Hello! For work (where I’m compiling a big Rust monorepo in a loop), I’m on a Thinkpad T14 Gen 4 and it’s getting a bit annoying.
My only constraint is that I’d like a 13-14 inch device
The AI memory situation is probably to blame, but after sending a LLM fetch specs and GeekBench results, it seems that for sustained multicore performance, the only thing that comes close (within 10% but still lower) of a M5 MacBook Pro (currently ~3200 EUR), is the HP Elite G1a (ARM linux is OK), but its list price is much higher (up to 4900 EUR for 64G memory and 1TB disk) for arguably worse speakers and battery on top.
So right now it’s hard to justify not getting a MacBook and that makes me sad…
Is there a laptop you’re using right now that comes close enough to a MBP in long compilation jobs, but runs Linux and doesn’t somehow cost more ?
What are you really looking for? You mention GeekBench, multi-core performance, battery life, speakers, memory, and screen size. But you mention Rust compilation. So I’m confused, do you want a powerful machine or one that has a long battery life or a machine that looks and sounds good?
I see EUR and fr-fr, so you must not be in the US but in France or Belgium.
Also your comparison is not really equal. You’re comparing a M5 Macbook Pro with an M5 Pro 18-core CPU, 20-core GPU with 24 GB system memory (3200€), and 1TB SSD to a HP Elite G1a with 64GB and 1TB SSD (4900€). If you give the macbook pro 64GB of memory, then you’re at 4300€. If you take the default M5 Max config (18 core CPU, 32 Core GPU) and the default 36GB RAM, you’re also at 4800€.
What is your budget?
Looking at nano review and sorting by multi-core performance, this is the list:
CPU Multicore Cinebench Core Ultra 9 285HX 2240 Core Ultra 9 290HX Plus 2195 Ryzen 9 9955HX3D 2160 Ryzen 9 9955HX 2091 M5 Max (40-сore GPU) 2073 M5 Max (32-сore GPU) 2066 M4 Max (16-Core) 2044 Core i9 13980HX 2014 Snapdragon X2 Elite Extreme (X2E-96-100) 1967 M5 Pro (18-Core) 1949 (why not geekbench for multi-core?).
As you can see, AMD and Ryzen do well (and that snapdragon), all the way to Intel Core i9 13980HX.
Here’s a Tuxedo laptop with a Ryzen 9 9955HX and 36GB RAM for ~2800€. Make it 64GB and you’re at ~3500€.
Or here, a StarFighter Ultra with Intel Core Ultra 9 285H, 64GB RAM, 1TB storage.
Otherwise there’s linux preloaded where you can find stores that sell laptops.
I meant that I don’t care about battery life/speakers/screen, it’s only a tiebreaker if all other things (price, CPU sustained perf) are mostly equal.
I missed the memory part indeed, going up to 64GB changes a lot, I haven’t found a HP g1a with ~48GB of memory which is what I think is the good amount for…
…My budget which is 3000-3500 EUR
Indeed I’m in France
Thanks for those links, I knew GB was flawed (mostly because it’s user-provided data, if nothing else), but had no idea about alternatives
AI
🍿
I honestly just SSH into my workstation from my laptop, into a tmux session, for all my heavy LLM tasks.
I’m not even using it for AI or heavy local inference or so. I just happen to easily fill up and use 40-is GB, and when I want ram in a package that performs well CPU wise it gets giga expensive fast :(
How big is just the repo?
Currently 260k loc in the workspace
I mean in MB/GB
I spend my day in RDP/SSH
Unfortunately not really. I use a framework as my day to day and offload builds to a Macbook pro.
I’m using a laptop with strix halo as my main machine (ryzen 395 in hp zbook ultra g1a). Came from macbook pros and compile large c++ repos frequently. This thing is a beast! Running debian 13.
The only thing it’s noticably worse than a macbook is its battery life. Running local llms, play high end games. All on 55-70W TDP.
Was a part time linux user since 2005 but Apple’s “Liquid Ass” finally made me switch full time to Linux :D
yeah this is why I have a server for compiling. would never compile on my rigs, would take forever. I mean you can score an offsite server for cheap if you shop around. A LOT cheaper than buying another machine that’s for sure.
Yes. There is no “best laptop for compiling Rust”, with a reasonable price tag.
All laptop CPUs are extremely weak versions of desktop CPUs and rust is famously heavy on CPU during compilation.
The winning move here is to not use a laptop.
Framework if you dont mind relying on the Community for coreboot support.
Otherwise System76, Novacustom, Starlabs, Nitropad, … there is a handful of sellers of coreboot laptops which is a market you should totally invest in. Coreboot is amazing and there is quite a variety!
Also not to forget Chromebooks, but they are extremely limited and mostly unrepairable.
Found one.
The closest comparison to the Macbook Pro is probably the Framework 13 Pro with Intel Core Series 300. The newest Panther Lake chips are close to Apple’s M series silicon in terms of efficiency (and therefore much better battery life than previous models) and competitive when it comes to performance. They even market it as the MBP for Linux users.
The 13 Pro has excellenr build quality and a haptic touchpad, just like the Macbooks. And unlike the Macs, it’s super repairable, upgradeable memory, has modular ports, and supports Linux really well! And especially when you spec it with more RAM and storage (even in memory crisis), the Framework is incredibly close to the price of the equivalent base M5 Macbook (it goes more to show how obscene Apple’s memory pricing is really).
Alternatively, you could get an older M1 or M2 Mac and run Asahi Linux (but then you can’t upgrade the RAM or storage)
What was really the problem? If you compile it the first time, the second time it only compiles the files you’ve changed. No matter how big your monorepo is, as long as it compiles overnight, there shouldn’t be any impact to the speed of work.
So there’s this fun thing in rust where if you don’t have exactly the same feature set for all components of the workspace when you go to a subfolder for a component you can invalidate cached artifacts and trigger a lot of jobs. Same thing when you check out a branch then compile. And as I have a fair volume of crates in the workspace, changing mtime on the files for a checkout means I still have 5~10 minutes each time I want to run tests locally (a fresh run takes ~30 minutes to ~45}
I’ve never worked with a rust codebase that big. I’ve never had to deal with issues like that but damn, 10 minutes to run a test is crazy.
the framework laptops https://frame.work/
I looked at phoronix benchmarks for the 13 Pro first, but I wasn’t impressed (even acknowledging that Phoronix should have asked for the Ryzen one)
Well if Panther Lake doesn’t work for you, I don’t think any x86 machine will. You might want to try Strix Halo, but those are more power hungry and aren’t in many laptops (I think it’s the weird ROG tablet and some HP machine that isn’t all too great). You might also be interested in the Framework Desktop, a fun little Striz Halo box, but it’s not the most portable
Stay away from Snapdragon machines, they don’t perform as good as M series silicon and Linux support is pretty poor.
Your best alternative is probably an M1 or M2 Macbook Pro running Asahi Linux, since you are fine with ARM.
Doesn’t matter too much.
I highly recommend avoiding anything designed specifically around Linux because you’re going to be paying hundreds of dollars extra for inferior hardware.
Framework is a scam. System76 is a scam. Anyone who disagrees is a useful idiot.
I’m not saying you’re wrong, but saying “anyone who disagrees is a useful idiot” is not a sign of being particularly bright either.
