Benchmarks

Time it takes to solve an every day problem, like saving a document as a PDF or decrypting an HTTPS response.

>>59189629
>Use each instruction on the cpu in asm a few million times moving data around and preforming on it, this would give you a baseline of each instruction's actual performance
>load cache full of something designed to fit in cache and call it a million times
>load ram full and flip bits, see if ram fails total checksum of what you've loaded
>average all of these into individual scores, do not combine anything, show the actual performance of each instruction
>Render some opengl stuff
>Upload a bunch of textures to gpu using multiple pipelines (ati fails here)
>check cpu for clock drift (old amd fails here, and new may still, not sure, nobody tested yet) and if drifts, invalidate results as they are i n v a l i d unless using an alternative clock source to monitor the time it takes to benchmark stuff
>top it off with a file transfer test from a in-memory driver proxy caching a file loaded from disk and then flushed each test, so after about a million reads of the file you have ram perf and northbridge IO perf if the physical storage medium remains the same thats a nice baseline

Fuck benchmarks

Just buy Intel and Nvidia for guaranteed quality and performance.

>>5919033
I'd call this bait if it weren't a legitimate opinion

>>59190285
This is exactly the sort of shit I'm talking about. These are actual metrics, but how will the average /v/irgin migrants that infest this board ever understand what those numbers mean?
It leaves them in the dark but gives people who actually give a damn food for thought. It's like starving the intellectually dull, something society should be doing as a whole.

>>59190285
These are the kinds of tests we need to see. Sadly all these fucking TechTubers do is gaymes and "rending" since everyone and their mom makes shitty youtube videos or "photo editing"