Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Apple M5 chip

apple.com

Post reply on HN
Print view View original

401–410 of 1001 posts

Re: Apple M5 chip

#401

Post by jeffbee »

timeon wrote: post #317
jeffbee wrote: post #254

Earlier quoted context omitted.

I'm interested in reading about your low-carbon lifestyle that is so efficient you got to the point of giving up machine inference.

Depends where you are. People in some countries have lot of catching up: https://ourworldindata.org/grapher/co-emissions-per-capita Maybe they are in USA - every little think counts there.

No, in the USA it is the opposite. The little things do not and cannot add up to anything. The only things that make a difference are motor fuels and hamburgers.

Re: Apple M5 chip

#402

Post by tylerhou »

paxys wrote: post #392

M5 is 4-6x more powerful than M4, which was 5x more powerful than M3, which was 4x more powerful than M2, which was 4x more powerful than M1, which itself was 6x faster than an equivalent Intel processor. Great! Looking at my Macbook though, I can say with utmost certainty that it isn't 4000x faster than the Intel one I had 5 years ago. So, where is the disconnect here? Why is actual user experience not able to keep…

> M5 is 4-6x more powerful than M4

In GPU performance (probably measured on a specific set of tasks).

Re: Apple M5 chip

#403

Post by condiment »

paxys wrote: post #392

M5 is 4-6x more powerful than M4, which was 5x more powerful than M3, which was 4x more powerful than M2, which was 4x more powerful than M1, which itself was 6x faster than an equivalent Intel processor. Great! Looking at my Macbook though, I can say with utmost certainty that it isn't 4000x faster than the Intel one I had 5 years ago. So, where is the disconnect here? Why is actual user experience not able to keep…

It's GPU performance.

Spin up ollama and run some inference on your 5-year-old intel macbook. You won't see 4000x performance improvement (because performance is bottlenecked outside of the GPU), but you might be in the right order of magnitude.

Re: Apple M5 chip

#404

Post by liuliu »

zozbot234 wrote: post #265
liuliu wrote: post #253

Earlier quoted context omitted.

Faster compute helps, for things like vision language model that requires bigger context to be filled. My understanding is that ANE is still optimized for convolution load, and compute efficiency while the new neural accelerators optimized for flexibility and performance.

The old ANE enabled arbitrary statically scheduled multiply-add, of INT8 or FP16. That's good for convolution but not specifically geared for it.

I am not an expert on ANE, but I think it is related to the size of register files and how that is smaller than what we need for GEMM on modern transformers (especially these fat ones with MoE).

Re: Apple M5 chip

#405

Post by tester756 »

paxys wrote: post #392

M5 is 4-6x more powerful than M4, which was 5x more powerful than M3, which was 4x more powerful than M2, which was 4x more powerful than M1, which itself was 6x faster than an equivalent Intel processor. Great! Looking at my Macbook though, I can say with utmost certainty that it isn't 4000x faster than the Intel one I had 5 years ago. So, where is the disconnect here? Why is actual user experience not able to keep…

Because this is bullshit, lies, marketing

Re: Apple M5 chip

#406

Post by freehorse »

paxys wrote: post #392

M5 is 4-6x more powerful than M4, which was 5x more powerful than M3, which was 4x more powerful than M2, which was 4x more powerful than M1, which itself was 6x faster than an equivalent Intel processor. Great! Looking at my Macbook though, I can say with utmost certainty that it isn't 4000x faster than the Intel one I had 5 years ago. So, where is the disconnect here? Why is actual user experience not able to keep…

They are not 4x more powerful than the previous generation at everything, or even at the same thing every time, so it does not stuck up. Here 4x refers sth wrt LLMs running on the GPU.

I use both an M1 max and an M3 max, and frankly I do not notice much difference if you control for the core count in most stuff. And for running LLMs they are almost the same performance. I think from M1-M3 there was no much performance increase in general.

Re: Apple M5 chip

#407

Post by qnpnp »

bamboozled wrote: post #27
littlecranky67 wrote: post #11

And here I am, selling my Macbook M4 Pro to buy a Macbook Air and a dedicated gaming machine. I've tried gaming on the Macbook with Heroic, GPTK, Whiskey, RPCS3 emu and some native. When a game runs, the performance is stunning for a Laptop - but there is always glitches, bugs and annoyances that take out the joy. Needles to mention lack of support from any sort of online multiplayer, due to the lack of anticheat sup…

Sometimes I just feel like buying the latest and greatest game, I have an m4 too, the choices are usually quite abysmal. I agree.

My solution is cloud gaming in that case, such as GeforceNow (for compatible games), or Shadow (for a whole PC to do as you please).

Re: Apple M5 chip

#408

Post by random3 »

paxys wrote: post #392

M5 is 4-6x more powerful than M4, which was 5x more powerful than M3, which was 4x more powerful than M2, which was 4x more powerful than M1, which itself was 6x faster than an equivalent Intel processor. Great! Looking at my Macbook though, I can say with utmost certainty that it isn't 4000x faster than the Intel one I had 5 years ago. So, where is the disconnect here? Why is actual user experience not able to keep…

The disconnect is that you're reading sideways.

First line on their website:

> M5 delivers over 4x the peak GPU compute performance for AI compared to M4

It's the GPU not the CPU (which you compare with your old Intel) and it's an AI workload, not your regular workload (which again is what you compare)

Re: Apple M5 chip

#409

Post by hannesfur »

fooblaster wrote: post #329
hannesfur wrote: post #258

Earlier quoted context omitted.

Oh, I overlooked that! You are right. Surprising… since Apple has shown that it’s possible through CoreML ( https://github.com/apple/ml-ane-transformers ) I would hope that the Foundation Models ( https://developer.apple.com/documentation/foundationmodels ) use the neural engine.

The neural engine not having a native programming model makes it effectively a dead end for external model development. It seems like a legacy unit that was designed for cnns with limited receptive fields, and just isn't programmable enough to be useful for the total set of models and their operators available today.

That's sadly true, over in x86 land things don't look much better in my opinion. The corresponding accelerators on modern Intel and AMD CPUs (the "Copilot PCs") are very difficult to program as well. I would love to read a blog post on someone trying though!

Re: Apple M5 chip

#410

Post by amluto »

hangonhn wrote: post #345
sosodev wrote: post #331

Earlier quoted context omitted.

It's easier to develop a game for a mac in some ways but you reach a tiny fraction of gamers that way.

I wonder how that might look once you factor in Apple TV devices. They're pretty weak devices now but future ones can come with M-class CPUs. That's a huge source of potential revenue for Apple.

The current Apple TV is, in many respects, unbelievably bad, and it has nothing to do with the CPU.

Open up the YouTube app and try to navigate the UI. It’s okay but not really up to the Apple standard. Now try to enter text in the search bar. A nearby iPhone will helpfully offer to let you use it like a keyboard. You get a text field, and you can type, and keystrokes are slowly and not entirely reliably propagated to the TV, but text does not stay in sync. And after a few seconds, in the middle of typing, the TV will decide you’re done typing and move focus to a search result, and the phone won’t notice, and it gets completely desynchronized.

Post reply on HN