Sources:>https://www.techpowerup.com/347906/intel-binary-optimization-tool-changes-code-execution-with-heavy-vectorization>https://www.tomshardware.com/pc-components/cpus/geekbench-investigates-up-to-30-percent-jump-with-intels-ibot-performance-gain-attributed-to-newly-vectorized-instructions>https://videocardz.com/newz/geekbench-6-7-released-with-core-ultra-200s-plus-bot-detectionWhy are people arguing if it is cheating or not when the real issue is that this is an explicit arbitrary remote code execution vulnerability? I mean, even assuming that the code always produces the exact same same results, it is still a very lame coverup for allowing them to also do anything they want as long as the user does not notice it.
>>108562502>NOO YOU CAN'T MAKE MY COMPUTER FASTER IT'S UNFAIR
>>108562502Isn't like 90% of a gookbench score:A) How fast you can hardware encode/decode JPG.B) How fast you can hardware encode/decode H.264.C) How fast you can hardware encode decode AES encryption?Anyway x265 seems like the go-to if you're not a brainlet since it just tests how fast the actual CPU is, not hardware encoders.
>>108562521I just made my computer faster with -mcpu=native --param=l1-cache-size=32 --param=l1-cache-line-size=128 --param=l2-cache-size=1024 -falign-loops=32:7:16:3 -falign-jumps=16 -O2 -fvect-cost-model=dynamic.Thank you for your attention to this matter.
>>108562601would compiling the kernel with these flags help much
>>108562568likely would to have 8K pron watch it with 720P tv and compare gpu temperatures
>>108562641Probably not, aligning loops and jumps can get you like 7-8% performance uplift depending on the workload, but it can also make your performance worse, too.
>>108562641The Kernel in particular is already fine tuned with lots of compiler hints and layout decisions.Even in generic terms, using explicit cache and alignment values can backfire horrendously if there is a mismatch. I would only use -march=native, which implies mtune, instead.
>>108562502hyper threading led to spectre and meltdown, what nuclear clusterfuck will this corner-cutting lead to?
>>108562521Replacing all the code with mov [score], MAX_SCORE doesn't make the former faster and is called cheating.
