1
At half the performance, which is an irrelevant comparison here. Generally AArc...
source link: https://twitter.com/andreif7/status/1415341266506854405
Go to the source link to view the article. You can view the picture content, updated content and better typesetting reading experience. If the link is broken, please click the button below to view the snapshot at that time.
Don’t miss what’s happening
People on Twitter are the first to know.
Tweet
See new Tweets
Conversation
Please note I'm referring just to things that completely didn't make it or got ground-up overhauled, and not to new things introduced in a64 -- those are orthogonal to my point. amd64 was made with one goal in mind -- max BC while introducing 64-bit addressing and regs; a64 was \
made with another goal in mind -- a next-gen ISA with zero BC.
What kind of practical win does dropping backward compat give? BC doesn't look expensive, while being really important for adoption (see vmware example. and to go further, linux and windows)
The practical win is a64 can reach perf levels a32 never could.
Practically x86-64 designs also reached perf levels that x86 (32-bit) couldn't. Most of it looks unrelated to ISA (bigger ooo structures/caches, better predictors). Same for A32/64, though A32 may have had more weird things (predication) that made high perf design difficult
By 'never could' I mean implemented as a modern-day uarch, not in comparison to something from the '80s when ia32 got introduced. Predication in a32 is but one example -- register sizes and organisation, mem consistency models, POD types, most useful ops -- our understanding of \
Can you bring evidence to support that claim? I can believe there's some perf impact, just very very little compared to the big fish (better predictors, bigger caches, bigger core structures, and process improvements allowing all that at higher clocks)
Just check any of IPC charts on http://anadtech.com -- pay close attention to the gen-over-gen perf gains of a certain fruit SoC vendor using a64 as their sole ISA.
That's not evidence of a64's impact. Pwr/perf targets are a better explanation. AT's A14 review shows S865/A77 achieving better power efficiency despite being a process node behind.
At half the performance, which is an irrelevant comparison here.
Generally AArch64 should be +10% perf.
Where does that +10% perf figure come from? How much of that comes from extending to 64-bits?
Perhaps ISA can have a notable impact if you have enough cruft affecting common instrs. Predication may be a culprit. But A32 vs A64 is not something I looked into
From the people building the cores.
Show replies
Relevant people
Follow
Gamer who analyzes tech too much
Software engineer at Microsoft, opinions here are my own
Follow
An ISA archaeologist by day, an ISA futurologist by night.
Trending now
Trends for you
Trending in South Korea
산리오 콜북
20K Tweets
Food · Trending
1,432 Tweets
Trending in South Korea
5,344 Tweets
Trending in South Korea
정은경 청장
20.6K Tweets
Trending in South Korea
1,780 Tweets
Show more
Recommend
About Joyk
Aggregate valuable and interesting links.
Joyk means Joy of geeK