For the best experience on desktop, install the Chrome extension to track your reading on news.ycombinator.com
Hacker Newsnew | past | comments | ask | show | jobs | submit | history | nexle's commentsregister

Breakdown of the (semi-clickbait) 208MB cache: 16MB L2 (8MB per die?) + 32MB L3 * 2 dies + 64MB L3 Stacked 3D V-cache * 2

For comparison, 9950X3D have a total cache of 144MB.


> 16MB L2 (8MB per die?)

It is indeed 8MB per compute die but really 1MB per core. Not shared among the entire CCD.


I wouldn’t be caught dead with less than 200MB of cache in my desktop in 2026.


they already paid 10x more to their lawyers to ensure that torrenting for LLM training is perfectly legal, why they want to pay more?


> Singapore is an odd country

The reason you find it odd is because you really can't find another country that the citizen have such a high trust towards the government and let the government do (almost) anything they wanted, yet the government doesn't abuse this power (mostly, at least) and continue focus on long term benefits of the country (rather than short term gains because the political party need to survive the next election in few years time)

> One of the most militarized countries (#3 by military spending per capita) in the world, yet their military has barely been used.

Ther reason is quite simple: Singapore is a very small country and it is very easily to be invaded. The high military spending is more of a deterent.

> What would you even call their socioeconomic system?

It is very much a free market capitalism with some state intervention, similar to many other countries. If anything, I would say Singapore is more free market than many western countries due to the fact that the government is very pro-business as the country is heavily rely on foreign businesses to survive.


No need to abuse anything until shit hits the fan.


High Yield has a video that deep dive into the 395 chip on the silicon level: https://youtu.be/maH6KZ0YkXU


It is mindblowing how AMD managed to squeeze such a powerful iGPU (essentially a low-end dGPU) into an APU while much more energy efficient compared to a dGPU (and thus requires less heat sink = more compact).

Of course the major problem about Strix Halo is the price. I'm just wondering how much the iGPU contributed to the insane price tag compared to the NPU. If AMD can release a similar APU without the useless NPU (at least in Linux) with a more accessible pricing (e.g. 8745HS), they can easily dominate the low-end mobile dGPU market.


The pushing on their cloud offering almost everywhere (main page: https://linkwarden.app/, GitHub README: https://github.com/linkwarden/linkwarden, and installation guide: https://docs.linkwarden.app/self-hosting/installation) just give me a bad taste about it.

I understood an open source project need revenue to survive, but the reason why this project grew so large is because of the self-hostable nature, and the push of the cloud offering is the opposite of that.

I really hope this is not the first steps towards enshittification...


Nah, I just see this as a sustainable way to keep the project alive :)


slightly off topic but it should be a crime for a website hijacking the back button


It should be a crime for web browser letting the back button be hijacked in the first place!


It shuld be a crime for web browsers to download and execute code as a matter of loading a page.


Nothing is "hijacked"; it just sets the hash to allow permalinks. It should probably actually load the state when pressing back (or replace the current entry instead of adding a new one). But that's just a bug and not malice, as some seem to assume.


It actually is augmenting the history so "hijack" is correct.

Weird that back isn't restoring the state. Just stays the same for me.


Thanks for the link, seems like the loophole is already there since the introduction of the package visibility restriction, and almost everyone and their mother knows how to bypass this restriction.

> Google refuses to patch this

While I don't believe Google engineers are not aware of this widely used loophole, do you have any source that they refused to fix it?


That loophole was published 5 years ago, it hasnt been fixed since.

Do you need someone from Google to explicitly write an official note, notarized, indicating they are refusing to fix it?


> refusing to fix it

Google addressed similar isolation concerns (without breaking a tonne of APIs in incompatible ways) with Private Space and Work Profile: https://source.android.com/docs/security/features/private-sp...


If it's a security issue fix, they should release it in one of the monthly security patch.

I also think that private space do not fix the underlying issue. If you have four apps and you don't want them to know about each other you can put one of them in main profile, work profile, app locker and you run out of profile for the last one. The way app locker work doesn't scale to tens of sandbox.


I know you didn't ask for this sort of answer, but you could use user profiles for this.

You can have more users on the "standard" AOSP Android as well, but with a certain AOSP-derived you can also have notifications forwarding.

Until they add Application List Scopes (I believe it's on the road map), in the exactly the same way users can now lie to apps they have only specific contacts in their contact list and only one or two specific folders in the Storage.


that proves bad faith.

they keep releasing overly complicated features to sidestep the obvious reported vulnerability, to silence power users and please corporate enterprise sysadms.

the rest of the 99.9 of users keep the vulnerability, which is very profitable for ad networks. wonder why an ad networks who maintains android would do that.



That's a lot of steps just to convince Ubuntu to stop shoving snap through our ass. More the reason to switch away from Ubuntu.


Consider applying for YC's Summer 2026 batch! Applications are open till May 4

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

Search:

HN For You