3 ms·

I'm also considering that P14s AMD Gen4. But to do some ML stuff i need the damn Nvidia cuda cores :/ It has a USB4 port and could in theory run a eGPU but i ca

I'm also considering that P14s AMD Gen4. But to do some ML stuff i need the damn Nvidia cuda cores :/ It has a USB4 port and could in theory run a eGPU but i can't find any infos if it would work. Also it seams that the [1] intel 1360p is a bit faster? and i'm on the safer side because of real thunderbold support. [1] https://nanoreview.net/en/cpu-compare/intel-core-i7-1360p-vs-amd-ryzen-7-7840u https://nanoreview.net/en/cpu-compare/intel-core-i7-1360p-vs...

Baeocystin 3y ago
Every time a potential eGPU build has come up for me (~half a dozen times over the past few years) it's always made more sense in the end to simply build a generic ATX box, stuff the GPU in there, and just remote in. FWIW. The two times I tried anyway, because hey, fun new challenge, the primary lesson learned was that it was not worth it, most recently being ~6 months ago.
- delijati 3y ago
  Currently i do exactly that. I have a ATX box (AMD 2700x + Nvidia 3090) and access it remote, but i hate that it is idle most of the day and to transfer the data that i'm working on to it. (My notebook is my daily driver with company access) I just bought a eGPU case and will give it a shot, will report back :)
  - somerandomqaguy 3y ago
    I had a similar problem with needing a more powerful computer but not wanting to idle it all day long. The desktop however supported Wake on Lan, and the router stays on 24/7 anyways. So I ended up configuring SSH and software to send the magic packet on the router, and then sleeping the desktop. SSH in, send the wake up packet, then remote into the desktop. Put it back to sleep when I was done. There's still some quirks. You can't wake the desktop from a shutdown state, only a sleep state so you have to be a little careful only to put the desktop to sleep. But worked remarkable well and didn't require any additional hardware at least for me. The data transfer portion I don't know if I can offer any good ideas for you. My workflow was to treat the desktop as the primary workstation most of the time. Be curious to hear your experience with eGPU as well.
  - Baeocystin 3y ago
    Please do! I'm looking forward to hearing how it works out.
    - delijati 3y ago
      Got my Razer Core X (arch linux cuda was already installed as i have in my T14 a NVIDIA GeForce MX450) so far it works pretty well, thunderbold cable in set CUDA_VISIBLE_DEVICES=1 and have a lot more cuda cores ... yay. I use it with my spare 1070ti but will add my 3090 soon.
- yjftsjthsd-h 3y ago
  > The two times I tried anyway, because hey, fun new challenge, the primary lesson learned was that it was not worth it, Not cost effective or too buggy?
  - Baeocystin 3y ago
    Yes to both. On top of that, having the laptop be physically stuck while doing a long CUDA run was a worst-of-both-words situation. Much rather launch something and then be free to take the laptop somewhere else while the run percolates in the server room.

Baeocystin 3y ago

Every time a potential eGPU build has come up for me (~half a dozen times over the past few years) it's always made more sense in the end to simply build a generic ATX box, stuff the GPU in there, and just remote in. FWIW. The two times I tried anyway, because hey, fun new challenge, the primary lesson learned was that it was not worth it, most recently being ~6 months ago.

delijati 3y ago
Currently i do exactly that. I have a ATX box (AMD 2700x + Nvidia 3090) and access it remote, but i hate that it is idle most of the day and to transfer the data that i'm working on to it. (My notebook is my daily driver with company access) I just bought a eGPU case and will give it a shot, will report back :)
- somerandomqaguy 3y ago
  I had a similar problem with needing a more powerful computer but not wanting to idle it all day long. The desktop however supported Wake on Lan, and the router stays on 24/7 anyways. So I ended up configuring SSH and software to send the magic packet on the router, and then sleeping the desktop. SSH in, send the wake up packet, then remote into the desktop. Put it back to sleep when I was done. There's still some quirks. You can't wake the desktop from a shutdown state, only a sleep state so you have to be a little careful only to put the desktop to sleep. But worked remarkable well and didn't require any additional hardware at least for me. The data transfer portion I don't know if I can offer any good ideas for you. My workflow was to treat the desktop as the primary workstation most of the time. Be curious to hear your experience with eGPU as well.
- Baeocystin 3y ago
  Please do! I'm looking forward to hearing how it works out.
  - delijati 3y ago
    Got my Razer Core X (arch linux cuda was already installed as i have in my T14 a NVIDIA GeForce MX450) so far it works pretty well, thunderbold cable in set CUDA_VISIBLE_DEVICES=1 and have a lot more cuda cores ... yay. I use it with my spare 1070ti but will add my 3090 soon.
yjftsjthsd-h 3y ago
> The two times I tried anyway, because hey, fun new challenge, the primary lesson learned was that it was not worth it, Not cost effective or too buggy?
- Baeocystin 3y ago
  Yes to both. On top of that, having the laptop be physically stuck while doing a long CUDA run was a worst-of-both-words situation. Much rather launch something and then be free to take the laptop somewhere else while the run percolates in the server room.