NumPy / Sklearn performance comparison - macOS, Ubuntu, Windows 10 - use Anaconda

One of the advantages to use a Hackintosh is that you can try the same software on the same hardware across three operating systems: how does NumPy - and therefore Sklearn - perform across different operating systems and mathematics libraries?


I created a simple Sklearn routine, which utilises a single thread for its calculations.


Results

Hardware Os Python Math library Comment Time (s)
Macbook Air m1 MacOS Big Sur 3.9.1 CBLAS/LAPACK Anaconda install 22
Intel 9600k Ubuntu 20.04 3.8.5 OpenBLAS standard apt and pip install 27
Intel 9600k Ubuntu 20.04 3.9.0 OpenBLAS standard apt and pip install 27
Intel 9600k MacOS Catalina 3.8.5 MKL Hackintosh with Clover,
Anaconda install		 28
Intel 9600k Windows 10 3.8.5 MKL Anaconda install 28
Intel 9600k MacOS Big Sur 3.9.1 OpenBLAS Hackintosh with OpenCore,
brew and pip install		 29
Intel 9600k MacOS Big Sur 3.8.7 OpenBLAS Hackintosh with OpenCore,
brew and pip install		 32
Intel 9600k MacOS Catalina 3.8.7 OpenBLAS Hackintosh with Clover,
brew and pip install		 33
Intel 9600k Ubuntu 20.04 3.7.9 MKL Anaconda Intel proprietary
Python+Numpy		 34
Macbook Pro i5-7360U MacOS Catalina 3.8.5 MKL Anaconda install 38
Intel NUC N8280 Ubuntu 20.04 3.8.5 OpenBLAS standard apt and pip install 181
Raspberry Pi 4 Raspbian Linux 10 3.7.3 181


Visually, displaying only the same hardware (9600k)





Findings


There were three surprising findings:

First, the way Python and NumPy are installed does matter a lot: Anaconda's libraries are optimised and will give you an obvious performance edge compared to brew install. Install these using Anaconda instead of brew on macOS.

Second, setting up the development environment on Ubuntu was by far the easiest. Not only apt installed everything right away (no need for brew on Anaconda), the installed libraries were highly optimised and gave the extra performance that on macOS only the Anaconda install provided.

Third, the Apple M1 is really fast, even though it was fiddly to install NumPy - requires Anaconda, brew does not work. 


Takeaway message


Use the Anaconda installer - it is not only convenient, it is also highly optimised. Brew on macOS will not give you a good machine learning performance.

Comments

Post a Comment

Popular posts from this blog

MurMurHash3, an ultra fast hash algorithm for C# / .NET

Image
Finding good hash functions for larger data sets is always challenging. The well know hashes, such as MD5, SHA1, SHA256 are fairly slow with large data processing and their added extra functions (such as being cryptographic hashes) isn’t always required either. Performance and low collision rate on the other hand is very important, so many new hash functions were inverted in the past few years. One of the most notable ones is MurMurHash3, which is an improved version of its predecessor (v2). The new version can create both 32 bit and 128 bit hash values, making it suitable for a wide range of applications. 64 bit architecture The 32 bit hash isn’t too important for large amount of data matching, so I only worked with and tested the 128 bit version. The latter one is optimized for x86-64 architecture, giving amazing speeds for data processing. As a 128 bit value consists of two 64 integers, the processor is fairly good with working with such large hash values – they are not byte
. . .

Read more »

Quick select algorithm - find the Kth element in a list in linear time

Image
Quick select algorithm (Hoare's selection algorithm) – select the Kth element or  the first K element  from a list in linear time Working with large datasets is always painful, especially when it needs to be displayed in a ‘human readable’ format. It is a very frequent task to display only the largest, newest, most expensive etc. items. While sorting the whole dataset definitely gives a correct result, it is much slower than it needs to be – it needs at least O(n*log(n)) time and an it often uses recursion for the sorting, so in practice it can be quite slow. The quick select algorithm can get the top K element from a list of N items in linear time, O(n), with a very reasonable multiplication factor. The quick select does not use recursion so the performance is great for even large datasets. Algorithm The idea of the quick select is quite simple: just like with quicksort, select a random element from the list, and place every item that is smaller to the first half of the
. . .

Read more »

Octoprint as a systemd service - running it with high priority

Image
Octoprint provides remote access to your 3D printer, and as it's written in Python, you can run it on practically any operating system. If you are using a standard Raspberry PI image (for instance, because you run it on an Orange PI ), rather than the customised Octoprint image, you can manually install it and create a systemd service to run it: Add this to /etc/systemd/system/octoprint.service  [Unit] Description=Octoprint After=network.target [Service] ExecStart=/home/pi/octoprint/bin/octoprint serve WorkingDirectory=/home/pi/octoprint StandardOutput=inherit StandardError=inherit Restart=always User=pi CPUSchedulingPolicy=rr CPUSchedulingPriority=80 [Install] WantedBy=multi-user.target Then run sudo systemctl daemon-reload sudo systemctl enable octoprint This will start Octoprint on every reboot with a high priority (so unlikely to stutter during printing).
. . .

Read more »