UserBenchmark - what's so bad about it?

[Guest]

I use the UserBenchmark website quite a bit when comparing GPUs and CPUs just to get a general understanding of how big of a difference some hardware are from one and other. But I've seen a lot of people saying this site is garbage and should not be used. One guy claimed they're biased against Intel. Can someone give me an ELI5 of this? Should I not use this site?

[zeusthemoose]

I believe it is something about the formulas being completely wrong. But @BlueScope819 knows much more about this then me, maybe they can help explain it.

[Haro]

they fake numbers. not reliable and is over all fake, please avoid it. 

[minibois]

17 minutes ago, RonnyDouglas said:

I use the UserBenchmark website quite a bit when comparing GPUs and CPUs just to get a general understanding of how big of a difference some hardware are from one and other. But I've seen a lot of people saying this site is garbage and should not be used. One guy claimed they're biased against Intel. Can someone give me an ELI5 of this? Should I not use this site?

There are two things offending about the UserBenchmark scores.

 

Problem 1

The first and most egregious would be their CPU ranking, which is one of the first percentages you will see on the page:

Spoiler

Simply put, this number is comprised from different scores they gather and the distribution of importance is something like this:

Single core speed: 75% of the final score

4-core speed: 20% of the final score

64-core speed: 5% of the score

(don't quote me on those percentages, this is kind of illustrative and somewhat near what the weightings actually are)

 

This makes it so, as you can see in the screenshot above, a 4c4t CPU like an i3 9100F can rival a 12c24t CPU like the R9 3900X in overall score.

According to UB this is because gaming is their primary focus, which I understand; but I think putting so much importance in single core is not realistic nowadays.

 

Solution 1

The solution to this is simply ignoring the overall score and just looking at single core speeds and then thinking about how many cores/threads you will use and making a score for your use like that.

 

Problem 2

Another issue with the website is the way people can run benchmarks. I can run UserBenchmark on my PC, with a highly overclocked system, running very cool and with absolutely no background tasks.

Someone can also run UserBenchmark with no OC, while their system is overheating, with their RAM at the default 2133Mhz speed while watching a 4K60fps YouTube video.

 

UB will just present these two benchmark runs as equal, which they are not. This sorts of poisons the UB database and thus results.

 

Solution 2:

Look at UB for a vague idea of performance, but don't look at it as a definitive score.

[NineEyeRon]

If online comparison tools are used as a small part of wider research then I never see an issue.

 

If used in isolation then they are potentially misleading as it is just one perspective, and potentially skewed for various reasons such as sample sources, population skewness and test validity.

 

The same goes for pretty much all other similar online tools, including PSU calculators and bottleneck calculators. Along they are a poor way of making a decision but when used to be put in context, and their results not taken at face value but as comparisons within their data sets, they CAN be helpful.

 

Just don’t use them on their own to back up an argument, that will never end well.

[SavageNeo]

2 minutes ago, RonnyDouglas said:

I use the UserBenchmark website quite a bit when comparing GPUs and CPUs just to get a general understanding of how big of a difference some hardware are from one and other. But I've seen a lot of people saying this site is garbage and should not be used. One guy claimed they're biased against Intel. Can someone give me an ELI5 of this? Should I not use this site?

well i3 9350k is better than threadripper 3990x according to the site. thats all what you want to know. 

[5x5]

Userbench's owner has engaged in direct insults and attacks on reputable outlets. He has also skewed, faked and manipulated numbers. The site is a massive meme and is banned on multiple online forums because of it.

[5x5]

Exhibit A:

 

[mahyar]

as said before userbenchmark is bs and i use cinebench 3dmark and passmark something like passmark is like 30 years old and people used it and trusted it for more than my age

[5x5]

Exhibit B:

 

This is what they link to:

 

 

[Guest]

Thanks @minibois and everyone else for your replies. I will think a second time before trusting the numbers on the site.

[Morgan MLGman]

The site is garbage because of its "Effective speed" rating, because of its ranking and few other things but it can be useful to people.

There are two useful figures to check when you're comparing two CPUs on userbenchmark:
1. 1-Core speed metric should give you a rough estimate of singlethreaded performance of the chip

2. 64-Core speed number should give you the estimated multithreaded performance of the chip

The rest is pretty much useless garbage information.

[5x5]

Just now, Morgan MLGman said:

The site is garbage because of its "Effective speed" rating, because of its ranking and few other things but it can be useful to people.

There are two useful figures to check when you're comparing two CPUs on userbenchmark:
1. 1-Core speed metric should give you a rough estimate of singlethreaded performance of the chip

2. 64-Core speed number should give you the estimated multithreaded performance of the chip

The rest is pretty much useless garbage information.

I would actually say even those are a wash. I've had UB give me massive variation on two separate systems with identical specs. It just spat out incredibly distant numbers, especially for single-core speeds (both same i7, both same RAM speeds, both same BIOS settings and software installed).

[Morgan MLGman]

1 minute ago, 5x5 said:

I would actually say even those are a wash. I've had UB give me massive variation on two separate systems with identical specs. It just spat out incredibly distant numbers, especially for single-core speeds (both same i7, both same RAM speeds, both same BIOS settings and software installed).

In my experience they're more-or-less accurate, at least relatively speaking as these tell you nothing as absolute numbers, I only compare them in relation to one another.

They're somewhat arbitrary of course, but those two metrics seem more accurate than what most other "comparison" websites provide - the benchmarking tool itself is garbage though and has many bugs and quirks itself, it's terrible for storage performance when you're running multiple drives and can kind-of be tricked into showing higher, unrealistic performance numbers, which isn't acceptable for benchmarking software

[Ralf]

18 minutes ago, SavageNeo said:

well i3 9350k is better than threadripper 3990x according to the site. thats all what you want to know. 

That's because it shows how good the Intel cpu's are. /s

The 4core i3-9350KF is only a few percent behind the 10core i9-10900K.

[5x5]

1 minute ago, Morgan MLGman said:

In my experience they're more-or-less accurate, at least relatively speaking as these tell you nothing as absolute numbers, I only compare them in relation to one another.

They're somewhat arbitrary of course, but those two metrics seem more accurate than what most other "comparison" websites provide - the benchmarking tool itself is garbage though and has many bugs and quirks itself, it's terrible for storage performance when you're running multiple drives and can kind-of be tricked into showing higher, unrealistic performance numbers, which isn't acceptable for benchmarking software

I honestly don't remember what caused the massive disparity but I couldn't resolve it. Even nuking both OSes didn't resolve it. Only after I swapped boards did the scores also swap. I have no idea why an Asus board would produce better results than an ASRock board given all settings, clocks and etc were identical.

[manikyath]

userbenchmark tried to take a concept that is good for quick comparisons and trends over time, and tried to turn it into the be all and end all of metrics, to the point of toxicity.

 

if you look at just the bare numbers, they're actually pretty useful, its just that those numbers get burried deeper and deeper in an absolutely horrible pile of worthless fluff.

[SavageNeo]

2 minutes ago, Ralf said:

That's because it shows how good the Intel cpu's are. /s

The 4core i3-9350KF is only a few percent behind the 10core i9-10900K.

hope you are joking...

[SpookyCitrus]

1 minute ago, SavageNeo said:

hope you are joking...

Look at the rig in their signature, I doubt they are. Unless that too is a troll.

[SavageNeo]

Just now, SpookyCitrus said:

Look at the rig in their signature, I doubt they are. Unless that too is a troll.

9900ks with H310 board...  wait that has to be fake. NVM

[Ralf]

9 minutes ago, SavageNeo said:

hope you are joking...

 

7 minutes ago, SpookyCitrus said:

Look at the rig in their signature, I doubt they are. Unless that too is a troll.

Here, only 5% slower in games: https://cpu.userbenchmark.com/Compare/Intel-Core-i3-9350KF-vs-Intel-Core-i9-10900K/4055vs4071

[Morgan MLGman]

11 minutes ago, BlueScope819 said:

Your contradicting yourself?

I believe it's a pretty legitimate statement - the tool is garbage, true, but it's equally garbage for everyone, so if you assume that the average of scores for all CPUs is skewed in the same way, you can still use the scores as relative performance numbers.

[SpookyCitrus]

8 minutes ago, SavageNeo said:

9900ks with H310 board...  wait that has to be fake. NVM

Not fake but completely ridiculous. I've seen people do it before, but the main thing I'm confused by is it's a DDR3 H310 motherboard, wtf is this combination. 9th Gen Intel doesn't support DDR3 last time I checked. :/

[SavageNeo]

2 minutes ago, SpookyCitrus said:

Not fake but completely ridiculous. I've seen people do it before, but the main thing I'm confused by is it's a DDR3 H310 motherboard, wtf is this combination. 9th Gen Intel doesn't support DDR3 last time I checked. :/

well you can do some odd bios modding. 

 

 

[SpookyCitrus]

3 minutes ago, Ralf said:

 

Here, only 5% slower in games: https://cpu.userbenchmark.com/Compare/Intel-Core-i3-9350KF-vs-Intel-Core-i9-10900K/4055vs4071

Maybe you should read up on the rest of the replies in thread, posting a link to a Userbenchmark when we are talking about how unreliable and skewed Userbenchmark is, doesn't make a whole lot of sense.