Now that we can have 64GB/8GB in a shiny 16" MBP do you think it's best to buy a fully spec'd out 16" MBP for computing intensive DS/ML/Blockchain development?
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
MBP for Data Science/Machine Learning/Blockchain development
- Thread starter patent10021
- Start date
- Sort by reaction score
No Nvidia GPU mean no GPU acceleration for training models. A MacBook Pro would not be my choice of ML/AI unless you use it as a terminal to some sort of cloud system with acceleration.
Last edited:
Unfortunately, the lack of Nvidia support is a killer for intensive ML workloads, which no beefy specs will be able to compensate for. Solutions:
- Use your MBP as a terminal for writing and debugging code, then offload the training of your models to an external machine with GPUs (either a local server, if available, or a cloud-based one). This is what I do atm.
- Use a different OS, in which case there are plenty of options that offer much better bang for the buck (specifically, I'd rather build a dedicated workstation than spec out a laptop).
- Use your MBP as a terminal for writing and debugging code, then offload the training of your models to an external machine with GPUs (either a local server, if available, or a cloud-based one). This is what I do atm.
- Use a different OS, in which case there are plenty of options that offer much better bang for the buck (specifically, I'd rather build a dedicated workstation than spec out a laptop).
I would have to disagree about other OSes. MacBooks are used for DS at Google. Models are moving to and will most likely all exist in the cloud. I barely know any engineers and scientists who are using anything other than MacBooks. That includes robotics, DS, analytics, blockchain etc.
Anyway, I'm just talking about enough power to run notebooks, dashboards, visualizations, etc. Not so much big data. Besides, GPUs are highly prevalent in Deep Learning and Deep Learning is a small part of ML. That's why I was thinking 32GB/8GB might be enough for private research.
If we're using MBP as a Terminal then we might as well use an iPad Pro. Which is what I'd like to do if I didn't need Xcode as well.
Btw, AMD ROCm looks promising.
towardsdatascience.com
medium.com
AMD added code to TensorFlow a long time ago. There is support for TensorFlow. Amd GPUs support TensorFlow so this will be great for future AMD MBPs.
Anyway, I'm just talking about enough power to run notebooks, dashboards, visualizations, etc. Not so much big data. Besides, GPUs are highly prevalent in Deep Learning and Deep Learning is a small part of ML. That's why I was thinking 32GB/8GB might be enough for private research.
If we're using MBP as a Terminal then we might as well use an iPad Pro. Which is what I'd like to do if I didn't need Xcode as well.
Btw, AMD ROCm looks promising.
Train neural networks using AMD GPUs and Keras
Getting started with ROCm platform
towardsdatascience.com
AMD ROCm GPU support for TensorFlow
Guest post by Mayank Daga, Director, Deep Learning Software, AMD
AMD added code to TensorFlow a long time ago. There is support for TensorFlow. Amd GPUs support TensorFlow so this will be great for future AMD MBPs.
Last edited:
Yes, the MBP is the standard issue computer in the industry (and for good reasons: it still offers the best all-around environment for a scientist/engineer/developer), but this doesn't change the fact that the lack of Nvidia support for macOS is hampering itI would have to disagree about other OSes. MacBooks are used for DS at Google. Models are moving to and will most likely all exist in the cloud. I barely know any engineers and scientists who are using anything other than MacBooks. That includes robotics, DS, analytics, blockchain etc.
Anyway, I'm just talking about enough power to run notebooks, dashboards, visualizations, etc. Not so much big data. Besides, GPUs are highly prevalent in Deep Learning and Deep Learning is a small part of ML. That's why I was thinking 32GB/8GB might be enough for private research.
If we're using MBP as a Terminal then we might as well use an iPad Pro. Which is what I'd like to do if I didn't need Xcode as well.
Btw, AMD ROCm looks promising.
Train neural networks using AMD GPUs and Keras
Getting started with ROCm platform
AMD ROCm GPU support for TensorFlow
Guest post by Mayank Daga, Director, Deep Learning Software, AMD
AMD have added code to TensorFlow. There is support for TensorFlow which is progress towards AMD supporting AI libraries.
For my local ML and data science stuff, I went with 32 GB and the extra memory does help from time to time, but bumping up the GPU does not really give any tangible benefit for that kind of workload. On the other hand, one could argue that the upgrade is relatively inexpensive on the new 16", so might as well go for it.
On the topic of AMD-based acceleration: ROCm has been around for a while, and while it might be a step in the right direction, the reality is that Nvidia has too huge of a lead and no professional in the field uses AMD atm.
Also, last time I checked ROCm worked on Linux only. The only library I know of that supports AMD acceleration for macOS is PlaidML, which works decently on the MBP (although it's limited in the number of frameworks you can use with it). I wouldn't use it on the job, but for side projects and personal research it's fine.
Last edited: