Apple Leads Global PC Growth With 21% Shipment Increase

The funny thing is if you want a PC that has AI capabilities then you build a PC with Nvidia GPU. Not buying a low powered ARM PC/laptop.

AI will excite some but this is just a temporary boost (not AI itself). Most people still prefer their phone/tablet. It's good that laptops are a focus here but it's not going to last.

A lot of people underestimate just how demanding true AI is. You want to do AI stuff? You need 32-64GB RAM and as much VRAM as you can afford and I'm talking 24, 48, 96GB .... maybe even 100GB+ lol. You need a good x86 processor too. Last time I checked, Apple still starts at 8GB unified memory lol. You can do basically nothing with 8GB across cpu/gpu.

If Apple wants to continue their success they need to start increasing memory specs. 16GB would be a good start. I'd say 24-32GB ideally and even that seems like little. But of course Apple will not give you true AI.

What you're talking about is hardware necessary to build a model-- much less hardware is required to run/use a model. Very few people are building models, much less routinely.

That said, Apple's unified memory is largely a strength for building models. There is not a lot of computers out there with 24+ gb of VRAM on their GPU. Easily <1% of computers sold. EVERY mac that has that much ram has that much VRAM. Yes, the base Macs start at 8gb, but nobody training a model is going to (or should) buy one of those. If you professionally build models day in, day out, yes, you should specifically build a computer for that purpose (optimize for CUDA cores). For "normal" people, that rarely do so-- Apple's setup gets you a long way there, for a much lower price, and does so for a wide swath of their product line.

The average price of a new PC is $632, from the latest data I can find. That is the market Apple is competing with, with the 8gb Macs. Which computer do you think would do AI better?

No benefit to an x86 processor. If anything, the opposite-- "Apple intelligence" is running on every Apple computer since the move off x86 to ARM in 2020. Similarly, Microsofts "Copilot+" requirements, thus far, are only met by 2024+ ARM processors. So, if anything, thus far both Microsoft's and Apple's AI models can only run on ARM based PCs. That'll likely change on the PC side, but it certainly doesn't imply that x86 is the stronger starting point for running AI models :p

The funny thing is if you want a PC that has AI capabilities then you build a PC with Nvidia GPU. Not buying a low powered ARM PC/laptop.

An RTX 4090 has 24GB ram which won't run models bigger than 20GB.

An M3 Max 48GB will run 35GB size models fine and consume about 8 times less energy than a PC with a 4090. The highest end Macs can offer well over 128GB to video memory.

If you want that much VRAM with a PC you'll need to string a bunch of very expensive GPUs together and spend a mother load on electricity.

You need a good x86 processor too.

Nonsense.

Last time I checked, Apple still starts at 8GB unified memory lol. You can do basically nothing with 8GB across cpu/gpu.

That's a starter office PC/Mac. It's not for any advanced AI tasks.

If you're going to act like an expert then get facts right.

The funny thing is if you want a PC that has AI capabilities then you build a PC with Nvidia GPU. Not buying a low powered ARM PC/laptop.

AI will excite some but this is just a temporary boost (not AI itself). Most people still prefer their phone/tablet. It's good that laptops are a focus here but it's not going to last.

A lot of people underestimate just how demanding true AI is. You want to do AI stuff? You need 32-64GB RAM and as much VRAM as you can afford and I'm talking 24, 48, 96GB .... maybe even 100GB+ lol. You need a good x86 processor too. Last time I checked, Apple still starts at 8GB unified memory lol. You can do basically nothing with 8GB across cpu/gpu.

I think the problem has been that AI has become a term that’s so hard to pin down because it encompasses a huge variety and various levels of technology. It’s a catch all term. Marketing. Hardcore AI is indeed very power and memory intensive.

The funny thing is if you want a PC that has AI capabilities then you build a PC with Nvidia GPU. Not buying a low powered ARM PC/laptop.

AI will excite some but this is just a temporary boost (not AI itself). Most people still prefer their phone/tablet. It's good that laptops are a focus here but it's not going to last.

A lot of people underestimate just how demanding true AI is. You want to do AI stuff? You need 32-64GB RAM and as much VRAM as you can afford and I'm talking 24, 48, 96GB .... maybe even 100GB+ lol. You need a good x86 processor too. Last time I checked, Apple still starts at 8GB unified memory lol. You can do basically nothing with 8GB across cpu/gpu.

If Apple wants to continue their success they need to start increasing memory specs. 16GB would be a good start. I'd say 24-32GB ideally and even that seems like little. But of course Apple will not give you true AI.

Market excitement around AI…. A lot of users are more concerned than excited

The only reason there is market excitement is that it’s the next big thing to push through a load of new PC sales and throw another load of e waste into landfill 😂

Growth above all else… until it’s too late.

The funny thing is if you want a PC that has AI capabilities then you build a PC with Nvidia GPU. Not buying a low powered ARM PC/laptop.

AI will excite some but this is just a temporary boost (not AI itself). Most people still prefer their phone/tablet. It's good that laptops are a focus here but it's not going to last.

A lot of people underestimate just how demanding true AI is. You want to do AI stuff? You need 32-64GB RAM and as much VRAM as you can afford and I'm talking 24, 48, 96GB .... maybe even 100GB+ lol. You need a good x86 processor too. Last time I checked, Apple still starts at 8GB unified memory lol. You can do basically nothing with 8GB across cpu/gpu.

If Apple wants to continue their success they need to start increasing memory specs. 16GB would be a good start. I'd say 24-32GB ideally and even that seems like little. But of course Apple will not give you true AI.

I don't know what your idea is of "real" AI, but to me this sounds a bit like a "no true scotsmen" argument. I have a masters degree in AI (don't work in the field anymore) but back in the day as a student (and this was last decade, not a hundred years ago) we had to build and run models on our laptops. Both rudimentary language models but also more complex things (like models to make robot wheelchairs navigate unknown environments including traffic based on instructions from noisy EEG readings). All of that was "real" AI, yet almost none of it required more than 16GB of RAM. Did some things take a long time to build? Yes. Were they unusable once built? Nope. Not even on an 8GB RAM laptop.

It's funny how you talk about "REAL" AI, then give the example of Stable Diffusion, something that will run on any modern MacBook, even the ones with a humble M1 soc. To give another real-world example, the autopilot computer in a Tesla got bumped to its 4th iteration last year. The HW4 Autopilot has double the RAM of the previous version: it went from 8 to 16GB. And that's an AI system that is literally in charge of people's lives.

I think A) you need to reconsider what "real" (as opposed to what?) AI is, and B) consider that no, for most REAL AI, you don't need the highest specs, unless you want to do some very data-intensive things within a relatively short time.