Deep Learning Laptop configuration critique elicited
I am seeking to purchase a home computer for general use and deep learning, which requires significant GPU resources.
1. Mobility is highly preferred, e.g., laptop
2. Price is important, e.g., no "the sky is the limit"
3. Energy efficiency is a goal. If I'm not using it specifically for deep learning, I'd greatly prefer to have a low consumption computer (for a low carbon footprint).
4. I'll be happy to start with Win 10, though I will undoubtedly dual boot it to Ubuntu like I have to 4 other machines in past, and use whichever OS provides the best overall results (noting some configuration and fan speed quirks with Linux and eGPUs).
I had considered a purpose built deep learning desktop, though it has no mobility and draws too much power in 'normal' operation.
I am considering the following;
- Alienware 17 Gaming Laptop
- Intel Core i7-8750H
- 16G DDR4 RAM
- GeForce GTX 1070 Max-Q 8G
- 1 TB hybrid drive
- 1 Thunderbolt port, 3 USB 3.0 "Superspeed" ports
This gives me the flexibility to;
- Use the embedded GPU as at least a model experimentation starting point, where I can then shift to a more powerful eGPU with more memory, an AGA, an AWS p3 VM, or Kaggle.
- Upgrade eGPUs as they improve in power, memory, and price reduction
As deep learning can run 24/7 at full GPU load putting a significant thermal demand on laptop components, I'm paying especially careful attention to various cooling approaches, and will monitor CPU/GPU temperatures closely.
I may also add an SSD drive to store the learning data on during training, to avoid long waits for batch pulls from the HDD.
I don't know how to tell if the motherboard (R5?) contains the Thunderbolt circuitry, or if it is on a daughter board.
All thoughts/critiques welcomed!
I wouldn't put Alienware 17R5 in the energy efficient or low carbon footprint category. However for the other needs, you can't really do better than an Alienware laptop. You can actually have up to 4 discrete graphics cards if you wanted to in Linux. Besides the dGPU you have Thunderbolt 3 port, AGA port, and a second NVMe M.2 slot. AGA should be your first choice for adding the eGPU because of its relatively reasonable cost and good performance. Thunderbolt 3 is next due to convenience. M.2 eGPU adapter is your last choice but it provide good performance and low cost.
Pop!_OS is a good Linux distro that gives you an option to disable the dGPU right out the box to save energy. Windows will take some work getting multiple eGPU going. I've tried two Nvidia eGPUs with a 15R3. The Thunderbolt 3 components are all on board. Alienware laptops have the best Thunderbolt 3 connection (TB3 controller directly to CPU). The AGA connection also enjoys direct CPU attachment.
I am working on a blog entry addressing this exact setup theme. I am using a Dell XPS 15 9500, which should come cheaper/roughly the same as the Alienware, and it is somewhat more efficient. The XPS13 is definitively low power, and they both have two thunderbolt .
The amount of RAM you have is probably a little too low. 24/32GB will be better. Otherwise looks promising! I would also consider the Mantiz Saturn Pro II as it has the best performance-dollar ratio. but AMA with alienware has great co-support.
Love the idea of Pop!_OS. I am currently working with Ubuntu and its been...troublesome.
Good luck. looking forward to hearing about your progress