Hacker News Reader

Comment by ilaksh | Hacker News Reader

Are PC laptops going to catch up? I wonder if there is some kind of conflict of interest with Nvidia holding back more capable PC chips or software integration.

View on HN

>Are PC laptops going to catch up?

x86 has to scale from embedded boards to laptops to servers. AMD made different trade offs than Apple.

Apple chose to achieve a level of performance by keeping the frequency low and having a high IPC. High IPC means that cores consume a larger size of the die. That helped with lower power consumption.

x86 makers like AMD, did the other way around. They dedicated a smaller portion of the die to a core. That makes it cheaper. To make up for that, they used a higher frequency, which also means a higher power consumption.

Apple also has a node advantage, which also makes their CPUs more expensive.

AMD could totally do what Apple did: have faster IPC at lower power consumption if they would have release more expensive CPUs targeted at the mobile market.

But AMD and Apple have different goals.

If we stick to the mobile world, Qualcomm is doing with their Orion cores just what Apple did with their M cores. In the first half of this year we will see laptops with Orion cores that will have both strong IPC and low power consumption.

>AMD could totally do what Apple did: have faster IPC at lower power consumption if they would have release more expensive CPUs targeted at the mobile market.

Doubtful. I'm not sure they have the talent/knowledge to pull it off while Apple Silicon is a moving target with a new generation introduced in each new iPhone.

AMD and Apple are in a different business. What makes sense for Apple doesn't make sense for AMD.

And AMD employed some of the most talented people in the area like Jim Keller. They totally can do it. Intel, too. Qualcomm is doing it with Orion and Qualcomm has less expertise and resources.

Well, Qualcomm at least claims that. Whether they can deliver is to be seen. Wouldn't be the first time they fail to do so...

Granted, Keller was at Apple for a while too...

Qualcomm literally bought the team if Apple engineers who led the M1 design in Nuvia. That’s Oryon.

For the last 10 years, Qualcomm couldn't come close to Apple designed SoCs. Now they might come close after buying a team of ex-M1 designers.

It sounds ridiculous to say that AMD, Intel, and Qualcomm can do what Apple is doing. If they could, they'd be able to compete with Apple Silicon now. It's been 3 years. It's like saying AMD, Intel, Qualcomm can do what Nvidia is doing if they want to.

Processor design isn't black voodoo. Its all configuration, most of which is now designed with CAD, and targeting particular use cases.

Its kinda like the evolution of bitcoin mining-everyone was doing it with GPUs, AMDs were better because of some hardware optimization where they could hash something in less cycles, then people started making ASIC which are purpose built and the fastest possible way to run the necessary compute. There were no technical advancements here, just economic forces driving the design and manufacture of components that made making ASICs profitable.

Apple did the same thing except they designed the chip and the computer as one, and they succeeded because it allowed them to optimize stuff further, and you see this in results where they are the leader but not by much, and often behind systems that are overclocked in single core performance. Power consumption on macs, also varies greatly depending on what you are running, which is again a result of optimization for particular use cases.

The iPhone is the most profitable product in the world. It gets massive resources to design a new SoC every single year. That resource spills over to the M series which is just based on whatever the A series does that year.

AMD does not have that.

And yet, despite the advantages if all the Apple silicon, its not being used to run the backbone of the web, while AMD and Intel are. Making something work well in general purpose is just as "advanced" as making something super optimized.

Apple Silicon isn't designed to be server chips. The M1 team thought it was a good idea but management did not want to do it. So some of the team broke off and started Nuvia to design ARM chips for servers. Qualcomm acquired Nuvia and that became the Qualcomm Oryon SoCs.

Anyways, ARM chips are being used for the backbone of the internet. They're called Amazon Graviton chips. Ampere Computing also has ARM server chips. So do many Chinese hyperscalers.

> Apple also has a node advantage, which also makes their CPUs more expensive.

M2 is on TSMC N5. AMD's Zen 4 mobile processors are made with TSMC N4. From die shots, I believe the P-cores in the M2 are about the same size as Zen 4 cores. And yet Apple's M2 laptops are a good deal more power efficient than Zen 4 laptops.

> a good deal more power efficient than Zen 4 laptops

Zen4 is competitive when it comes to actual computing. When doing (almost) nothing, the Apple system comes out on top, but that also has a lot to do with its software.

Yeah, Apple have been massively optimizing for heterogenous cores and having efficient power states when not doing any work.

I think macOS is also just better at actually being idle when the user isn’t doing anything. Windows seemingly always has something shuffling around in the background.

This is definitely part of it, as anyone who’s run Linux and Windows on the same machine can probably attest Windows is constantly doing… something? It is quite odd as it doesn’t seem to offer much in exchange for all these wasted cycles.

> Zen4 is competitive when it comes to actual computing.

Source for this?

Geekbench and SPEC heavily favors Apple Silicon in both raw performance and perf/watt. The perf per watt is 2-3x.

The only CPU benchmark Zen4 wins in is Cinebench R23, which uses Intel Embree engine inside and heavily optimizes for x86 AVX instructions.

> x86 has to scale from embedded boards to laptops to servers

Isn’t this also true of ARM?

Yes, though Arm is probably a bit more robust at the lower end (think microcontrollers and very low-power industrial cores), and a little weaker at the high end (per-core performance on Neoverse still doesn't seem to match AMD/Intel).

Spot on assessment. In fact, AMD has gone the opposite direction by shrinking die size through their Zen 4c architecture (https://news.ycombinator.com/item?id=36458678).

To me the more interesting aspect is how memory architecture will be approached going ahead by both parties. Apple can afford to go more bespoke and integrated, but how can x86 maintain a good mix between performance and design flexibility.

Is software ever going to become more responsive? I'm tired of Slack constantly fucking up its UI because I'm faster than it is, and that's on an M2.

Are people going to be less lazy and use a compiled language like C, C++, Rust instead of Javascript for desktop and mobile apps?

Are companies going to pay for that development?

Nobody really cares.

I’m not usually the type to complain about input lag; most modern apps on a recent machine are snappy enough for me.

But I’ve been messing around with egui lately, and it just feels… fast. It’s weird.

But nobody really cares.

unlikely

Who will develop a cross-platform UI system as capable, flexible and well-embedded as HTML/CSS and as responsive as native UI? Google did its shot with Flutter, it's not exactly eating the world. The performance issue will long be solved before somebody comes up with a capable contender.

> The performance issue will long be solved before somebody comes up with a capable contender.

I wish I had such a positive outlook.

New highways are built and they are quickly filled with traffic.

Trying to do that with Slint: https://slint.dev No need to use compiled language like C++, still possible to use a language like JavaScript/TypeScript, just no need to bring a browser.

Flutter suffer, badly, to be tied to a specific lang.

If it where an API directly usable by any lang, then it could fly further.

There is https://gioui.org/ for Go.

This doesn't seem to cover a single point in the comment you're replying to.

> Who will develop a cross-platform UI system as capable, flexible and well-embedded as HTML/CSS and as responsive as native UI?

History shows that you can only choose one. If you're not using the platform's native UI stack, you're going to end up with a slower, less capable solution that will look and feel foreign and clumsy in comparison. What the industry really needs is a React Native-type solution that isn't tied specifically to React.

Isn't this what WebAssembly or Rust is used for? IIRC Figma's web app uses WebAssembly and I prefer that way more than the native app.

Maybe when native languages manage to come up with a single good UI framework

Macs have Cocoa which is a good UI framework. There's your single good UI framework for them.

iOS has UIKit, which is a good UI framework. There's your single good UI framework for them.

Linux has GTK and Qt, though I'm not sure which one is in vogue these days, but both of which seemed fine when I used them in the past.

etc

But it sounds like what you're actually saying is "I only want one OS to exist".

Every OS has its own UI API, because every perform has different UI design language, UI interaction, and UI behavior. You cannot make one UI that is good on every platform because every platform is different and those differences are what make people prefer one over the others, but a single UI is (as Java and Chrome wrappers demonstrate) going to produce a bunch of exceedingly mediocre apps that fail numerous basic UI tools on their respective platforms.

If you want a cross platform apps you have cross platform core logic, and per platform UI.

If you want a single UI framework across platforms, it cannot be good, and it will inherently at best produce the kind of mediocrity we get with chrome apps, or you only have a single platform and have the same UI toolkit across all of them: desktop, phone, tablet, etc. I'm going to guess that's going to produce a fairly terrible UX on at least one of the options if it's good on others, or maybe it will just be mediocre across the board.

Then nobody will go back to making native apps

Yes, the problem is easy crappy apps are cheaper for developers even if they're worse for users.

The alternative "single UI framework" is "everyone must run the same OS".

As long as people keep buying/using crapware apps that's all that will happen. The fact that so many people are ok with that mediocrity is depressing, but that they do so while then complaining about things being slow is actively annoying.

It is totally possible to have a "runs on anything" UI framework; it simply has to have bindings for every applicable platform. SDL does this, for example, in 2D Graphics. Java has a few "run anywhere" UI frameworks, and Microsoft has .Net Maui (as well as the 3rd party avalonia UI). The problem with most native frameworks, however, is they are nowhere near as flexible as HTML and CSS (which many pre-teens taught themselves as a way to customize their tumblr and myspace pages). Other frameworks, by comparison, require programming knowledge or more setup than just a text editor (usually requiring an IDE and the rest of your toolchain).

I can see native software making a comeback if it is:

1. Declarative: only software engineers like imperative UI systems

2. Flexible: some frameworks (like any of the XAML-based ones) are strongly typed and you can only combine certain XAML elements with other certain XAML elements, and will fail to compile entirely if you mess something up

3. Unopinionated: HTML, CSS, and Javascript don't require you to implement any particular paradigm for getting data on the screen; that is left up to the coder

4. Dead ass simple: the litmus test is that a 7th grader should be able to learn the basics of it on their own, and the tooling should be ideally no more complicated than a text editor

5. Totally sandboxed: You can't trust 99% of frontend devs with security

The people you need to appeal to are not hardcore software engineers, but hobbyists and newbies, since they end up being the majority of frontend devs anyways.

Honestly I think you could get pretty far with just adding declarative languages (read: XML) to existing native frameworks, like gtk and Win32.

Your post is thoughtful. Why do you care so much about declarative? I tried SwiftUI last summer to see what it was about and I find the whole style rather awkward. Writing a desktop app in XML sounds even worse. And I'm not seeing the benefit to the user. With an imperative coding style, you can more directly control the user experience and give them the best one.

With an imperative style, you need to learn a programming language before you can start making applications. With HTML, you can start making applications immediately by just writing HTML, and then you can learn the programming language (javascript) later. It's a barrier to entry for non-engineers

The problem is not putting bits onto the screen, that's all SDL does. It does not integrate at all with the host platform, to an even greater extent that chrome based crapware.

1. That's an opinion, not a fact, and most "declarative" UI frameworks have poor performance and can't do anything not explicitly models by the declarative framework. The fact that people want to just draw a UI doesn't mean that the resulting UI is either good, or works correctly, and the idea that it's "easier" simply implies lower understanding of what is being done and that you aren't considering any of the edge cases you still need to handle.

2. Not sure what to say here, as I don't know your point? it seems more about the type system of your language and not at all related to the ui toolkit

3. This is BS. html, css, and the DOM all have a very clear model for how and what a UI is, claiming it isn't means that you haven't apparently spent much time working in any other system. Much like I keep meeting people who say that they don't have an accent, everyone else does.

4. Application development is conceptually complicated unless your application is not doing anything interesting. I'm also curious what makes a browser ui more "simple" than say Cocoa?

5. Sandboxing is a part of the OS, not the UI framework. All applications on iOS are sandboxed, that vast majority on macOS are.

Application development is hard, and it's not because of the UI framework you're using. The overwhelming amount of any actual application is the application logic, not the UI.

Chrome based apps are all abysmal because to make an actual application using a browser engine, you need to reimplement, in html+JS, a huge amount of work the host OS would otherwise just provide for you.

If all you want is a webpage, just make a webpage, don't force everyone to download a 600Mb copy of chrome.

> It does not integrate at all with the host platform

SDL does in fact make system calls and have OS-specific code (DirectX implementation, Metal implementation), even if most of them don't directly manipulate OS UI controls, though it can initialize a window with a rendering context on any of its OSs.

I feel like you're missing the point that performance matters less than accessibility for UI frameworks. HTML and CSS are the most popular choices because a child can learn them quickly. Native apps aren't as popular because all the frameworks out there require you to be an engineer to use them.

> 1. the idea that it's "easier" simply implies lower understanding of what is being done and that you aren't considering any of the edge cases you still need to handle.

I don't see how that follows at all. HTML is declarative and powers most applications at this point.

> 2. Not sure what to say here, as I don't know your point? it seems more about the type system of your language and not at all related to the ui toolkit

Web browsers are very forgiving with HTML errors, native UI frameworks (especially those which simply convert the markup to imperative statements) often refuse to run entirely if there's a single markup error. A native framework which seeks to become as popular as HTML would need to be as flexible as HTML

> 3. html, css, and the DOM all have a very clear model for how and what a UI is, claiming it isn't means that you haven't apparently spent much time working in any other system. Much like I keep meeting people who say that they don't have an accent, everyone else does

You're not understanding what I'm saying. You can run an ajax request and get some data, but to put that data into the UI would require you to run a query on the dom and manually set the innerHTML of the element to the new data. Some other UI frameworks will force you to bind specific variables or lists to a control in order to auto refresh when the bound variable changes. This is an opinionated framework: one which tells you how to manage updating the UI rather than allowing the developer to do it themselves. Such an opinionated framework would become more difficult for people to learn

> 4. Application development is conceptually complicated unless your application is not doing anything interesting. I'm also curious what makes a browser ui more "simple" than say Cocoa?

I have no experience with cocoa because I don't use a mac, but from what I can tell it requires you to know Objective C and run XCode. HTML requires, at the very least, notepad to write. It is extremely easy to get setup with HTML and start learning it as a lay person.

> 5. Sandboxing is a part of the OS, not the UI framework. All applications on iOS are sandboxed, that vast majority on macOS are.

You would want to make sure that programs written in this framework are sandboxed away from any resources. On Windows this means AppContainers and on Linux there are different implementations (snap, docker). But the host which runs and loads the UI should run all applications sandboxed by default (i.e: like a web browser)

Gnome is 120FPS butter on Javascript.

As far as I know, Gnome profiled a bunch of their biggest UI slowdowns and none of them would have been (much) improved on native.

Bad code is bad, JavaScript isn’t the main culprit here.

This.

I've wrote plenty of (very) high performance code using a Javascript/WebGL combo.

Most people bitching about JS are not even good coders to start with.

Truth to be told, there's a lot of terrible js out there, and some of it is mine.

At least you're honest about it. I don't have any JS out there period, so your terrible JS is much better than I could have done.

sometimes I do think about my first contributions. how i could have done this and that completely differently. kind of like going to first grade again with what you know (and are) as an adult.

yeah but look at freaking chromium, 3.2GB compressed archive. That is absolutely insane, and cannot be defended in ANY WAY.

look at what they started with with webkit? chromium is an abomination the likes of which definitively proves there is no god, or that god HATES humanity, as god would have 100% for sure schmited us all down for the atrocity that is chromium.

Yeah, agree with that.

I recall reading about a project who was stripping down chromium to its bare essentials, it seemed quite nice but cannot find it atm.

Contempt culture is so ugly. The bullshit asymmetry principle is so in play, makes it so hard to reel in or temper the ardent zeal.

You're also off by 15x. Installed size is 247MB on Debian. https://packages.debian.org/trixie/chromium

Chromium is one of the most amazing fully featured bundles of computing the planet has ever built. We should be in awe. But, "this has made a lot of people very angry and been widely regarded as a bad move." Contempt is rampant.

Yes, and people often forget that web browsers provide 100% backwards-compatibility to 100s of legacy web standards. Of course that bloats things up, see also Windows.

Also, parent is maybe refering to the source code size? Which also includes a huge chunk of ChromeOS code.

you might want to look at what the source archive costs your diskwise :)

One doesn’t even need to go as far as C++ or Rust… many best in class Mac apps were written in Objective-C (which is notably slower than either) and were plenty responsive on single core PowerPC G3s, G4s, and G5s.

I think a great deal of the issue comes down to bloated frameworks, bad application code, and a prevalent attitude of nonchalance towards optimization that compounds with dev machines being powerful enough to mask most performance problems.

I think it's probably more productive to think of this in economic terms; "nonchalance toward optimization" <-> "I can write this code in an hour instead of three hours". And, from a company's perspective: "I can get this code running acceptably at 1/3 the dev cost".

Surely we’ve crossed a threshold of diminishing returns on this by now though? There has to be a point where the gains in convenience/DX/speed are too tiny for the costs involved and the weight of the average app ceases to grow.

IRC clients have been around for ages, and are typically more responsive.

Yes, but this seems fairly off topic when my complaint is about Slack and general software bloat. I know there exists responsive software, but much of it isn't.

Nah, my company full of people who can't even open an excel file are definitely going to be magically able to use an IRC client as their main form of communication.

Use Matrix? It's a lot closer to Slack UX and you can self host it.

I don't think it has a particularly snappy client, though it wouldn't be hard to beat Slack.

if you are suggesting you have employees who are unable to be using an irc client, perhaps you should suggest to management that a round of layoffs are in order?

how hard can it be? spend 5 minutes and anyone thats worth keeping can figure it out.

There will always exist (always have) bloated and non-bloated programs. I don’t think anyone here knows anything about Slack’s particular future, but I bet it will not be getting any better.

This is what I'm trying to achieve. I'm going to build a cross-platform ecosystem of high quality apps (in terms of UX and performance) based on Qt C++ and QML. Starting with an alternative to the resource-hag Electron app: Notion, I'm building Plume[1]. Got some plans for what will come next.

[1] https://www.get-plume.com/

This looks great, thank you! I've joined the wait list.

Thanks!

Maybe if AI can code a native interface quickly on whatever platform you are on today. The problem with human programming is it takes time so these cross-platform abominations are the current solution. We need a hands-off, useable conversion amongst native platform toolkits.

Why though? If JS execution is wasteful, maybe there's another way to find the waste and eliminate it.

Why would it? People are voting with their wallets that this is okay, that they'd rather have Slack and all that it has instead of some competing solution that's more responsive but lags in other areas, so there is no market pressure to improve this.

Stop accepting these BS "apps" that are just web pages being loaded in a full copy of chrome. It's mind blowing to me that so many people accept these "cross platform" apps that are just as bad today as java "cross platform" apps were a few decades ago (if not actually worse).

Just a plethora of enshittified apps developed to be cheap rather than good.

All AMD Zen 4, including the laptop 7x4x and 8x4x series support bfloat16 too.

The laptop models also include a separate ML/AI inference accelerator besides the CPU and GPU, but for now that has software support only for Windows.

If I recall correctly, these integrated "AI accelerators" from both Intel and AMD are actually rather underpowered in comparison to an actual GPU (even the iGPU), at least in their current forms. So they don't have any performance advantage, but was put in place for power efficiency.

The new amd agpus are on par.

I think most chip manufacturers have plans to do that. But it's going to take some time for that to get to market. AMD, Nvidia, Samsung, and a few others are working on arm based chips for laptops. It will be interesting to see if Intel adapts. These things don't happen instantly and Apple seems to have caught most of the industry off guard by doing such a good job with the m1 and its successors. They put a stake in the ground for high performance computing on Arm and raised the ambition level quite a bit.

Nvidia of course has the added complexity that they were going to buy Arm and then that got blocked. They had to adjust to that as well. And of course they are the GPU provider of choice for GPUs to many AI developers and researchers. On paper they have all the right tech and ideas here.

There's a recent announcement of them working on this now: https://www.reuters.com/technology/nvidia-make-arm-based-pc-...

AMD, Samsung and a few others have similar plans. My guess is there will be several options in the market in a few years. We'll have to see how good those are.

You can buy a laptop with a 4060 with 8 GB of VRAM for under $1500. For many ML tasks that’s plenty of memory for inference and training (assuming you’re willing to deal with a lower batch size).

What about power consumption?

1 hour at 40w tdp or 8 hours on 15w for a simple SD Lora training...

Qualcomm is supposed to launch new laptop chips(snapdragon x series) and they are supported to be 21% master than Apple M3 muti-core performance.

https://www.macrumors.com/2023/12/18/qualcomm-snapdragon-x-e...

> Qualcomm is supposed to launch new laptop chips(snapdragon x series) and they are supported to be 21% master than Apple M3 muti-core performance.

... while using 4 times the power of an M3.

Qualcomm has not given much details but it will be interesting to see performance numbers for 23Watt profile. Even if you can match M1/M2 performance that would be big step up from x86

Edit: looks like 23 watt performance is more or less the same https://www.anandtech.com/show/21112/qualcomm-snapdragon-x-e...

It's difficult for them to catch up when the market for Windows laptops can't bear a 2k+ price tag. Expensive Windows laptops do exist and are pretty good, but they're few and far between.

I know giant gaming laptops exist, but in order to keep that power and go smaller, the price is going to jump up in the moon.

There are plenty of expensive Windows laptops, they're just sold as "gaming" laptops with added blinkenlichten.

I mean that was my second point there, but I did enjoy blinkenlichten.

Apple’s best selling laptop, by far, is the Air. And most likely the cheapest model.

So that’s $1100 or $1200.

Yes other models go way up from there. And many PCs are sub $500. But $2k is not the normal Apple price.

With that being said, the base-model Macbook Air is absolutely designed to leave you unsatisfied. 256gb of storage is a decade-running joke, and 8gb base RAM in an age of unified memory might as well be planned obsolescence.

The upgraded models feel like a different beast entirely. Apple wants you to feel like you're justified paying for 16gb of memory and 1tb of storage because... you are. Those upgraded machines run smooth and stay running nice after multiple years of software updates. MacOS is a storage-hungry OS, 3 years of daily-driving a base-model Macbook Air will make you wish you spent $2k on your laptop.

The Air is designed to satisfy the target market for the Air, which it does, and is why they sell in large numbers. It runs Safari, MS Word, and TurboTax just fine.

For comparison, a base Surface Pro has even lower specs than a basic MacBook Air. (8GB/128GB)

But both of these devices are absolutely useable for general tasks.

    With that being said, the base-model Macbook 
    Air is absolutely designed to leave you unsatisfied. 
    256gb of storage is a decade-running joke

I spent a couple of years doing Rails and Python work on a 2018 MBP (local dev) with 256GB. It's doable, barely, if you are willing to leave a low-profile thumbdrive permanently inserted. Or if you are willing to velcro an external SSD to your laptop lid =)

8GB+256GB is fine for a lot of use cases though. If you have a remote development setup, or if you are just kind of using your laptop as kind of a glorified $1099 chromebook.

While I think it’s dumb to sell 8GB devices, I’m coming to the conclusion 256GB may not be a problem for normal people.

If your pictures are in iCloud or Google photos, your music is streamed, and other stuff is in iCloud or Google drive or whatever so most people need a ton of space these days?

I’m not arguing Apple’s RAM/storage prices are sane. They’re not. But with so much in the cloud or just being done in the browser instead of a large downloaded app i’m not sure normal users really benefit.

Paying for cloud storage may be a better deal, especially since I suspect the cloud is less likely to randomly lose data than if you forget your laptop somewhere or break it without a backup.

If you want to do video editing or a lot of programming that’s probably not gonna work. But if you’re mostly going to do “office task“ kind of things of typing and surfing and listening to music or watching YouTube maybe 256 GB isn’t so bad.

Base configuration should still be 16/512 in my opinion. But I’m not sure it’s as bad as it would’ve been 10 years ago to go low on storage.

I think the problem is Microsoft and Intel. Intel isn't too interested in making an ARM chip like the M2/M3, and Windows only runs on x86. If Linux had a significant marketshare on the desktop/laptop, this wouldn't be so much of a problem, and it'd be possible for someone to challenge Intel/AMD in non-Apple CPUs, but that's not the world we live in.

I hope they never catch up to Apple's prices per RAM and SSDs on laptops.

Snapdragon is making big claims about the Snapdragon X Elite. It is an ARM chip instead of x86, meant for the laptop market. I haven't delved deep into it yet so I can't make any comparisons or give my own take.

https://www.anandtech.com/show/21112/qualcomm-snapdragon-x-e...

On battery life? Its hard to say.

On things like GPU and LLM performance, I'd say they already have, and often at significantly lower prices.

People seem to have taken "Apple made a very performant laptop" and factored it down to "the M1/2 series of chips beats everything every time" - which is obviously not the case.

For the particular use case of "a laptop on battery that doesn't die in ten minutes" Apple is decently far ahead. But everything else you can easily beat it if you make different trade-offs.

The newer AMD chips have already caught up in terms of performance/efficiency. Software is more of a challenge.

At both ends of the spectrum? My intuition is that on the low end they’d take a hit in efficiency by not having low-power cores to schedule background tasks onto like Intel and ARM have been pushing

Fair point. They do have efficiency cores (Zen4c) in some of the latest mobile and server chips. However, even when they do have dedicated efficiency cores, it's not clear that this would be a major advantage in the Windows software landscape.

What do you mean?

Any modern CPU, made within the past 5 years or so (perhaps longer), is plenty fast, but the software either doesn't take advantage of that speed, or is does so in a very wrong way.

There is no reason for Windows to be sluggish on any mid-range CPU or up, yet here we are. I should not, I haven't used Windows for a good while, but I see demos and people are amazed by the speed of which windows open and coming from the Mac or a light weight Linux desktop, I can't say that I'm impressed.

Apple still holds the advantage that they just need to target a very small set of hardware configurations, all of which they have designed and control. It does still allow them to build the most pleasant desktop experience of all the current offering, even with less amazing hardware (mostly their GPU).

But this is HN, what people probably mean is: LLM are still better with an Nvidia GPU and neither Apples nor AMDs APIs and libraries are as good as Nvidias CUDA.

>It does still allow them to build the most pleasant desktop experience of all the current offering

I wouldn't call MacOs desktop experience as most pleasant. Not if you are a power user.

In what way? Can you elaborate?

From a poweruser PoV, the only complaint I have with MacOS is its lackluster tiling support which I need to install separate applications to fix.

What do you use or recommend for tiling?

I don’t experience this difference with Windows vs Mac performance. Actually my desktop machine is much faster in most tasks than the similar vintage M1 Macbook Pro I use for work. Probably mostly due to faster storage in the desktop.

Which laptop with AMD CPU I can use to watch movies for 20 hours or browse the Internet for the same amount of time without recharging?

I am not an Apple or MacOs fan but I recently bought a MacBook Pro just for battery life. So I can watch movies and browse the Internet on the sofa without having to plug the laptop.

If you benchmark the AMD 7xxx series aganst the Apple chips while considering their power budget, AMD is on par or on top.

Battery life is strongly tied to software though, and here Apple has the upper hand. However, if you use Chrome vs Safari, you will notice degraded battery life on Mac OS too.

I use Edge and I can browse the Internet for about 20 hours.

> However, if you use Chrome vs Safari, you will notice degraded battery life on Mac OS too.

Still way better than non-M chips.

I've done the runaround and used Safari, Orion, Chrome, Arc, and back to Firefox on my M1 MBP. Some are better than others, but the machine overall is still superior.

The ASUS ZenBook 14 has been tested to have about 18 hours of battery life for basic web browsing [1]. In the same test the 2020 MacBook Air 13" only scored about 14 hours. The Zenbook also easily beat the M1 in multicore performance, though single core is still where Apple shines.

Unfortunately, this test didn't include a more recent Macbook, so I can't tell how it stacks up against the M3. Still, for about €500 less than the M2 Air, I know what I would pick. The 1600p screen and the soldered-on memory is a bit of a downside, but the thing sells for less than half the price of a new M2 Macbook Air when you actually configure it with 16GB of RAM and 512GB of storage; whatever improvements the M2 chip made don't seem all that interesting for that price difference.

Based on their historical data [2] for another guide, the Zenbook also seems to beat the M2 in terms of multi core performance and browsing battery life when recorded in another test, but I can't tell if those numbers are representative.

[1]: https://tweakers.net/reviews/11472/4/laptop-best-buy-guide-o...

[2]: https://tweakers.net/reviews/11610/4/laptop-best-buy-guide-d...

Before buying a MacBook Pro 13", I've seen lots of reviews on notebookcheck. I bought my MacBook used and it really gives me 20 hours of battery life like in their test.

https://www.notebookcheck.net/Apple-MacBook-Pro-13-2020-Lapt...

If that Zenbook lasts for 18 hours, it is a good contender. I don't care for having the fastest CPU, just a decent one and plenty of battery life.

Every website tests different settings. There's a certain standard for brightness, the websites browsed, the software used, and so on. You can't compare battery tests unless all the test parameters are the same.

For instance, Tweakers uses a screen brightness of 180 cd/m² while testing, whereas NotebookCheck uses 150 cd/m². That alone should account for a sizeable difference in test results. I find it hard to tell what 30 cd/m² means in practice.

> The ASUS ZenBook 14 has been tested to have about 18 hours of battery life for basic web browsing. In the same test the 2020 MacBook Air 13" only scored about 14 hours.

Because the ZenBook has a 75 Wh battery, in comparison to the MBA's 52.6 Wh battery. That's one of the reasons it's heavier, too.

Reply on news.ycombinator.com