Budget (including currency): ~3000$
Country: EU
Games, programs or workloads that it will be used for: Python, Jupyter Notebook, Tensorflow, Keras, PyTorch.
Other details (existing parts lists, whether any peripherals are needed, what you're upgrading from, when you're going to buy, what resolution and refresh rate you want to play at, etc):
So hello everyone, I am a PhD Student, doing machine learning research and I am in the market for a new laptop. I currently have a early 2015 Macbook Pro 13", which is, well, aged enough to make me consider an upgrade. So I would like your help and suggestions to make a choice.
Effectively what I am looking for is laptop that's gonna last me for quite some time, is going to be reliable and have a lot of battery life (under normal work loads and not neural network training obviously). I mention training neural networks, which should hardly be the case in most cases as it is unreasonable to train complete models on a laptop. My most demanding workload will be consisted of development/coding and testing (training) smaller iterations of my models before I deploy them on a GPU server that is gonna handle the days/weeks of training required.
To that end, I have concluded that my 13" screen is too small to spend so much time on it and I am aiming for something like a 15"-16" laptop. I have narrowed it down to 3 choices that I would like to list in a "best first" manner:
1) Macbook Pro 16" - Late 2019 - with a dedicated GPU in the form of 5600M (8GB) and 32GB or ram.
2) LambdaLabs Tensorbook (details here: https://lambdalabs.com/deep-learning/laptops/tensorbook )
3) Razer Blade 15 Advanced Edition with RTX 2080
However, I am facing a lot of tough choices here.
While I would prefer the Macbook, Apple doesn't seem to play well with Nvidia which I would require for running experiments (even small ones) locally on GPU. There is an alternative to using, let's say tensorflow, with an AMD card with ROCm but I would have to boot Ubuntu - which pretty much negates the point of buying a mbpro. I could buy a nvidia eGPU but there's no guarantee it's going to work in the long term and every MacOS update could potentially brake the compatibility. And last and worst I could use Keras with PaidML but for anyone doing active research would know that Keras is not the way to go if you want to do innovative research and publish your work (so that's off the table).
As my "best option" for the mbpro is to book on Ubuntu, why pay the premium of apple and not just go for the LambdaLabs Tensorbook which seems to be tailored to my needs (per say). The only reason I'm sceptical about LambdaLabs is because a) I have never heard of them before today and b) what's going to happen if I have a service issue given that I'm based in EU and the company is based in the US. I can't go without a laptop for a long period of time...
Razer Blade comes third only because I feel like I'm paying extra for literally features (like a high refresh rate monitor) that I don't need. Otherwise I feel like it has a good enough build quality and overall specs to do what the Tensorbook would do for me (given that it's going to run Ubuntu anyway).
Some quick quirks of mine:
- I hate working on windows and I refuse to do so. For anything but work, Windows is OK, I just can't stand working on windows.
- I understand that if you read this far you might be confused as to why I need a GPU at all if I am going to run the majority of my workload to remote servers, but I would like to afford the luxury of running small experiments in my laptop before I just go off and wait 72hours for results. I could technically do it in CPU (which all 3 could cover as they are Intel based), but time is money and money is what I need (being a student and all), so I'm looking for more effective options so to speak.
Also, should I wait for the next generation of macbook pro's coming out in a short while? I am clearly not thinking about Apple Silicon systems which will most likely not be compatible with my needs, but for the new Intel processors. Is it worth the wait, performance wise, if I end up using CPU for the experiments that is.
P.S. I would especially love to hear what you're rocking if you're a machine learning developer or data scientist or what have you.