Hacker Newsnew | past | comments | ask | show | jobs | submit | more sydney6's commentslogin

I can imagine that the power draw in S3 could come from having (replacable) DDR Memory instead of (mostly soldered) LP-DDR Memory.

edit: typo.


The problem is that current Intel laptops don't use S3 anymore, they use S0ix a.k.a. "modern standby", an abomination where the CPU doesn't really sleep and the battery drains fast.

Dell, Lenovo, HP etc. all have the same problem.


I have a Thinkpad X1 from 2018 and by default it came with S0ix enabled and Lenovo later on added the S3 sleep state option through a BIOS update, called "Linux compatibility something".. Before that, one had to manually edit the DSD table to get rid of this evil burning-sleeping-laptop-in-backpack-feature called S0ix.

Does the Framework Laptop, or other popular models from the other manfacturers you mentioned, not have a S3 sleep state option these days, i.e. S0ix only?


So it was believed that S3 is deprecated on TGL, but it probably works. I remember reading about it on some coreboot channels. Starlabs may have enabled it. Grep for S3 on https://support.starlabs.systems/kb/bios-and-firmware/bios-a...


S0ix has stronger requirements on the "correct" interaction between firmware/BIOS and OS, it offloads more work to the OS. Poorly implemented S0ix will drain the battery faster, but correctly implemented S0ix is as good as S3 or even better.

Lenovo put out a buggy S3 implementation on some systems that hasn't been tested well because it's only an optional "Linux suspend" setting. Drains twice as fast as Linux with correctly implemented S0ix. And the worst thing is, nobody except Lenovo do can fix it because it's all on the BIOS level, and their China-based firmware team has other priorities.

Well-implemented S3 is nice. But it's going to disappear. Both Intel and AMD are switching away with full force, vendors won't have S3 options in the BIOS going forward and the ones that remain will likely suck. On the other hand, S0ix support is coming together even on AMD platforms which were a little late to the party. Once it's working decently, I'd rather trust my OS than my laptop manufacturer's firmware team to suspend components correctly.


The most common reason i've heard has been: "Yeah, but it's so much more biutiful than Windows 10." But then again, this appears to work quite well for Microsoft.


Perhaps AWS just hasn't paid their networking egress fees to Amazon, you know?


> "The punk ethos is primarily made up of beliefs such as non-conformity, anti-authoritarianism, anti-corporatism, a do-it-yourself ethic, anti-consumerist, anti-corporate greed, direct action and not "selling out"."

Substitute "Punk" for e.g. "Group" and read that sentence again.


And?? With the substitution reads like someone describing a punk/anarchy group.


I think he means something like this: (monty Python)

https://youtu.be/KHbzSif78qQ?t=35


E.g. accelerating video decoding/encoding, OpenGL applications, etc..


RedHat's NetworkManager already does something pretty similar [1].

[1] https://wiki.archlinux.org/title/NetworkManager#Checking_con...


Backblaze, IIRC, doesn't distribute data across multiple regions.


People aren't skeptic about EVs, people are sceptic/afraid of change.


I think the question here isn't "is this a good laptop?", but rather "is this business model sustainable for the company?". I truly hope so..


To get a grip around the numbers, i quote from George Neville-Neil's Talk at BSDCan '15 "Measure Twice, Code Once" [1]:

- 10 Gbps is 14.8 million 64 byte packets per second - 67.5 ns per packet or 200 cycles at 3 GHz - Cache miss is 32 ns

[1] https://www.youtube.com/watch?v=LE4wMsP7zeA


You're not running a continuous stream of 64-byte packets in a home or SME setup. Also, assuming a 1:1 mapping to packet processing is a false dichotomy these days, NICs are doing an unbelievable amount of preprocessing, particularly grouping related packets together.


No, of course not. A good starting point for real world performance benchmarking could be e.g. IMIX [1].

The example above represents the solely theroretical worst case as a means to establish a baseline for performance benchmarking.

Anyway, if you are referring to HW offloading capabilities of "modern" NIC's, using techniques like LRO would break the "end-to-end"-principle of a router.

[1] https://en.wikipedia.org/wiki/Internet_Mix


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

Search: