What should I use a benchmark test for?

[StunnaFan34]

I thought I could use benchmark tests like timespy and passmark to check if my system is performing as expected. But someone told me that you shouldnt use benchmark tests to check for that, only to see differences in overclocking. Is this true? Are tests like passmark and timespy (3DMark) not good to check for under performance? In other words, if my score is around the same as others does that mean im good?

[Moonzy]

3 minutes ago, StunnaFan34 said:

if my score is around the same as others does that mean im good?

if your score is within 5%-ish of others with the same setup, it's good enough

5%-ish tolerance because some people might have better/worse clock speeds on their GPU/CPU

[StunnaFan34]

Just now, Moonzy said:

if your score is within 5%-ish of others with the same setup, it's good enough

5%-ish tolerance because some people might have better clock speeds on their GPU/CPU

ah ok so the tests are good to see if youre performing as expected? thanks for the quick response

[StunnaFan34]

1 minute ago, DutchGuyTom said:

Yes, I use them all the time to see if my systems are performing as expected. 3DMark is a good all-around test for gaming systems and you can check scores for comparable systems using their new search function.

 

It will show you a graph but you have to keep in mind that the upper half of scores consists mostly of people overclocking their systems. So if your score is not in the top half, that doesn't reflect poorly on your system. If you're in the bottom 25%, you should look into.

thank you for the response. is passmark also a good test?

[StunnaFan34]

51 minutes ago, DutchGuyTom said:

I have heard pass mark is pretty good too but I have less experience with them because, if I recall correctly, it is only free for the first month. I used up that free month already and was quite impressed with how thorough it is. But since my free trial expired, I primarily use 3DMark demo now.

As long as you don't use UserBenchmark, you're fine.

ok thank you. yeah i dont trust user benchmark

[StunnaFan34]

4 hours ago, DutchGuyTom said:

Yes, I use them all the time to see if my systems are performing as expected. 3DMark is a good all-around test for gaming systems and you can check scores for comparable systems using their new search function.

 

It will show you a graph but you have to keep in mind that the upper half of scores consists mostly of people overclocking their systems. So if your score is not in the top half, that doesn't reflect poorly on your system. If you're in the bottom 25%, you should look into.

one mor question. It says i scores in the 68% percentile. with an overall score of 11446 while the 3dmark average for mine is 11445. should i look at the average? (this is from the link you sent me with all the overclocks)

[Kisai]

5 hours ago, StunnaFan34 said:

I thought I could use benchmark tests like timespy and passmark to check if my system is performing as expected. But someone told me that you shouldnt use benchmark tests to check for that, only to see differences in overclocking. Is this true? Are tests like passmark and timespy (3DMark) not good to check for under performance? In other words, if my score is around the same as others does that mean im good?

Not all benchmarks are equal.

 

Passmark only tells you the cpu power... relative to other cpu's, just like geekbench. It's also entirely possible to fudge numbers, hence they are not reflective of real world performance, only relative performance to other cpu's.

 

Timespy, and various "game-like" benchmarks often hit a ceiling to which they no longer represent what a real world game would perform like. For example, if you dig out the 1997 Final Reality benchmark and somehow manage to get it to work (it's directx6) it will just hit a cap and GPU/CPU parts from 2020 will have the same scores as they had in 2006. You eventually hit a ceiling where the game engine can't process anything faster due to design limitations in the original game engine. You can also do this with the FFXIV (V1.0) Benchmark. It just generates the same score regardless of the machine once you use past a certain point.

 

Hence these benchmarks should not be able to hit 60fps in their default settings, because once they do, they are no longer reflective of anything.

 

With Cinebench and similar software, these actually use the same engine used in the commercial product that it's designed for, so they are far more reflective of the CPU power in that software, and similar software, but you can't go "oh I get X score in cinebench, my Metro game should be 120fps", because these are two completely different use cases.

 

A more typical thing is to use passmark/geekbench to benchmark the CPU and compare it to other people with the same CPU. The actual tests are designed to run in the CPU's L1/L2 cache so that the system memory and chipset do not affect the score. You can't really do the same with GPU's since there is no standard means of doing this. Even an OpenCL/Vulkan program will generate different results on different driver, gpu and memory configurations.

 

If you are overclocking, then just use the benchmark to determine if the settings you are using are better/more stable than the previous.

[StunnaFan34]

20 minutes ago, Kisai said:

Not all benchmarks are equal.

 

Passmark only tells you the cpu power... relative to other cpu's, just like geekbench. It's also entirely possible to fudge numbers, hence they are not reflective of real world performance, only relative performance to other cpu's.

 

Timespy, and various "game-like" benchmarks often hit a ceiling to which they no longer represent what a real world game would perform like. For example, if you dig out the 1997 Final Reality benchmark and somehow manage to get it to work (it's directx6) it will just hit a cap and GPU/CPU parts from 2020 will have the same scores as they had in 2006. You eventually hit a ceiling where the game engine can't process anything faster due to design limitations in the original game engine. You can also do this with the FFXIV (V1.0) Benchmark. It just generates the same score regardless of the machine once you use past a certain point.

 

Hence these benchmarks should not be able to hit 60fps in their default settings, because once they do, they are no longer reflective of anything.

 

With Cinebench and similar software, these actually use the same engine used in the commercial product that it's designed for, so they are far more reflective of the CPU power in that software, and similar software, but you can't go "oh I get X score in cinebench, my Metro game should be 120fps", because these are two completely different use cases.

 

A more typical thing is to use passmark/geekbench to benchmark the CPU and compare it to other people with the same CPU. The actual tests are designed to run in the CPU's L1/L2 cache so that the system memory and chipset do not affect the score. You can't really do the same with GPU's since there is no standard means of doing this. Even an OpenCL/Vulkan program will generate different results on different driver, gpu and memory configurations.

 

If you are overclocking, then just use the benchmark to determine if the settings you are using are better/more stable than the previous.

ah ok thank you. So I can trust my results if i compare it to people who arent overclocked?

[StunnaFan34]

21 minutes ago, Kisai said:

Not all benchmarks are equal.

 

Passmark only tells you the cpu power... relative to other cpu's, just like geekbench. It's also entirely possible to fudge numbers, hence they are not reflective of real world performance, only relative performance to other cpu's.

 

Timespy, and various "game-like" benchmarks often hit a ceiling to which they no longer represent what a real world game would perform like. For example, if you dig out the 1997 Final Reality benchmark and somehow manage to get it to work (it's directx6) it will just hit a cap and GPU/CPU parts from 2020 will have the same scores as they had in 2006. You eventually hit a ceiling where the game engine can't process anything faster due to design limitations in the original game engine. You can also do this with the FFXIV (V1.0) Benchmark. It just generates the same score regardless of the machine once you use past a certain point.

 

Hence these benchmarks should not be able to hit 60fps in their default settings, because once they do, they are no longer reflective of anything.

 

With Cinebench and similar software, these actually use the same engine used in the commercial product that it's designed for, so they are far more reflective of the CPU power in that software, and similar software, but you can't go "oh I get X score in cinebench, my Metro game should be 120fps", because these are two completely different use cases.

 

A more typical thing is to use passmark/geekbench to benchmark the CPU and compare it to other people with the same CPU. The actual tests are designed to run in the CPU's L1/L2 cache so that the system memory and chipset do not affect the score. You can't really do the same with GPU's since there is no standard means of doing this. Even an OpenCL/Vulkan program will generate different results on different driver, gpu and memory configurations.

 

If you are overclocking, then just use the benchmark to determine if the settings you are using are better/more stable than the previous.

basically a test wouldnt actually be reflective in performamance on games but i can still figure out if im underperforming by comparing to other build or looking up the average score of a cpu or gpu non overclocked?

[Kisai]

1 minute ago, StunnaFan34 said:

ah ok thank you. So I can trust my results if i compare it to people who arent overclocked?

Pretty much.

 

If you have an i9-10900K and another person has a i9-10900K and the scores are within 1% of each other, then that's probably where it should be. If your score is more than 5% off, then that other score might be overclocked.

[StunnaFan34]

Just now, Kisai said:

Pretty much.

 

If you have an i9-10900K and another person has a i9-10900K and the scores are within 1% of each other, then that's probably where it should be. If your score is more than 5% off, then that other score might be overclocked.

ah ok so through comparison on official websites i can see if im performing alright?

[Kisai]

Just now, StunnaFan34 said:

ah ok so through comparison on official websites i can see if im performing alright?

yes

[StunnaFan34]

Just now, Kisai said:

yes

ok thank you. Sorry i just want to make sure. Thank you for the detailed response

[StunnaFan34]

18 hours ago, Kisai said:

yes

just looking back at the forum i just wanted to ask one more thing. So do you meean by your first message that comparing GPUs is bad because we get different results for different drivers? Does that mean if I compare my Graphics score results with other people with the same GPU it wont be useful?