Hacker News new | past | comments | ask | show | jobs | submit login

Computer architectures routinely see 3x performance jumps across different power budgets. This rule has held over decades.

Clock speeds alone can probably increase by 30%. Caches and internal datapaths can double or more. Then you can start to add in more execution units or more expensive branch-prediciton or even new more power-hungry instructions.

A 4 Watt Intel Pentium 4410Y Kaby Lake for mobile devices gets about 1800 on Geekbench, while a 115 Watt Intel Core i7-7700K Kaby Lake for desktops gets 5600.

I'm just going to say it: the Apple laptop CPU is going to get Geekbench score... above 9000!

And, yes, I do have a CPU design background.




So artificial benchmarks already do a very poor job of capturing performance. The apple laptop cpu does not exist. If it did exist it would likely suffer a very substantial performance hit if forced to emulate x86 software. So you are going to speculate on the meaningless benchmark numbers of an imaginary cpu that will take a wholly unknown hit if everyone does not rewrite everything why?


Artificial benchmarks do a great job of capturing performance, since they're more controlled and eliminate unnecessary variables.

Once you understand this, then you can understand how CPU designers work to predict future performance. CPU designers use artificial testbenches.


You making up numbers doesn't appear to be a useful endeavor.


I suspect if Apple designs a desktop CPU, performant x86 emulation will be a key design criteria. I know very little about CPU design, but I imagine it would be possible to have hardware optimisations for x86 emulation just like we have today for video codecs.

Or even further they could bake a "rosetta" into the chip's microcode and have their CPU natively support the x86 instruction set along with ARM or whatever they come up with.


Which is the previous gen and was sandbagged as the 50% increase in cores for coffee lake show.


If someone disagrees they should state why




Join us for AI Startup School this June 16-17 in San Francisco!

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: