8350K-vs-AMD-Ryzen-9-3950X/3935vs4057

Please click the green human to continue - UserBenchmark

https://gpu.userbenchmark.com/Compare/Nvidia-RTX-3060-vs-AMD-RX-6600/4105vsm1680901 (6600 is always within 10% of the 3060 and beats it in some games)
at this point is user benchmark unreliable? Is there any kind of verification when people run it? why are some of these results so messed up when its real people doing this and not fictional benchmarks?
Id love to hear peoples thoughts on this

It is useful to see DELTA changes on a single PC. It is not reliable to compare different systems, IMO and it should not be used to determine a PC's "underperformance" or performance vs cost.
It is useful to allow others to see a standardized overview of a PC's performance.

 

[hotaru.hino]

The data Userbenchmark receives is user submitted. And since there is likely a leaderboard style of listings, people want to game the system by trying to submit high scores. This can cause an inflation of how hardware is supposed to perform. This is also, to a lesser degree, a problem with 3D Mark.

I'd argue at best, it's useful for trying to get a comparison between two parts that none of the reputable hardware sites has tried. Like trying to compare a 10 year old top of the line GPU to a recent budget or integrated GPU.

 

[Unolocogringo]

It gets used here a lot because it can show all of your hardware and how it is running in your system.
Lots of posters have no clue what hardware they have. Or what speeds things are running at.
That is about as useful as it gets for it.
Everything is way off the mark as far as comparisons.

Some people buy a dell gaming machine. It has substandard parts everywhere in it.
1 stick of ram.
Poor cooling for processor .
Almost no case cooling.
Single 120 AIO for processor as exhaust.
And fan curves to let the hardware run hot and throttle in favor of decreased noise.
I spend weeks tinkering and overclocking to the max limits of complete stability. With very good case cooling and high performance quiet fans.
Even though the hardware is technically the same ,my system would blow theirs away from a performance/benchmark perspective.

 

[Zerk2012]

you can't really use it for comparison with parts but it helps some here to see what they have and if the CPU is throttling or XMP not enabled.

 

[punkncat]

The site has been called out numerous times for a bias towards Intel. Just the same, and as others have already pointed out, it has its uses.

 

[USAFRet]

What also gets peoples panties into a wad is a report like this:

"Way below expectations", or "Excellent" ?
Which is it?

 

[alceryes]

What also gets peoples panties into a wad is a report like this:

"Way below expectations", or "Excellent" ?
Which is it?

This comes from a misunderstanding of how Userbenchmark reports its results, Userbenchmark's poor documentation about these numbers, AND the fact that they update their benchmarks to be more difficult, on the fly.
The 89.3% is how that CPU performed against Userbenchmark's standard test - this score is practically useless.

The performance percentage in the actual component description is how this component fares against all the other Ryzen 5 5600X's that have been tested. This test is useful, to a point. It is only comparing that particular model CPU but does NOT take into account CPU OCs, or different memory speeds, or BCLK OCs, etc. That is why I like to say that if you are 50% or higher in this score you're doing okay.

 

[alceryes]

It's has its uses, but yeah, measuring performance isn't one of them.
You can't get an idea of 'normal' performance when:
A)Bob's 10600K with PL1 125w + PL2 250w no OC gets compared to a bunch of random folks...
12900Ks overclocked on LN2
FX 8350s power/thermal throttling all the time
Ryzen 5600Xs with PBO on
and so on

B)Or take Bob's 3060Ti FTW3 Ultra...
2080Tis on LN2
3060Ti FE
Thermal throttling 3090s
and so on

3d Mark is a little cleaner in that regard, but still impossible to compare a cpu or gpu to 'stock', because there's still overclocks and poorly cared for systems in the mix.

This isn't true if you are looking at each component's percentage. The bench score itself is completely useless, yes.

 

[alceryes]

I've always used user benchmark when comparing CPU and GPU. Just not AMD vs intel cause it's widely known its way off. But recently I've noticed some more odd results.
https://gpu.userbenchmark.com/Compare/Nvidia-RTX-3080-Ti-vs-Nvidia-RTX-3090/4115vs4081
examples of CPU brand misrepresentation https://cpu.userbenchmark.com/Compare/Intel-Core-i3-8350K-vs-AMD-Ryzen-9-3950X/3935vs4057

Please click the green human to continue - UserBenchmark

https://gpu.userbenchmark.com/Compare/Nvidia-RTX-3060-vs-AMD-RX-6600/4105vsm1680901 (6600 is always within 10% of the 3060 and beats it in some games)
at this point is user benchmark unreliable? Is there any kind of verification when people run it? why are some of these results so messed up when its real people doing this and not fictional benchmarks?
Id love to hear peoples thoughts on this

It's a useful tool for a quick diagnostic look.

 

[USAFRet]

Exactly.
In my report above, the difference between "Poor" and "Great" is 20 percentage points.
86% to 106%.

So that Ryzen 5 is running at 89.3% of a perfect 100.
By any standards, that is running just fine, nothing to be "fixed".

But people will see that "Way below", and lose their mind.
It is being compared to systems that are overclocked and gamed to get the best numbers.

 

[alceryes]

Exactly.
In my report above, the difference between "Poor" and "Great" is 20 percentage points.
86% to 106%.

So that Ryzen 5 is running at 89.3% of a perfect 100.
By any standards, that is running just fine, nothing to be "fixed".

But people will see that "Way below", and lose their mind.
It is being compared to systems that are overclocked and gamed to get the best numbers.

Don't listen to their descriptions. Ignore them completely. Just look at the numbers. If 50% of that model of CPUs tests in a 5% range in the bench score (let's say 88% to 93%) then scoring just below that will almost guarantee a rotten 'description' because you're slower than at least 50% of the R5 5600X CPUs tested.

However, the percentage saying that you are in the 13th percentile of R5 5600X CPUs would be something to look at. Click on that CPU to get a list of that recent CPUs tests. Go down to the bottem and hit the link to show all of that CPUs test. Ignore the top 100+ scores. Go to page 2-3 to find the CPU score that is at/around the 50% relative performance mark and see how those systems differ from your 13th percentile score. Check out those system scores in detail.

 

[Bob.B]

I've always used user benchmark when comparing CPU and GPU. Just not AMD vs intel cause it's widely known its way off. But recently I've noticed some more odd results.
https://gpu.userbenchmark.com/Compare/Nvidia-RTX-3080-Ti-vs-Nvidia-RTX-3090/4115vs4081
examples of CPU brand misrepresentation https://cpu.userbenchmark.com/Compare/Intel-Core-i3-8350K-vs-AMD-Ryzen-9-3950X/3935vs4057

Please click the green human to continue - UserBenchmark

https://gpu.userbenchmark.com/Compare/Nvidia-RTX-3060-vs-AMD-RX-6600/4105vsm1680901 (6600 is always within 10% of the 3060 and beats it in some games)
at this point is user benchmark unreliable? Is there any kind of verification when people run it? why are some of these results so messed up when its real people doing this and not fictional benchmarks?
Id love to hear peoples thoughts on this

It's easy to use and quick to run.
It gives some GP info on a few different areas of the machine.
If some area shows way out of wack it might be something to look deeper.

As for comparison of parts it's based on their test other test sites may have a different test and opinion.

It's difficult to compare a machine running stock settings against a machine that is overclocked which is where the UBM numbers may get a little screwy.

I used it often on this machine when I went into speed-it-up mode.
Change a part...ubm.
Change a driver...ubm.
Change a bios...ubm.
Change a setting...ubm.
All to make sure I was going forward and not backwards.

 

[USAFRet]

It's difficult to compare a machine running stock settings against a machine that is overclocked which is where the UBM numbers may get a little screwy.

And that is where people get all wrapped around the axle.

"I just ran a UBM and it says my system SUX! How do I fix it?"

There is no context. All people see is that scary red text.

 

[carocuore]

It sucks because it compares your build against heavily overclocked systems so it'll always show yours is "way below expectations"

 

[Bob.B]

And that is where people get all wrapped around the axle.

"I just ran a UBM and it says my system SUX! How do I fix it?"

There is no context. All people see is that scary red text.

Yes you have to know how to read it.

That's what the M's get paid big bucks for.......unwrap the axle.

 

[Karadjgne]

I like it for its specs, it'll say the things like ram running at 2133MHz when Op has 3600MHz etc, it'll say exactly what components they have, when Op is somewhat clueless.

And that's as much thought as I give it.