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

Dissecting the Apple M1 GPU, Part II

rosenzweig.io

Post reply on HN
Print view View original

71–80 of 195 posts

Re: Dissecting the Apple M1 GPU, Part II

#71

Post by pedrocr »

threentaway wrote: post #18
gnarbarian wrote: post #10

Earlier quoted context omitted.

Currently using a Renoir laptop. It's smoking fast but I had to install a bleeding edge kernel to get the display driver to work at all. That should be fixed in the next ubuntu release though.

Weird, I had no issues with the built in display on Ubuntu 20.04, but I had to update the kernel to 5.8 to get display out over USB-C to work. Now that Ubuntu 20.10 is out and uses 5.8, I'm just using that so I don't have to mess with custom, unsigned kernels.

I just installed 20.04.1 on a 4750U Lenovo T14s and everything just works as far as I can tell.

Re: Dissecting the Apple M1 GPU, Part II

#72

Post by The_Colonel »

gonehome wrote: post #61
monkmartinez wrote: post #56

Earlier quoted context omitted.

Are you saying you don't see much promise for AMD, Intel and Nvidia in the GPU space or with computers in general? I had a hard time following your logic. Apple may own their stack, but there are a TON of use cases where that stack doesn't even form a blip on the radar of the people who purchase computer gear.

My prediction is x86 is dead. External GPUs will remain and I think Nvidia has an advantage in that niche currently. The reason stack ownership matters is because it allows tight integration which leads to better chip design (and better performance/efficiency). Windows has run on ARM for a while for example, but it sucks. The reason it sucks is complicated but largely has to do with bad incentives and coordination pr…

> My prediction is x86 is dead.

I don't see that. At least in corporate environment with bazillion legacy apps, x86 will be the king for the foreseeable future.

And frankly I don't really see the pull of ARM/M1 anyway. I mean, I can get a laptop with extremely competitive Ryzen for way cheaper than MacBook with M1. The only big advantage I see is the battery, but that's not very relevant for many use cases - most people are buying laptops don't actually spend that much time on the go needing battery power. It's also questionable how transferable this is to the rest of the market without Apple's tight vertical integration.

> I'd argue Apple's M1 machines are already at this level and they're version 0

Where is this myth coming from? Apple's chips are now on version 15 or so.

Re: Dissecting the Apple M1 GPU, Part II

#73

Post by monkmartinez »

gonehome wrote: post #61
monkmartinez wrote: post #56

Earlier quoted context omitted.

Are you saying you don't see much promise for AMD, Intel and Nvidia in the GPU space or with computers in general? I had a hard time following your logic. Apple may own their stack, but there are a TON of use cases where that stack doesn't even form a blip on the radar of the people who purchase computer gear.

My prediction is x86 is dead. External GPUs will remain and I think Nvidia has an advantage in that niche currently. The reason stack ownership matters is because it allows tight integration which leads to better chip design (and better performance/efficiency). Windows has run on ARM for a while for example, but it sucks. The reason it sucks is complicated but largely has to do with bad incentives and coordination pr…

Ok, I still have questions.

To start... How would a city with tens of thousands of computers transition to ARM in the near future?

The apps that run 911 Dispatch systems and run critical infrastructure all over the world are all on x86 hardware. Millions if not Billions of dollars in investment, training, and configuration. These are bespoke systems. The military industrial complex basically custom chips and x86. The federal government runs on x86. you think they are just going to say, "Whelp, looks like Apple won, lets quadruple the cost to integrate Apple silicon for our water system and missile systems! They own the stack!"

Professional grade engineering apps and manufacturing apps are just going to suddenly rewrite for apple hardware, because M2 or M3 is sooooo fast? Price matters!!!! Choice Matters!!!

This is solely about consumer choice right now. The cost is prohibitive for most consumers as well, as evidence by the low market penetration of Apple computers to this day.

Re: Dissecting the Apple M1 GPU, Part II

#74

Post by uncledave »

spideymans wrote: post #69
gonehome wrote: post #43

Earlier quoted context omitted.

You might have to bite the bullet and get a mac - I don't see much promise from others in this space. Intel's CEO change may save them, but they're definitely facing an existential threat they've failed to adapt to for years. Amazon will move to ARM on servers and probably some decent design there, but that won't really reach the consumer hardware market (probably - though I suppose they could do something interestin…

I just got my M1 Air. This thing is unbelievably fluid and responsive. It doesn't matter what I do in the background. I can simultaneously run VMs, multiple emulators, compile code, and the UI is always a fluid 60 fps. Apps always open instantly. Webpages always render in a literal blink of an eye. This thing feels like magic. Nothing I do can make this computer skip a beat. Dropped frames are a thing of the past. Th…

So much this. M1 mini here. I am absolutely chuffed with it. It’s insanely good.

I’m going to be the first person in the queue to grab their iMac offering.

Re: Dissecting the Apple M1 GPU, Part II

#75

Post by war1025 »

chucky_z wrote: post #58
gjsman-1000 wrote: post #55

Everyone, remember that in 2019 Alyssa was in _High School_. Where do people get this knowledge? It's astounding. She was literally reverse-engineering Mali for Panfrost starting in her Sophomore year.

Where did you get your knowledge? With places like HackerNews along with folks like yourself and myself writing posts about this kind of stuff it becomes trivial to walk down a rabbit hole for anyone of any age. Think back to how much you accomplished when you were younger because you were walking in the footsteps of others. :)

A magical thing about high school / college that people fail to appreciate is how few actual responsibilities you have at that age.

It's why kids are able to devote so many hours to video games and other time waster stuff.

Only difference is this person managed due to luck / skill / timing to end up on a more productive path.

Time commitment is the big impediment for most open-source projects. Late adolescence / young adulthood is one of the few periods of life where you both are competent enough to do useful things and have enough free time to actually commit to that work.

Re: Dissecting the Apple M1 GPU, Part II

#76

Post by floatboth »

unix_fan wrote: post #45
stefan_ wrote: post #31

Earlier quoted context omitted.

They are also on a different process shrug

Process node improvements don’ty bring that much performance.

Even evolution within the same process node can bring noticeable performance improvements. Launch day AMD Zen 2 (TSMC 7FF) chips could barely clock to 4.2GHz, ones manufactured months later can often do 4.4.

Re: Dissecting the Apple M1 GPU, Part II

#77

Post by floatboth »

unix_fan wrote: post #47
yepthatsreality wrote: post #11

Earlier quoted context omitted.

And Apple will court (allow) developers until competition shows up. Then they will shut it down and obfuscate. The fact that Apple refuses to provide FOSS drivers themselves indicates this.

Apple has never provided drivers for other operating systems, including windows. The drivers used for BootCamp aren’t even developed by them.

Well, they clearly at least commissioned them from someone else, since they are the ones distributing them.

Re: Dissecting the Apple M1 GPU, Part II

#78

Post by cwhiz »

vbezhenar wrote: post #33
cambalache wrote: post #6

OK I love this, but I am pretty sure my chances of getting a mac in the future are zero. From which side should I expect the PC response? Intel? AMD? Microsoft? ARM? Samsung? I think Apple willingly or not has just obliterated the whole consumer pc market.

What exactly do you expect? x86 is quite competitive. M1 might be slightly better, but it's not like it's miles ahead.

Fiddling around on an M1 MBA it felt faster than my 2020 16" MBP. It's half the weight, seems to get double the battery life, and costs less than 1/3.

I just can't even imagine what the gap is going to look like when Apple really refines this down.

Re: Dissecting the Apple M1 GPU, Part II

#79

Post by jrockway »

monkmartinez wrote: post #34
zaptrem wrote: post #15

Earlier quoted context omitted.

M1 cores outperform Comet Lake cores and are basically tied with AMD Vermeer despite using a fraction of the power.

Who cares if you have the darn thing plugged in anyway? Does the M1 outperform the Threadripper 3990x?

I think to understand the M1, you have to be a Mac laptop user. For years, Mac laptop performance lagged years behind high-end desktop performance -- they have been stuck on 14nm+ process chips with a mobile power budget, while desktop users have had 7nm chips that can draw 500W with no trouble. As a result, what M1 users tell you is fast is what PC desktop users have had for ages. A Threadripper and 3090 will blow the M1 out of the water in raw performance (but use a kilowatt while doing it, which a laptop obviously can't do).

At my last job, they issued us 2012-era Macbooks. I eventually got so frustrated with the performance that I went out and bought everyone on my team an 8th generation NUC. It was night and day. I couldn't believe how much faster everything was. The M1 is a similar revelation for people that have stayed inside the Mac ecosystem all these years.

Re: Dissecting the Apple M1 GPU, Part II

#80

Post by floatboth »

astrange wrote: post #52
monkmartinez wrote: post #34

Earlier quoted context omitted.

Who cares if you have the darn thing plugged in anyway? Does the M1 outperform the Threadripper 3990x?

Personally I like the noise level in my room being 32 dB and don't like PCs having to run the fan at full speed to show the desktop picture.

Any desktop PC, even with the beefiest Threadripper, won't run the fan at full speed when you're just browsing the web or watching videos.

Heck, if you're not squeezing the absolute maximum performance from a chip by overclocking (I am :P) you can run a 5950X on a regular tower cooler with the fan curve never coming close to 100% and it will still be incredibly fast.

Post reply on HN