Is userbenchmark a bad source?

[Okjoek]

Just curious because I have viewed it a number of times in the past, but recently I saw someone saying it wasn't a good source. I was under the impression that it was based on user submissions of some sort, not that I'm sure of how you go about this.

[LienusLateTips]

Yes. They have a huge bias towards Intel CPUs, and in general their data isn't accurate because it's user collected. 

 

Example:

https://cpu.userbenchmark.com/Compare/Intel-Core-i3-9100F-vs-AMD-Ryzen-TR-1950X/4054vs3932

 

>when i3 9100F is faster than TR 1950X

[Juular]

Raw user submitted data are good, it's just how they're making a rating with this data that heavily disregards multi-threaded performance despite modern games become more and more multi-threaded. So say, 4-thread CPU performs the same as 16-thead one in their eyes. That's partially true, for older games if two CPUs have comparable single-thread performance, the one with 4 thread will perform the same as 16-thread one but that's not true with most modern games.

[LienusLateTips]

Just now, Juular said:

Raw user submitted data are good, it's just how they're making a rating with this data that heavily disregards multi-threaded performance despite modern games become more and more multi-threaded. So say, 4-thread CPU performs the same as 16-thead one in their eyes. That's partially true, for older games if two CPUs have comparable single-thread performance, the one with 4 thread will perform the same as 16-thread one but that's not true with most modern games.

Except their user data doesn't filter out outliers. My stock 8700K is shown as being below average when it's perfectly fine and actually shown as above average among stock 8700K in other programs.

[Juular]

1 minute ago, LienusLateTips said:

>when i3 9100F is faster than TR 1950X

In gaming it is. But i have no idea how they've calculated that +18% difference when there's only +8% on ST.

 

TLDR: Just disregard ratings, look at raw data and paint a picture yourself.

[Juular]

1 minute ago, LienusLateTips said:

Except their user data doesn't filter out outliers. My stock 8700K is shown as being below average when it's perfectly fine and actually shown as above average among stock 8700K in other programs.

It's because OC'ing and just general performance variation between thousands of systems, it's a good instrument for comparing two CPUs, just disregard what it's saying about performance of your specific machine, in most cases these +-10% of difference are nothing to worry about.

[GoldenLag]

its at best a very very rough comparison tools for GPUs. for CPU i would just ignore it 

[Juular]

Back to your question.

It is bad as a source.

It's good for quick rough comparison.

To get accurate data look at reviews of CPUs in question.

[LienusLateTips]

7 minutes ago, Juular said:

It's because OC'ing and just general performance variation between thousands of systems, it's a good instrument for comparing two CPUs, just disregard what it's saying about performance of your specific machine, in most cases these +-10% of difference are nothing to worry about.

The thing is: that's the issue. If 1000 people overclock one CPU on LN2, it could massively skew the result in favour. Or, if 1000 people underclock to 0.8GHz, it could go that way, too.

[Juular]

1 minute ago, LienusLateTips said:

The thing is: that's the issue. If 1000 people overclock one CPU on LN2, it could massively skew the result in favour. Or, if 1000 people underclock to 0.8GHz, it could go that way, too.

Same with any unlocked CPU, it'll be equally skewed across the board. It's inaccurate by design. And i doubt that there's a lot of submissions using LN2, people that do that don't care about Userbenchmark.

[minibois]

If you want to know CPU performance, it can be a ... Sort of okay way to quickly have a glance. But when you want to know actual performance, you need to check proper reviews.

Apparently AnandTech has a database of scores from their CPU benchmarks. Might be a good alternative for a quick check.

 

It's also possible to check the Cinebench R15 Google Doc on this forum (CPU, motherboard and memory section, see one of the pinned threads). That can give rough performance metric (single and all core) speeds too.

 

Of course it's best to check how the CPU performs in your workload, but the others can be a decent alternative.

[Mark Kaine]

It's really only ok to compare parts like GPUs etc,  it's not a good benchmark at all however. Not comprehensive or detailed enough,  and their "ratings" are quite questionable as well. 

 

 

[Tan3l6]

3 hours ago, Arika S said:

it's good to get a rough idea, but should never be taken as a definitive source for anything. you're better off finding multiple websites that do full benchmarks of the components you are looking at and draw your own conclusions from it

Exactly - gives you the window of perception.

[Jimjamesjimothy8]

I wish it was more reliable because I like how comprehensive it is.

[Bombastinator]

19 minutes ago, TrainFan2019 said:

I wish it was more reliable because I like how comprehensive it is.

It’s got parts that are more useful than other parts.  It’s very bad at some things which is why it gets the horrible press it does.  I find that even the cpu reports aren’t useless.  It’s just their synthetic compilation metric that has no value.  Looking at the shape of the whole of the graph Can be useful.  For example the intel vs AMD graph posted as an example of how bad it is:  the graph of the intel chip is very narrow and the tip points to the right.  This implies than many users are pushing the thing as hard as it can go, while with the AMD it is much more spread out, showing it’s getting a lot more general use.  The people with the intel chip aren’t getting the performance out of it they want and are pushing their chips as hard as they can go looking for just a bit more where the AMD users don’t seem to feel they have to as much.

 If you look at the graph as more of an indicator of what various people are doing with the thing rather than a hard number it becomes less misleading.  It’s the synthetic compilation number on which cpu is “faster” which is often crap.

[Mark Kaine]

On 12/19/2019 at 8:54 AM, Juular said:

Back to your question.

It is bad as a source.

It's good for quick rough comparison.

To get accurate data look at reviews of CPUs in question.

I agree it's definitely not good for,  well,  "benchmarking"  but it's good to find out if something maybe wrong with parts in a PC, but it's still not very conclusive and doesn't test things like PSU and motherboard (which may be asking a little to much lol)

 

 

It definitely helped me fixing my under performing Kingston SSD,  but it took a very long time to actually figure out and fix it,  I just knew  "something"  was wrong with my hard-drives -- for some reason the Kingston even affected my Seagate Barracuda and Samsung Evo, which also work much better since I fixed the Kingston.

 

I'd say my gut feeling is userbench is mostly trash,  but it does have some value for specific things too. 

[Fasauceome]

Their data is not true random, and the variables aren't controlled, and is therefore irrelevant. Can't be used to meaningfully compare anything or test performance

[seagate_surfer]

The data colelcted by the community, good as a starting point, but should not be used as the final word, researching from several different resources and reviewers will give you a better and more critical opinion than to rely on a single source of information...

[Bombastinator]

6 hours ago, Fasauceome said:

Their data is not true random, and the variables aren't controlled, and is therefore irrelevant. Can't be used to meaningfully compare anything or test performance

This isn’t a bad way to think about it.  Their data isn’t useless but it’s dirtier than it looks and their conclusions are often worthless.  It can have value only when compared with other sources