Passmark Questions GPU testing accuracy, linear?

[CryoSage]

hello, and how are you all. I had a few questions i was hoping you could help me answer.

Is the GPU Passmark Test a "linear" test? and what defines a "linear" test?

how accurate are these benchmarks? i see that the GTX680 out shines the GTX690.... what is the reasoning behind that?

How much do you all trust the benchmarks when you are comparing performance?

Thank you for all of your help, as these are BURNING questions in my head. Cheers.

[David (PassMark)]

Is the GPU Passmark Test a "linear" test? and what defines a "linear" test?

Is isn't logarithmic or exponential, if that is what you are asking? If you double all aspects of performance, then the score will roughly double.

i see that the GTX680 out shines the GTX690

See this old post
http://www.passmark.com/forum/showth...690-on-the-top

How much do you all trust the benchmarks

See this page,
http://www.videocardbenchmark.net/graph_notes.html

[CryoSage]

Thank you very much for your quick reply. These links are very informative, and much appreciated.

However, I am still struggling to identify whether or not the passmark tests are what professionals in the field would call "linear"

The reason is because I have a buddy of mine that keeps insisting that passmark is indeed linear in nature. I am struggling to find the definition of a GPU test that is linear, vs. non linear. any help would be greatly appreciated.

And lastly, I just want to know that the passmark benches are indeed the GO TO benchmarks for overall performance and reliability, despite the scaling with different setups / mobos / ram / OS's, etc.

personally, after everything I have learned i think that these test give you a BETTER overall reflection of a CPU / GPU since they are a sum of many tests, and not just a single system being tested. what are your thoughts?

[David (PassMark)]

There is no fixed definition of what a linear benchmark is. It isn't a commonly used term (in benchmarking). So I suspect people have different idea about the meaning. Some people might interpret it as, if you over clock your CPU by exactly 10%, does the CPU benchmark score also increase by 10%. Or if you double the number of cores then does the CPU benchmark result also double, etc.. In this sense nothing will ever be linear. If for example you keep doubling the number of CPU cores (in isolation) then eventually the RAM, Cache & bus speed will limit performance increases. Leading to logarithmic or even negative performance increases.

No single benchmark can reflect the performance of a system in all possible usage scenarios. If for example you only purchased a PC to play World of Warcraft, then the best benchmark is to use World of Warcraft itself as the benchmarking tool, as nothing else matters.

If you want to do lots of different things with your computer, then a general hardware benchmark, like PerformanceTest, is a good option. (As it doesn't make sense to install 20 different application and individually test each one on all the hardware permutations).

[CryoSage]

I see what you mean. you have been of great assistance, and thank you again for your support and help. take care comrade. this community seems to be quite amazing and generally wholesome. cheers.