I understand that you want to know about whether the 18GB of RAM in your new MacBook Pro will be sufficient for running parallel MATLAB simulations using `parfor`. Let's break down the considerations:
2. Physical Cores: The MacBook Pro you mentioned has 6 performance cores. If you were to follow MathWorks' recommendation strictly, ideally you would want 24GB of RAM to fully utilize all 6 performance cores in parallel computations (4GB per core).
3. Efficiency Cores: The efficiency cores are typically not as powerful as performance cores and are designed to handle background tasks and improve energy efficiency. It is not clear whether MATLAB will effectively utilize these cores for `parfor` loops, but if it does, you would theoretically need even more RAM to fully utilize them.
4. GPU: If your MATLAB simulations can take advantage of GPU computing, then some of the computational load will be offloaded to the GPU. However, the GPU will also need to share the unified memory, which could further constrain the available RAM for CPU tasks.
5. Other Applications: Running additional applications like LaTeX, Word, PDF, PowerPoint, and browser tabs will consume some amount of RAM, leaving less available for MATLAB.
Given these considerations, if your simulations are memory-intensive and require more than what is available per core, then you might indeed face limitations. However, it is also possible that your simulations will not require the full 4GB per worker or that not all cores will be used at maximum capacity simultaneously.
If you find that you are running into memory limitations, there are some strategies you could employ:
- Optimize Code: Ensure that your MATLAB code is optimized to use memory efficiently. This might involve vectorization, preallocating arrays, or using data types that require less memory.
- Scale Down: Use fewer workers in your `parfor` loops to match the available memory. This means not all cores will be used, but it can prevent out-of-memory errors
- External Memory: MATLAB allows you to work with data that doesn't fit into memory using techniques like `tall` arrays or datastores, which can process data in chunks
- Cloud or Server: Consider running your simulations on a cloud service or a server with more RAM if your local resources are insufficient.
In summary, while you may not be able to fully utilize all 6 performance cores for memory-intensive parallel tasks, this doesn't necessarily mean your purchase was wrong. It depends on the specifics of your workload. If your simulations are not as memory-intensive as the guideline suggests, you may still be able to run effective parallel simulations. If memory does become a bottleneck, you can explore the strategies mentioned above to mitigate the issue.
Thank you,
Shubham Shreshth.
