I feel like i'm alone to dislike M1/what is it good for (Merged Mega Thread)

[ric22]

The Max is faster than the Pro and the Ultra is faster than the Max. If you need 64GB you get the Max and if need 128GB you get the Ultra. Whether it's using all the bandwidth is irrelevant, as is the cost if you are earning good money with your computers. The faster M1 helps you earn more money than than the slower M1 and you deduct the purchase expenses on your taxes. It's a win-win for everyone except you.

You replied to me about a sarcastic comment, mocking me when I asked what you needed 400GB/s bandwidth for.... now you're the one saying it's irrelevant??!?!?!

You've demonstrated you don't know what you're talking about, and I don't like being trolled, so I'll block you now.

 

[darngooddesign]

You replied to me about a sarcastic comment, mocking me when I asked what you needed 400GB/s bandwidth for.... now you're the one saying it's irrelevant??!?!?!

You've demonstrated you don't know what you're talking about, and I don't like being trolled, so I'll block you now.

Ironic since your tone has been pretty sarcastic. Still, the fact remains that your theoretical concerns just aren't relevant to real world use.

 

[jdb8167]

Maximising single core performance is usually better for coding / programming. As much as all those cores are nice, compilers are nearly all single threaded or have minimal improvement running parallel compilation.

How to tell us you’ve never worked on a complex software project without saying, “I’ve never worked on a complex software project.“

 

[philstubbington]

Maybe its just for those who end up buying the best thats available at the time thinking that it will 'future proof' them for many years to come......but then in a few years they end up upgrading to a newer system anyways........[myself included, although trying to break that habit!] lol

I bought pretty much the top end M1 AirBook (maximum cores, memory and 1Tb storage) but I tend to hang onto them for quite a long time. As I did with my last two Macs.

 

[Tagbert]

This discussion seems to have devolved into some pointless hairsplitting and personal attacks. Adios.

 

[Krevnik]

How to tell us you’ve never worked on a complex software project without saying, “I’ve never worked on a complex software project.“

I seriously facepalmed when I read that. There are a lot of factors in play, for sure, and you should always measure your projects and find out what the bottlenecks are. But even my small Swift service that takes seconds to build scales nearly linearly with cores until linking starts dominating the build.

 

[Mr. Dee]

I think we are living in a golden age of abundance of computing power we just don’t know what to do with it. Even when Moores Law eventually maxes out, we will still not be able to max out what we have today. Considering a lot of folks are buying some of these expensive configs they plan on keeping for the next decade.

 

[Danfango]

How to tell us you’ve never worked on a complex software project without saying, “I’ve never worked on a complex software project.“

Define your metric of “complex software project” and I’ll see if your numbers match mine.

We have 500 engineers as a starter…

 

[jdb8167]

Define your metric of “complex software project” and I’ll see if your numbers match mine.

We have 500 engineers as a starter…

I take it you never do full builds locally? Everything done via a CI/CD environment?

It seems like you are the definition of, “I don’t do that so it doesn’t affect anyone.” To say that a full build with multiple compilation units doesn’t benefit from more cores is just bizarre.

 

[Danfango]

I think we are living in a golden age of abundance of computing power we just don’t know what to do with it. Even when Moores Law eventually maxes out, we will still not be able to max out what we have today. Considering a lot of folks are buying some of these expensive configs they plan on keeping for the next decade.

Of course we know what to do with it.

Firstly we design crappy unsafe languages and use them for 50 years. Then we build crappy browsers on top of them. Then we build crappy abstractions on top of those browsers like web assembly. Then we build more crappy languages on top of that and more crappy abstractions.

Then eventually we wrap it in electron which allows us to run several piles of dung like this on top of a poor machine that is crying it’s eyes out.

 

[AxiomaticRubric]

Coding - only if you target Apple/Web, otherwise just issues upon issues

The ignorance here is appalling.

Macs are the best hardware you can get for Microsoft development. .NET 6.0 for ARM is blazing fast on the M1. Visual Studio Code is the best IDE for web development, and the next version of Visual Studio for Mac is nearly out of the beta stage.

Docker offers endless possibilities for developing software that can target any Linux distribution (Java, .NET, Swift, Node.js, Python…. anything). You can develop SQL Server databases running in Docker containers on a Mac.

The latest versions of Node.js target native M1 ARM and work well on macOS.

Add to this all the great design tools like the Affinity suite, Sketch, Adobe products, etc. matched with the excellent user experience of macOS. There is simply nothing else like this for software development.

To address the original post: Yes there is a definite programming use-case for the M1 Pro, Max, and Ultra: Microservices architecture in Kubernetes requires a lot of memory and multi-core parallelization to run many Docker container VMs at the same time. The more powerful the Mac is, the closer the developer can get to simulating an actual cloud deployment of an intricate multi-container application.

 

[Danfango]

I take it you never do full builds locally? Everything done via a CI/CD environment?

It seems like you are the definition of, “I don’t do that so it doesn’t affect anyone.” To say that a full build with multiple compilation units doesn’t benefit from more cores is just bizarre.

Yes we do full builds locally (with exceptions see later) but the compilers do not noticeably benefit from running in parallel for the technologies we use (Go, C#). Even our old C++ stuff has limited scalability improvements on this using clang/llvm. This is because most of the compilation units are stacked in a dependency tree which cannot be parallelised at all because one thing has to exist before the other.

What really improves things is fracturing the design out into lots of smaller silos and using communication and composition to build a product. You reduce the compiler effort per change to only the dependency you are changing. Then you don’t need to contribute to the heat death of the universe every time you run the compiler. Our largest unit, which is a fairly complex microservice takes ~6s to compile on an M1 Pro.

Notably we don’t compile much. We like ~150 base containers into a VM running docker and drop out the service we’re working on for local development. No one is going to want to compile 149 containers and run them before your stuff even starts.

 

[jdb8167]

Yes we do full builds locally (with exceptions see later) but the compilers do not noticeably benefit from running in parallel for the technologies we use (Go, C#). Even our old C++ stuff has limited scalability improvements on this using clang/llvm. This is because most of the compilation units are stacked in a dependency tree which cannot be parallelised at all because one thing has to exist before the other.

And how do you reconcile that statement with this?

compilers are nearly all single threaded or have minimal improvement running parallel compilation.

Did you mean to say, “the compilers we use in the way we use them have minimal improvement?” What you wrote seems to be stated as a universal truth. That statement is flat out wrong for many situations. Much of what I do could easily benefit from 20 CPU cores.

 

[Ethosik]

Pity you failed to read the other posts and discovered the Mac can't even utilise all that lovely bandwidth, eh? Spend those 10s of thousands you earn on top spec Macs in forlorn hope of saturating the bandwidth...

And someone just using Word isn't using all of DDR5 bandwidth either. So we should just leave progress to the Elite of the Elite I guess?

I deal with a client that still uses a PowerPC Mac for photography processing (using Adobe CS2). What do you think the memory bandwidth is for that?

 

[Danfango]

And how do you reconcile that statement with this?


Did you mean to say, “the compilers we use in the way we use them have minimal improvement?” What you wrote seems to be stated as a universal truth. That statement is flat out wrong for many situations. Much of what I do could easily benefit from 20 CPU cores.

Let me summarise:

Our C++ traditional code, mostly integration and financial data processing, algorithmic, compute, totalling 20MLOC or so, doesn’t really gain more than 5% adding more cores at compile time because the expensive path is vertical dependencies. I’ve seen this across big data, compute, scientific and even desktop apps over the years.

What does have a ROI is isolating the dependencies into independent binaries which are faster to compile because they have a shorter dependency tree.

Honestly I have to optimise 500 people for frictionless development as they are the most expensive part of the organisation. Throwing more cores at them is a bad investment when there are architectural changes that can make a net gain.

We have to stop throwing compute at problems and solve them intelligently or we will just cruise down the happy path until physics kicks us in the balls and says no thanks.

When you have to scale out cores because you can’t make them smaller and faster, relying on parallel compilation isn’t going to scale with dependency graphs. Crush the graph first.

 

[Unregistered 4U]

Intel more likely said something like "This is kind of what we are hoping to do, but we promise absolutely nothing."

No, contracts of this type are rarely “promise absolutely nothing”. NONE of the vendors would contract in advance to buy chips if that were the case. Intel provides a roadmap indicating a feature set they will hit and vendors contract to buy a certain quantity. If Intel doesn’t provide a suitable chip, then there are clauses in those contracts that go into effect.

Thing is, those goals weren’t even some far fetched Intel pipe dreams. They were the same fairly modest goals that Intel had little problem hitting year over year… until Skylake. As a result, Intel began sliding back those delivery targets by 1 year, 2, 3… etc. Whatever long term contract Apple had with them, Intel’s failure to deliver gave them an out and now Apple’s on Apple Silicon. AND, having FAR less difficulty than Intel of hitting their TDP targets!

They almost certainly had many internal prototypes of thicker laptops better suited for Intel CPUs and AMD GPUs, but they chose not to sell them.

Oh, I’m sure we’d have seen them if they exist. What’s more likely is that Intel kept telling Apple over and over again for YEARS that they’re close to hitting the TDP’s Apple needed for their MacBook Pro’s, they just needed to wait a little longer.

And, Intel’s STILL missing their roadmap TDP goals even WITHOUT Apple as a customer.

 

[jdb8167]

Let me summarise:

Our C++ traditional code, mostly integration and financial data processing, algorithmic, compute, totalling 20MLOC or so, doesn’t really gain more than 5% adding more cores at compile time because the expensive path is vertical dependencies. I’ve seen this across big data, compute, scientific and even desktop apps over the years.

What does have a ROI is isolating the dependencies into independent binaries which are faster to compile because they have a shorter dependency tree.

Honestly I have to optimise 500 people for frictionless development as they are the most expensive part of the organisation. Throwing more cores at them is a bad investment when there are architectural changes that can make a net gain.

We have to stop throwing compute at problems and solve them intelligently or we will just cruise down the happy path until physics kicks us in the balls and says no thanks.

When you have to scale out cores because you can’t make them smaller and faster, relying on parallel compilation isn’t going to scale with dependency graphs. Crush the graph first.

Sure. But you ignored my question. Can you not see a situation where more cores does scale and has a nice ROI? Or is your dev env the only one that matters?

 

[ondioline]

Even our old C++ stuff has limited scalability improvements on this using clang/llvm.

This isn't really true these days, newer linkers like mold hugely benefit from parallelism. That is almost always the most time consuming part of a C++ or Rust build.

GitHub - rui314/mold: Mold: A Modern Linker 🦠

Mold: A Modern Linker 🦠. Contribute to rui314/mold development by creating an account on GitHub.

github.com

 

[Sopel]

hardware encoding useless? so funny

With nothing to stream, yes. What would you use it for?

 

[macacam]

3. 3D Modeling & Animation - You probably want to get a Quadro RTX card if you’re serious about this. M1 Max has no ray tracing capabilities, and its performance & cost is not as good as Nvidia cards. Software support on M1 is available in general but can’t compare to Windows PC. Windows with Nvidia is just better at this than M1.

What? We got a neophyte talkin mess on the internet. A windows pc is indeed better, but for absolutely none of the reasons you've mentioned!

 

[Danfango]

This isn't really true these days, newer linkers like mold hugely benefit from parallelism. That is almost always the most time consuming part of a C++ or Rust build.

GitHub - rui314/mold: Mold: A Modern Linker 🦠

Mold: A Modern Linker 🦠. Contribute to rui314/mold development by creating an account on GitHub.

github.com

Correct.

But that is all about squeezing some more juice out of large monolithic dependency chains.

I’m saying that’s not how we should build software because it’s solving the problem at the wrong end. It’s an architectural problem which is damage amplified by the compiler and linker.

 

[macsound1]

I'm not seeing the purpose in this post? What professional work do professionals do? That is your answer the work that professionals do. The fact that no one games (is that work?) on Macs who cares? Otherwise, virtually anything that people can do on a computer. sure some folks who are windows fanboys willl tell you certain software won't run - and they are right, but there are always other ways to do things. If your education and training are restricted to certain windows only things, then sure, go windows, go deep, run along, nothing to see here folks (Southpark credit).

Or maybe the fact that apple sales are increasing while windows is declining means no one is buying computers for any real work, because real work is done on Windows. Did I get that right, or am I rambling?

I play 1 game on my macs. Cities Skylines. I've been playing it on my 2013 15" MBP since the game came out in 2015 (I think).
As the game got updated, the mac version didn't work as well, so I installed bootcamp and played the game in windows on the same mac.
Then I got a base M1 Mac Mini, even though it had 8GB of ram instead of my laptop's 16gb, it ran so much better.
My gf got a 14" Max and 32Gb and it runs amazingly well.

 

[JimmyjamesEU]

With nothing to stream, yes. What would you use it for?

Is this a serious question?

 

[Sopel]

Is this a serious question?

Yes, it's a serious question. The compression is terrible with hardware encoders so they are useless for non-realtime encoding.

 

[Diablo360]

Bought my M1 Max for daytrading. Needed a fast powerful laptop for running apps and charts, and able to run 4 x 49” widescreen monitors. This replaced a 12 core Mac Pro 5,1. Best computer I’ve ever owned hands down.