You realize the framework laptop’s cpus are limited to 4 cores, right?

No $2000 laptop in 2020 should have only quad core cpus.

The M1 16gb will run circles around the i7-1185g7 (highest spec chip framework sells) compiling llvm with 64gb of ram.

Only using a shitload of idle VMs will have any apparent benefit over M1 MacBook pros.

The cpu is just too limiting for the 64gb of ram argument.

>The M1 16gb will run circles around the i7-1185g7

I have an M1 MBP and an 1185g7 based Linux laptop. Single core benchmarks, the i7 comes out slightly ahead. Multi core benchmarks, the M1 comes out slightly ahead. There are no circles being run around in performance - only in battery efficiency. For day to day developer usage, testing things literally side by side, I find absolutely no discernible difference in speed.

The M1 does win for battery life, but the battery still lasts for a full work day in my i7 laptop. Honestly at home I always reach for the Linux laptop, because I don't really need the battery life. But for going out and about away from power, the M1 MBP has advantages. Sadly the software (personally) lets it down, I'd probably use the MBP more if it has better Linux support in the future.

Prolonged compiling tasks will thermally throttle the 1185g7 chip long before the M1.

It will run circles around the 1185g7 compiling large codebases.

Passmarks across hundreds of benches show a 25% single and 40% multithreaded performance boost over the 1185g7.

I’ve got an i9 8/16 laptop chip and the 13” M1 is trading blows with my laptop despite 2-3x the thermal headroom and considerably less cores/threads across all workloads.

The frame.work people are very explicit that they’ve designed the cooling system to permit you to run the CPU fairly hard. It can dump 28W continuously according to https://frame.work/gb/en/blog/the-upgradeable-mainboard.

The i9 Apple laptops were notorious for not having a cooling system that could match the heat output of the CPU & throttling almost instantly.

Modern Intel laptop CPUs are no where near the performance / W of the M1, but they are considerably better than those that Apple stuffed into their laptops a few years ago with inadequate cooling I believe.

That was a firmware issue pre 2019 model that was corrected in a few weeks that got overblown. The throttling problem doesn’t exist anymore.

The only problem with the first generation framework laptop is intel’s quad core CPU because of thermal and power problems associated with their process nodes.

Hex core is really the lowest I will go now for anything over $1000. If this was a Ryzen 5800u, I’d have bought one already.

On my 1185g7 laptop, the browser benchmark Speedometer 2.0 scores 116. On my M1, I get 288. That's a bit more than "slightly ahead". The M1 absolutely smokes Tiger Lake and even Tiger Lake-H.

Sure, but that's expected because ARM essentially has hardware JavaScript acceleration built-in via FJCVTZS[0]. I personally don't find JavaScript specific benchmarks to be a good indicator of general computing power. (Ultimately, I spend little time in a web browser.)

FWIW, Cinebench multi-core benchmark is around 20% faster for me on the M1.

[0]: https://developer.arm.com/documentation/100076/0100/a64-inst...

> I personally don't find JavaScript specific benchmarks to be a good indicator of general computing power.

Many of us spend substantial periods of time writing up the technical documentation in Confluence or use JIRA. Confluence alone loads approximately 22Mb of mangled and compacted JavaScript for each open page in a separate browser tab.

Before M1, Confluence and JIRA were the single most source of a sustained battery drain, irrespective of the browser. On M1 based laptops, the battery impact is so negligible that it can be considered non-existent. How much of this subjected experience can be attributed just to the the ARM 8.3 JavaScript hardware acceleration is an interesting question as the JavaScript browser performance on Intel CPU's has also been affected by Spectre and Meltdown mitigations in the CPU microcode.

I doubt that instruction is relevant. It comes down to an abundance of microarchitectural resources on the M1. A 12th-gen Intel CPU gets 310 on this benchmark while having no particular accelerators. They just added tons of caches and buffers, like the M1 has.

Sure: 4 cores, but 8 hyperthreads.

LLVM builds eat memory in my experience & end up swapping if you don’t play games with the build system to reduce the parallelism of memory intensive parts of the build.

If you’ve got a VM or two sitting around then obviously that makes things worse.

Yeah I totally agree with that sentiment.

It’s often tempting to dismiss a product that doesn’t meet your own personal use cases but it’s a bit naive.

It’s actually fun to see products get popular that I know I wouldn’t buy myself. It doesn’t make them bad it just makes them not for me right now.

Is this really common? I mean using a laptop for such compute intensive tasks? I mostly use my macbook air m1 as a client to more powerful machines. I always have a few ssh sessions opened, a few tramp emacs buffers, some browser tab to jupyter notebooks running elsewhere...

I think it is quite common. I develop financial systems, and I compile apps regularly for running tests locally and large parts of the system is running locally using docker compose. This has the benefit of fast iteration for unit and system tests. Of course the tests and deployment is also run by a CI server but that is just for QA purposes so people dont have to remember to run their tests locally and to keep a clean main branch. I cannot imagine how long our feedback loop would be if we had to wait for the CI server for every change, or if we had a common dev server that everyone interacted with through a «dirty» development branch. Basically I think it is just different use cases. My experience with jupyter is that you use it for data analysis which is really more IO bound than cpu bound, also big data is often not possible to keep locally (because of the size and also in europe GDPR)

Note that I didn't necessarily refer to CI servers when talking about remote machines. Sometimes it's just a better specced workstation in the office, sometimes it's my own headless machine learning rig at home. Sure different engineers and different fields have different workflows but I believe it's still easier to build a good workstation from components than finding a laptop that excels in both portability and specs.

Interesting approach. I have some coleagues that actually remote desktops to their powerful desktop computer. I never did that because I fear all the tiny issues I have to deal with, but they tell me it works quite well. Also just offloading the heavy lifting through ssh or other types of build servers seems super interesting (I think some build systems have built in support for a common server, not sure if it is true, but it could be very helpful if it works well)

Yes, I specced my machines for local docker-compose of testing. It's pretty valuable, although different engineers use it different amounts.

Presumably, the people who were using desktop workstations pre-pandemic have started working from home with laptops - but their ways of working and tooling all center around doing things locally.

You sure this is true? I would bet the M1 is faster https://www.jeffgeerling.com/blog/2021/apple-m1-compiles-lin...

As he explicitly points out, the benchmark amounts to cross-compiling for the Intel CPU vs compiling native code for the Mac CPU, so it's not exactly even. I have no dog in this hunt, I don't have an M1 Mac and I don't care, but it felt wrong to let that stand without correction.

Kernel compiles are not particularly memory hungry. LLVM compiles on the other hand are very memory hungry, especially during linking.

I’m sure there are other 'workstation-ish' jobs that people would happily run on their laptop if it had > 16Gb memory. 64Gb is quite the leap.

1) There are 64Gb M1 MacBooks.

2) If you need 64Gb just sometimes, swapping on M1 MBP is insanely fast.

64GB ram for my lenovo legion was 250EUR, i could not configure a m1 with 64gb for less then 2900EUR i hate it that they bundled the ram with the m1 max i wanted to order the 64gb ram option but the difference between 32gb pro and 64gb max is 650EUR or more, and i just dont want to pay that to have more RAM.

Your laptop has DDR4 memory. I’ve checked prices for DDR5 memory 64Gb - around ~€500. So the difference with Apple is not so dramatic.

it cost 250EUR to go from 0GB to 64GB on lenovo, but it cost 650EUR to go from 32GB pro to 64GB max macbook pro 16, so it is dramatic for me. Im buying a macbook for 4-5 years.

You want to ignore the difference between DDR4 and DDR5, but the market doesn't ignore it and DDR5 is more expensive (I'm talking about non-Apple memory sticks).

Also, you want to ignore 400 Gb/s provided by Apple's 64Gb memory, but... you know.