Hacker Newsnew | past | comments | ask | show | jobs | submit | lemonish97's commentslogin

Or maybe it is just better?


I doubt it.

There’s three possibilities.

1) Intel is optimising for common cases inside the most dominant desktop operating system.. this is like apple having really good floating point in their cpu’s that makes javascript not suck for performance… and is why macbooks feel snappy with electron.

2) Intel and microsoft worked together when designing the CPU, so Windows is able to take advantage of some features that Linux is only just learning how to handle (or learning the exact way it works).

3) The way the operating systems schedule tasks is better in this generation for Windows over Linux, by accident.

“it’s better” doesn’t really factor, Windows has been shown repeatedly over the last half-decade to be so inferior as to be beaten by Linux when Linux is emulating Windows APIs. It’s difficult to be so slow that you’re slower than someone emulating your platform.


Neither of your three possibilities refute the parent hypothesis:

> Or maybe it is just better?


No, but if you continue reading the comment, you’ll read why I believe that’s not the case..


Well that too isn't exactly correct; Windows isn't getting beaten by Wine/Proton on Linux by any significant (nontrivial) margin; at best they're at par, and most of Linux's advantage comes from not having to bear the load of a thousand background processes (unlike Windows) when it's running a Windows app. I perfectly understand the appeal of desktop Linux being an avid user myself, but let's be real here, it's not very likely to run a Windows-native app or game much better than a debloated/LTSC Windows setup.


LTSC narrows it, sure. But 2025 benches prove Proton/Linux beats stock/debloated Win still. [0][1]

Handhelds crush: Bazzite/SteamOS +20-36% FPS on ROG Ally X/Legion Go. Cyberpunk 39FPS@20W Linux vs less Win; Returnal 33 vs 18. [2][3]

AMD desktop: CachyOS +15-80FPS (Cyberpunk/CS2) on RX5700XT vs Win11. Smoother 1% lows too. [2]

NVIDIA par/mixed, tweaks needed.

Vulkan+NTSync > DX overhead. 90% Steam games? Linux wins now.

Anti-cheat blocks the rest.

[0]: https://www.reddit.com/r/linux_gaming/comments/1pxtcv3/rx_57...

[1]: https://gamersnexus.net/gpus/rip-windows-linux-gpu-gaming-be...

[2]: https://arstechnica.com/gaming/2025/06/games-run-faster-on-s...

[3]: https://www.notebookcheck.net/Asus-ROG-Xbox-Ally-with-Bazzit...


I'd wonder about the performance impact of windows defender and the like.

I'd really wonder if one took a game that was on both xbox and linux. constructed a linux box to have basically as close to specs as on the xbox and then benchmarked the games against each othe, what would would see.

I'm not saying that linux is better than windows or that windows is better than linux, just that I think its very hard to make an apples to apples benchmark comparison and there are constant services on windows that run that one doesn't generally have running on a linux system that can cause problems.


I agree, but it sort of doesnt change anything.

You buy Windows as a product, and those subsystems are so spidered in that turning them off is not possible, and if it even was possible it would have some impact.

You buy Windows for games, thats been the consensus for years, the NT kernel could in theory run games 10x better, but it doesn’t mean anything because you only get it, with Windows.

So, an Apples to Apples comparison is Bazzite. The general purpose operating system you install and play games on. No need to apologise for Microsofts choices.


I don't see a comparison in [1]


The first video


It’s easy to be better when you only implement the mainline of the API


Glib, and not true.

Otherwise Windows could make WSL (1) faster than Linux, but they can’t because they don’t have the similar enough underlying operating system paradigms.

I could give examples, but I think just comparing native python performance on both platforms is the easiest case I can make without going into details.


WSL 1 was faster and better than WSL 2, but they abandoned it for its technical complexity and switched to containers / virtual machines, which create a slew of Other Problems.


yes, but was it faster than Linux?


I have no clue, I just know it was faster than Windows.


"Intel is optimising for common cases inside the most dominant desktop operating system."

- literally the history of Intel for more than 30 years and likely why we see this benefit now. gaming the compiler and hoping they wont get caught bought them a decade against AMD.

"Intel and microsoft worked together when designing the CPU"

- I guess the bitterness of Itanium doesnt last forever.


On only one laptop?


> On only one laptop?

That's how a good benchmark looks like. From ancient wisdom (Linux Benchmarking Howto): " 5.3 Proprietary hardware/software

A well-known processor manufacturer once published results of benchmarks produced by a special, customized version of gcc. Ethical considerations apart, those results were meaningless, since 100% of the Linux community would go on using the standard version of gcc. The same goes for proprietary hardware. Benchmarking is much more useful when it deals with off-the-shelf hardware and free (in the GNU/GPL sense) software. "


The oddity is that Windows is slower everywhere but on this one specific kind of laptop, as far as I understand. If it's not a quirk of the laptop, windows would be better everywhere.


Not sure if it's a TPU constraint, but according to this report it seems like the Gemini models have really poor TTFT and tps inference times.


It's great to see how we went from the first iteration of Claude Computer Use, to now being able to run it locally with just 7B params.


Yes, they mention that the model is aka nano-banana in the blogpost


Never thought I'd run an LLM released in 2025, on my phone, in full BF16. With ~80tps on an iPhone 16 pro btw.


How do you actually run this on an iPhone?


With something like PocketPal AI (https://github.com/a-ghorbani/pocketpal-ai). I'd love hear HN'ers opinions on the "best" LM Studio-like option for iOS devices.


For iOS, OpenCat. Has iCloud sync, and one universal app for MacOS and iOS devices.


I use PocketPal. Can run any gguf model off hf.


you can also run it on Cactus - either in Cactus Chat from the App/Play Store or by using the Cactus framework to integrate it into your own app


Honestly love gemini for zero-shot coding. But for some reason I'd still lean towards GPT4o for just natural conversations and day-to-day queries. Something about 4o's tone and behavior just clicks with me.


To be fair, HP buying Compaq in the early 2000s was pretty huge, considering they are doing decently well today because of their PC market share.


used this throughout my ML on graphs grad course this sem, it’s a really neat library


Fun fact, it is believed that Andrej Karpathy (OpenAI, ex AI Dir. at Tesla) had coined the word 'hallucination' in an RNN blogpost back in 2015. https://x.com/karpathy/status/1702916988891193460?s=20


Thought this was some ‘Proof of Work thing’. Def need a break from HN.


Me too but in fairness "Prisoner of War" doesn't come up in normal life all that much either.


Sadly in recent years it has definitely surpassed Proof of Work as a discussion topic.


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

Search: