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.
Haha ymmd ;)
Yeah, I thought it might either have been a misclick OR wrong information on my side (IF there is something wrong with it, I'd still like to know what...)
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.
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.
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.
I have been carrying miniPC's between locations instead of laptops when there are keyboard, mouse, and monitor at each desk that I can utilize when I am sitting there. Beefy power adapters which are already in place are good to have too, I might use multiple external HDDs.
Actually a miniPC alone can be more convenient to carry than a laptop.
This HP is about twice the size I consider "normal" but more than twice as powerful :)
A bit less than twice the weight, but more than twice the money though.
This size a brick would still be as feasible as when I occasionally carry two miniPC's, I know I could handle it, but then it barely may be more convenient than a laptop.
So I'm sold, even if I don't exactly need one right now.
Hopefully these will fly off the shelf and they'll come out with a much more powerful unit real soon for about the same price, and move up from there. Ideally a whole new generation of something trending could emerge that can be made desirable in ways not possible in 2025, allowing prices of this exact model to be slashed beyond the bone.
Hopefully before the next couple years is up, looking forward to it ;)
They're (over)due to start shipping miniPCs aimed at being AI/ML workstations, using a SoC that's a collaboration between Mediatek and NVIDIA, with 128GB of DRAM. They have not yet announced intentions to put those chips in laptops and run Windows on them, but the writing has been on the wall all year. It would be entirely unsurprising to see NVIDIA(+Mediatek) overtake Qualcomm for ARM Windows PC sales next year. But it's too soon to start making purchase plans around that.
https://www.tuxedocomputers.com/en/TUXEDO-InfinityBook-Pro-1...
I'll get myslef one of these as well when I cash in my invoice for last month's work.
Nice, I do love the configurableness, and the prices seem reasonable.
Yeah, totally.
Thank you! I'll check it out.
Framework, the AMD versions.
Can take the 48GB SODIMMs that are on the market now. That'll handle a spreadsheet or two.
Great - thx!
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:
If you wanna go mobile, there are 2 machines: 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
Can anyone tell my why I got downvoted? Is this wrong advice?
People can be jerks.
Haha ymmd ;) Yeah, I thought it might either have been a misclick OR wrong information on my side (IF there is something wrong with it, I'd still like to know what...)
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)
Excellent, thank you!
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.
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.
This is great - thank you!
Workstation laptops can be specified with powerful GPU’s.
You can spend $10k on a Dell with 4 drive capability, 128gb of ram and 24gb of gpu.
Yes but then you have a laptop edition card which will just never compete with desktop.
Super helpful - thank you!
Thank you!
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.
Thank you - Yes, I love the Lenovo ThinkPad!
[dead]
A gaming PC is probably your best bet. PC workstations for ML are extremely expensive.
I was kind of thinking this. thx
You get what you pay for.
Yeah, absolutely.
[dead]
Thank you!
What is your budget?
about $2-$3k.
https://h20195.www2.hp.com/v2/GetPDF.aspx/c09086887
Not sure how much the 128 GB RAM version costs or when it will ship.
I like the looks of those.
I have been carrying miniPC's between locations instead of laptops when there are keyboard, mouse, and monitor at each desk that I can utilize when I am sitting there. Beefy power adapters which are already in place are good to have too, I might use multiple external HDDs.
Actually a miniPC alone can be more convenient to carry than a laptop.
This HP is about twice the size I consider "normal" but more than twice as powerful :)
A bit less than twice the weight, but more than twice the money though.
This size a brick would still be as feasible as when I occasionally carry two miniPC's, I know I could handle it, but then it barely may be more convenient than a laptop.
So I'm sold, even if I don't exactly need one right now.
Hopefully these will fly off the shelf and they'll come out with a much more powerful unit real soon for about the same price, and move up from there. Ideally a whole new generation of something trending could emerge that can be made desirable in ways not possible in 2025, allowing prices of this exact model to be slashed beyond the bone.
Hopefully before the next couple years is up, looking forward to it ;)
Sweet - thank you!
That’s limited for the workload you describe.
What do you think is a more reasonable budget?
Enough to buy a workstation class machine.
Alternatively, a scope that matches your budget.
Thank you
I guess that means threadripper 9995 pro and 2x rtx 6000 is out.
I’m happy with my AMD framework. Though you said ML so maybe want Nvidia?
Cool - thx for the tip!
Wait for Q1 2026 and Nvidia will have probably announced their laptop product. Otherwise your best shot is a beefy Ryzen laptop.
I've been looking at Ryzen - thx for the tip!
A laptop product for ML and not for graphics?
They're (over)due to start shipping miniPCs aimed at being AI/ML workstations, using a SoC that's a collaboration between Mediatek and NVIDIA, with 128GB of DRAM. They have not yet announced intentions to put those chips in laptops and run Windows on them, but the writing has been on the wall all year. It would be entirely unsurprising to see NVIDIA(+Mediatek) overtake Qualcomm for ARM Windows PC sales next year. But it's too soon to start making purchase plans around that.
Thank you!
Unless it uses an entirely bespoke GPU architecture, presumably it will support both.
Thank you!
[dead]