HN Explorer

Time to upgrade my laptop – need advice please

16 points by zyruh ↗ HN
Looking for a high-performance PC (non-Mac) for data-heavy workloads — suggestions?

It's time to upgrade. I'm looking for a Windows/Linux-compatible machine (no Mac) that can handle:

Large spreadsheet workloads (think multi-GB Excel/CSV files)

4K+ video editing and exports

Possibly some ML experimentation later on

I was considering a high-end gaming PC for the GPU + RAM benefits, but not sure if that’s the best tradeoff for thermals, noise, and form factor. Open to desktop or laptop, though portability is a plus.

What builds or machines would you recommend in 2025 for blazing speed + stability?

19 comments

[ 3.8 ms ] story [ 37.0 ms ] thread
bigyabai
Wait for Q1 2026 and Nvidia will have probably announced their laptop product. Otherwise your best shot is a beefy Ryzen laptop.
benoau
There's really only two choices in terms of laptop: a built in discrete GPU aka a gaming laptop, or one with the ability to add an external GPU through Thunderbolt or OCuLink.

Pretty much any desktop is going to outperform a laptop, especially the GPU.

LorenDB
External GPU is actually a really good option here. Get a laptop with a beefy CPU (definitely Ryzen; Strix Halo chips are really good and would be decent at standalone AI as well) and Thunderbolt/Oculink, then grab an eGPU dock and get the best GPU you can afford. For gaming, eGPUs suffer from bandwidth constraints, but for LLMs it shouldn't matter as much since the model only needs loaded once; after that it can sit in VRAM.
chrsw
A gaming PC is probably your best bet. PC workstations for ML are extremely expensive.
zyruh
I was kind of thinking this. thx
dlahoda
for local ai seems unified mem may win.

no non mac laptop today has that with good thermals.

so some boxes has amd ai 390 or 395 with 128gb.

for laptops need to wait for good thermal releases.

so it feels that macs dominate these days in all your points, and has good resell price later, except being mac)

zyruh
Excellent, thank you!
mixmastamyk
I’m happy with my AMD framework. Though you said ML so maybe want Nvidia?
zyruh
Cool - thx for the tip!
seized
Framework, the AMD versions.

Can take the 48GB SODIMMs that are on the market now. That'll handle a spreadsheet or two.

sandreas
I personally like the Ryzen AI Max+ PRO 395 (Strix Halo), where you get non-upgradable, but unified RAM. This means that up to 96GB of 128GB can be used as VRAM, which is great for AI workloads and are the only machines similar to Apples approach of unified RAM.

The Framework Desktop is a really interesting machine, however it lacks the PRO in the name, which means that you're going a bit cheaper, but missing out ECC RAM option, which might be interesting for a Workstation that needs to be as stable as possible.

The HP Z2 Mini G1a Workstation has PRO available, which therefore also has unified ECC Ram up to 128GB, but this adds a premium of around 500 bucks. Maybe not worth the money.

So if you wanna go Desktop, you could pick:

  Framework desktop with Ryzen AI MAX+ 395+
  HP Z2 Mini G1a with Ryzen AI MAX+ 395+ PRO
If you wanna go mobile, there are 2 machines:

  HP ZBook Ultra G1a
  ASUS ROG Flow Z13
I would never buy an mobile ASUS device that expensive (ASUS support can be a nightmare) but at the moment these are the only mobile devices with this chip.

Otherwise you have to go Full GPU, which is non-unified and either very expensive or you get far less VRAM for AI workloads.

Here is a test with more info: https://www.youtube.com/watch?v=oyrAur5yYrA

rajkumarsekar
For your use case, I’d recommend looking at Lenovo ThinkPad P1 Gen 6 or Dell XPS 15/17 with an Intel i9 or Ryzen 9, 32–64GB RAM, and NVIDIA RTX 4070 or higher. If you’re open to desktops, a custom build with a Ryzen 9 7950X, 128GB RAM, and a decent RTX 4080.