Laptop for Machine Learning

[pascal]

The Windows Core Isolation - Memory Integrity option is enabled now.
No discernible performance hit with PyTorch (CUDA) running CNN's!

 

[Cyan]

have you solved the Intel GPU driver issue? I think my best friend IRL has the same problem with hit laptop, the GIGABYTE G5 KC-5ES1130SD.

 

[pascal]

I did not try. My focus was upgrading, testing all apps, improving security, and returning to ML tests.
Usually, the laptop is silent under regular office or programming use.
Only when running heavy ML tests that I hear the cooler working.

The Windows 11 Pro upgrade was automatic and smooth.
My subjective perception is that it is a solid OS, marginally faster than Win 10.
My daughter's L340 laptop also had an OS upgrade.

 

[Davros]

My daughter L340

What an unusual name she has

 

[pascal]

My daughter's L340 laptop also had an OS upgrade.

 

[Cyan]

My daughter's L340 laptop also had an OS upgrade.

it's a family thing then, preferring laptops. I also prefer laptops over desktop PCs, but maybe that's just me. Most of the people I know who have a PC usually prefer desktop computers. They have some advantages, sure, but the disadvantages don't make up for them and how complete and useful a laptop package can be.

 

[pascal]

Nowadays a 15" fast laptop is very good for an all-purpose computer (streaming, play, study, work).
You get used to the flexibility.

 

[BoardBonobo]

Have you tried running DirectML via WSL. This is what I do on my main machines (laptop i7 2060 and desktop 5900x and 6900XT) both running Win 11.

 

[pascal]

No, I haven't.
I use Anaconda with JupyterLab for small tests and Spyder for more robust tests.
Today more concentrated on testing some non-conventional CNN.
Think to combine it with EMs and Reinforcement Learning.

 

[pascal]

Waiting for Lenovo BIOS security corrections:

https://arstechnica.com/information...dels-fixed-to-prevent-unremovable-infections/

https://support.lenovo.com/us/en/product_security/LEN-73440#Lenovo Notebook

Maybe my next laptop will be a Dell.

 

[pascal]

Finally found the new BIOS.
I hope they will have more care about security in the future...

 

[Deleted member 2197]

Noticed this joint effort between Lambda and Razer for a ML laptop. Interested to see if the same is done with the Ada architecture.
April 2022

Lambda Teams Up With Razer to Launch the World’s Most Powerful Laptop for Deep Learning (lambdalabs.com)

Hardware

• 15.6” 2560x1440 165Hz display
• NVIDIA RTX 3080 Max-Q GPU with 16GB VRAM
• Intel i7-11800 Processor (8 cores, 2.3GHz to 4.6GHz)
• 64GB DDR4 memory
• 2 TB SSD storage
• Thunderbolt 4, USB 3.2, HDMI 2.1 ports
• Slim 4.4 lb aluminum unibody chassis
• 1080p webcam

Software

• Ubuntu Linux 20.04 LTS (Microsoft Windows dual boot optional)
• Lambda Stack with PyTorch, TensorFlow, CUDA, cuDNN, and NVIDIA Drivers
• One year of Lambda engineering support

Cost: From $3,499

 

[pascal]

This Razer/Lambda looks very good!

One possibility is have a Personal Cloud - PC

For example, continue to use a fast laptop with some GPU for development and enjoy the flexibility, but for big tests use a PC in LAN as ML server.
It could be a "cheap" PC like:
Intel i5 11400
32 GB RAM
RTX 3060 12GB (this is twice the memory of my laptop GPU have)
1 TB SSD

In general the big bottleneck for ML is the GPU´s RAM size! The bigger the GPU memory the bigger the ML Model it can run/train.
The CPU/Main Memory is for running some Python script driving the GPU computations, which is limited essentially by the GPU RAM size.
Reinforcement Learning and some Embbedings show some signs of CPU dependence. And it could be optimized by changing the language (C/C++ and Julia) and use some multithread, etc...

 

[pascal]

My impressions after two years of using the laptop: it works great as a multiple-purpose PC.
Wish list for a future upgrade (ascending order of priority): better display colors, lighter weight, less heat, and more GPU memory.

Also, it will be hard to find a gamer laptop upgrade with 2.5" slot available.