Are Passmark scores linear?

[nerd86]

ie. does a GPU with 12,000 points have 20% more measured performance than one with 10,000 points, for example? Are there diminishing returns on # of points earned? Does the test lose accuracy at extreme values (very high-end CPUs or GPUs)? Or does % higher score = same% more measured performance?

I understand there are a lot of nuances to measuring performance (in what context? etc.) of a piece of hardware. I'm mainly wondering, in the context of a Passmark test for CPU or GPU, does a % higher score indicate the SAME % higher measured performance or does it start requiring more and more performance to get fewer and fewer additional points as you climb up the scale (or is it the opposite??)


ie. does a GPU with 12,000 points have 20% more measured performance than one with 10,000 points or are there diminishing returns as you add more horsepower to a card...Or does the score start giving you exponentially higher results as you approach the extreme end?

How reliably can I look at Passmark results and say "X CPU benches 30% higher than Y CPU therefore X CPU is being measured by Passmark as being 30% faster"?

I know there are a lot of nuances to measuring performance and I'm not interested in that. I'm merely interested in comparing Passmark scores of different pieces of hardware: Does an X% higher score mean Passmark measured that device as being X% faster?

[David (PassMark)]

Are Passmark scores linear?

Roughly yes.

does a GPU with 12,000 points have 20% more measured performance than one with 10,000 points

Roughly yes.
But really depends on what software / game you are running. Some are really dependant on CPU performance, RAM performance, etc.. In other cases the scaling might depend on the monitor resolution you are running and vsync setting. So not all software will scale in an identical fashion to the 3DMark.

It is the same for CPUs. Each piece of real world software is different. Some are single threaded, some are multi-threaded to 4 threads, some are massively multi-threaded. Some are dependant on RAM or disk speed & some not.

The benchmark numbers are based on linear units of measurement. e.g. mathematical computations per second, frames rendered per second, megabytes transferred per second, etc..

[movieguy73]

i was wondering that too.
how do the scores from a 6 year old cpu compare to a new score?
ie.
https://www.cpubenchmark.net/cpu.php...00+%40+3.40GHz

http://www.cpubenchmark.net/cpu.php?...00+%40+3.40GHz

[David (PassMark)]

Yes, performance improvements & clock speeds have somewhat stalled over the last few years. Especially single threaded performance.

BUT, there are two other stories going one.
1) Integrated GPUs have been getting significantly faster.
2) Electrical power usage has been going down.

So for the two CPUs in question, you get 20% more power, at 30% less wattage, plus a 300%+ improvement in GPU performance.