UserBenchmark

[Spudwell06]

I wanted to compare two CPUs/Graphics Cards and UserBenchmark came up first.

Is it reliable/useful?

If not, what is a better website to use.

[gloop]

No, it is not a reputable website. 
 

Use trusted sites like Gamers Nexus or benchmarking tools (3D Mark, etc). 

[Dravinian]

15 minutes ago, comander said:

Their subtests can be ok. 

 

Their weighting is garbage. 

 

A year ago this would've been Intel's worst mainstream CPU vs their best high end desktop CPU. 

 

https://cpu.userbenchmark.com/Compare/Intel-Core-i9-9980XE-vs-Intel-Core-i3-9350KF/m652504vs4055

 

See how the budget part is ranked ahead? This is ridiculous. For modest workloads the two are similar. For heavy workloads the higher end part is almost 5x faster. Common sense says the latter is better, as do many real world tests including modem games. 

Except one costs $1,700 and the other costs $170.

 

The issues you raise are adequately and properly measured (so far as I know) in the sense that the 9980XE outscores the 9350 by a HUGE margin (295%) on anything multi-core and for workstation use.

But it is an expensive high-grade part that for 'most' users will not be a good value purchase, due to price and slower clock speeds.

 

I think saying that this is a comparison of purely CPU power, will lead you to say "well this is wrong" but that is not the sole factor and it is taking into consideration "value for money" where the 9350 is going to score WAY more than a 9980XE that outside of specific case uses isn't going to be value for money.

And the people with those specific case uses, won't be on userbench to see the comparisons, and they won't be comparing it to an i3 9350.

[KaitouX]

14 minutes ago, Dravinian said:

 

The 3600 loses to the 9400F in the "effective speed", even though it is faster(often by a lot) than the 9400F in everything but memory latency and both cost basically the same according to the site itself.

[Dravinian]

40 minutes ago, KaitouX said:

The 3600 loses to the 9400F in the "effective speed", even though it is faster(often by a lot) than the 9400F in everything but memory latency and both cost basically the same according to the site itself.

By 2%.

 

And the conclusion states this:

 

Quote

Comparing the i5-9400F to the Ryzen 3600 shows that the 3600 is 8% better for quad-core processing but it costs 40% ($50) more than the 9400F. Ordinarily higher quad-core performance would result in better gaming but the Ryzen 3000 memory controller, although significantly improved over previous Ryzen iterations, still has limited bandwidth and high latency. Gamers can invest the $50 savings in a better GPU, for example by upgrading from a 2060S to a 2070S

If you have issue with the statement, then fine, but it isn't just weighted for no good reason, the conclusion set out the reasoning.

 

That might be out of date, it may be incorrect. I don't know the answer.

 

It is also notable that looking through the list of comparison, AMD surpasses the Intel on every metric, bar one, and intel wins by 2%.

If you actually look through it all, when making a comparison, I doubt you would make the decision that the 9400f is a superior CPU for every workload but IF, and it is a big IF, the conclusion above is still true - for instance the prices may have changed - then a gamer may find the 9400f a better proposition.

 

I am not here to defend UserBenchMark, as I don't use it and I don't care about it.  Just saying, it isn't as clear cut as many on the internet wish it to be.

[Bombastinator]

The comments supporting userbenchmark seem to be with the accuracy of examples given, rather than the general conclusion which is that the userbenchmark relative scores aren’t good at reflecting reality and can be biased.  So using it as a “which is faster?” Metric can be misleading.

 

 My view is the combined score numbers of userbenchmark are not to be trusted, though their subset commentary on what parts of a cpu work better seem to be more accurate.  

[Haro]

15 hours ago, Spudwell06 said:

Is it reliable/useful?

No.

 

15 hours ago, Spudwell06 said:

If not, what is a better website to use

just watch a video from Gamer nexus or hardware unboxed etc.

 

11 hours ago, comander said:

Their subtests can be ok. 

no, no it cant.

Edited July 25, 2020 by TofuHaroto

[Nena Trinity]

I prefer passmark if a page has to be used.

 

https://www.cpubenchmark.net/compare/AMD-FX-8370-Eight-Core-vs-Intel-i5-9600K-vs-AMD-Ryzen-3-3300X-vs-Intel-i7-9700K-vs-AMD-Ryzen-5-3600XT/2347vs3337vs3716vs3335vs3781

 

[Bombastinator]

1 minute ago, TofuHaroto said:

No.

 

just watch a video from Gamer nexus or hardware unboxed etc.

 

no, no it cant.

quibbles with a by and large accurate statement. A question of granularity.

Re: usefulness

reliable? No.  Useful? Sometimes more often than not though no.

 

re: videos

would often require more than one video.  Time investment can get heavy.

 

re: subtests.  
depends on what is meant by subtests.

they will frequently compare parts of systems at the bottom of the page.  One thing I find interesting is the findings of those sub measurements (because quite often it looks more like reading of stats than testing) often disagree with the relative final scores. 

[Haro]

10 hours ago, Bombastinator said:

Useful? Sometimes more often than not though no.

getting wrong information is not really useful when you think about it. 

 

10 hours ago, Bombastinator said:

 Time investment can get heavy.

what? you're watching a video, requires 0 effort and time, you can literally just skip through it and look at the numbers. 

[Bombastinator]

7 minutes ago, TofuHaroto said:

getting wrong information is not really useful when you think about it. 

 

what? you're watching a video, requires 0 effort and time, you can literally just skip through it and look at the numbers. 

re: wrong information

Fair.  Especially if wrong information cannot be identified as such.  There is a limit to how inaccurate userbenchmark can be, but their testing variance range is quite large and they pretend that it isn’t, which is a problem.
 

Re video:

Zero is hyperbole.  Watching a video takes time. So does reading.  Video is slower than reading by definition generally, though it may depend on reading level.  It is generally less effort.  Skipping through stuff can be done with written stuff too.  Gamers nexus is very good about putting up a table of contents with time markers that improve things immensely.  Gamers nexus often puts up written stuff too.

[Haro]

10 hours ago, Bombastinator said:

 There is a limit to how inaccurate userbenchmark can be, but their testing variance range is quite large and they pretend that it isn’t, which is a problem.

they're simply wrong lol.

 

10 hours ago, Bombastinator said:

Zero is hyperbole.  Watching a video takes time.

why are you being so literal? 

[GoldenLag]

8 minutes ago, Bombastinator said:

re: videos

would often require more than one video.  Time investment can get heavy.

Actually Gamersnexus and Hardwareunboxed, as suggested above, have larger dataset for which to compare GPUs and CPUs in a single video.

 

Using the timeline to shorten the time it takes to find the graphs. GN also has written reviews of his videos. 

 

 

If one is only looking for a very comparison for GPUs in relative performance. Techpowerup have a "list" for comparative performance. Take this offcourse with a pile of salt, and allways confirm with the use of bencmarks in the given workload. 

[Bombastinator]

3 minutes ago, TofuHaroto said:

they're simply wrong lol.

 

why are you being so literal? 

It is a failing of mine to be literal. I have been accused of over precision in the past which makes me sensitive to it.

In this particular case I personally can read things a lot faster than I can watch things.  Nexus mods videos can be hard to get through for me.  I recognize  that may not be the same for everyone.