Compute Requirements

Welcome to the ibm-skills-ai-colab-sessions wiki!

Purpose: To support the hardware architectural and technical requirements for running Jupyter Notebooks locally and remotely, with the view define, stand up and execute Machine Learning models and define areas of constraints/limitations for such.

Remote Compute

The preferred remote compute platform is Google Research's Colaboratory: .

Quick Start

To run these repository notebooks, with zero configuration, go to Quick Start

Local Compute

Local hardware requirements for Machine Learning Frameworks

SciKit-Learn
HuggingFace's Accelerate, Transfomers
TensorFlow & Keras.io
PyTorch

Minimum Requirements

Framework/Library	CPU	RAM	Storage	GPU	Python Version
SciKit-Learn	1 GHz dual-core	4 GB	5 GB	Not required	3.8+
Hugging Face (Accelerate & Transformers)	2 GHz quad-core	8 GB	10 GB	Not required	3.7+
TensorFlow/Keras	1.5 GHz dual-core	4 GB	8 GB	Optional	3.8-3.11
PyTorch	1.5 GHz dual-core	4 GB	5 GB	Not required	3.8+

Optimum Requirements

Framework/Library	CPU	RAM	Storage	GPU	Python Version
SciKit-Learn	2.5 GHz quad-core+	16 GB+	20 GB+ SSD	Not required	3.8+
Hugging Face (Accelerate & Transformers)	3.5 GHz octa-core+	32 GB+	50 GB+ SSD	NVIDIA, 8 GB+ VRAM	3.7+
TensorFlow/Keras	3.5 GHz quad-core+	16 GB+	50 GB+ SSD	NVIDIA, 4 GB+ VRAM	3.8-3.11
PyTorch	3.5 GHz quad-core+	16 GB+	20 GB+ SSD	NVIDIA, 4 GB+ VRAM	3.8+

Notes:

Requirements may vary based on the size and complexity of models and datasets.
GPU requirements are generally for NVIDIA GPUs with CUDA support.
macOS has limited GPU support for PyTorch and TensorFlow.
For large models or datasets, more RAM and GPU memory will significantly improve performance.

SciKit-Learn

Here's a concise list of the minimum and optimum hardware requirements to run scikit-learn using Jupyter Notebook on Mac, Linux, and Windows:

Minimum Requirements (all platforms):

CPU: 1 GHz dual-core processor
RAM: 4 GB
Storage: 5 GB free space
Python 3.8 or higher

Optimum Requirements (all platforms) ^{1, 2, 3}:

CPU: 2.5 GHz quad-core processor or better ²
RAM: 16 GB or more ²
GPU: ³
Storage: 20 GB or more SSD
Python 3.8 or higher

Notes:

¹ These requirements are general guidelines and may vary depending on the size and complexity of your datasets and models.
² For large datasets or complex models, more RAM and a faster CPU will significantly improve performance.
³ GPU acceleration is not natively supported by scikit-learn, so a dedicated GPU is not necessary unless you're using other libraries that can utilize it.

HuggingFace's Accelerate, Transfomers

A concise list of the minimum and optimum hardware requirements to run Hugging Face Accelerate and Transformers:

Minimum Requirements:

CPU: 2 GHz quad-core processor
RAM: 8 GB
Storage: 10 GB free space
Python 3.7 or higher

Optimum Requirements:

CPU: 3.5 GHz octa-core processor or better
RAM: 32 GB or more ^5/sup>
GPU: NVIDIA GPU with at least 8 GB VRAM (e.g., RTX 2070 or better) ^{5, 6, 7}
Storage: 50 GB or more SSD
Python 3.7 or higher

Notes:

⁴ These requirements can vary significantly depending on the size and complexity of the models you're working with.
⁵ For large language models (e.g., GPT-3, T5-large), more RAM and GPU memory are crucial.
⁶ While it's possible to run Transformers on CPU, a CUDA-capable NVIDIA GPU is highly recommended for reasonable performance, especially for training.
⁷The Accelerate library is designed to help run models on various hardware setups, including multi-GPU and TPU configurations.

TensorFlow & Keras.io

A list of the minimum and optimum hardware requirements to run TensorFlow, TensorFlow Keras (version 2.12), and Keras.io v3:

Minimum Requirements:

CPU: 1.5 GHz dual-core processor
RAM: 4 GB
Storage: 8 GB free space
Python 3.8-3.11 (for TensorFlow 2.12)
GPU: Optional, but recommended for better performance

Optimum Requirements:

CPU: 3.5 GHz quad-core processor or better
RAM: 16 GB or more ¹³
GPU: NVIDIA GPU with at least 4 GB VRAM (e.g., GTX 1060 or better) ^{9, 10, 13}
Storage: 50 GB or more SSD
Python 3.8-3.11 (for TensorFlow 2.12)

Notes:

⁸ These requirements can vary based on the size and complexity of your models and datasets.
⁹ For deep learning tasks, a CUDA-capable NVIDIA GPU is highly recommended for significantly faster training and inference.
¹⁰ TensorFlow 2.12 supports CUDA 11.2 and cuDNN 8.1 or higher.
¹² Keras.io v3 is a multi-backend Keras that can work with TensorFlow, JAX, or PyTorch as the backend. The requirements might slightly vary depending on which backend you choose.
¹³ For large models or datasets, more RAM and GPU memory will be beneficial.

PyTorch

A list of the minimum and optimum hardware requirements to run PyTorch for macOS, Linux, and Windows:

Minimum Requirements (all platforms):

CPU: 1.5 GHz dual-core processor
RAM: 4 GB
Storage: 5 GB free space
Python 3.8 or higher

Optimum Requirements (all platforms):

CPU: 3.5 GHz quad-core processor or better
RAM: 16 GB or more ¹⁶
GPU: NVIDIA GPU with at least 4 GB VRAM (e.g., GTX 1060 or better) ^{14, 16}
Storage: 20 GB or more SSD
Python 3.8 or higher

Platform-specific notes:

macOS:

GPU support is limited. Apple Silicon (M1/M2) Macs can use MPS (Metal Performance Shaders) backend for GPU acceleration.
Intel Macs don't have native GPU support for PyTorch.

Linux:

Best GPU support, especially for NVIDIA GPUs with CUDA.
Some support for AMD GPUs through ROCm, but more limited than NVIDIA.

Windows:

Good support for NVIDIA GPUs with CUDA.
No official support for AMD GPUs.

^NOTES:

¹⁴ GPU is optional but highly recommended for deep learning tasks.
¹⁵ Requirements may vary based on the size and complexity of your models and datasets.
¹⁶ For large models or datasets, more RAM and GPU memory will significantly improve performance.

Reference

Anthropic Claude, Sonnet 3.5 (2024) (Free), "Hardware Requirements for (X Library) on Mac, Linux, Windows", Last Accessed: July 2029 https://claude.ai/chat/34691fca-04a4-499d-8871-fab59dbef0ab
- The stats and requirements were generated for efficiency and AI drive search/pull of information, as well as summation and formatting.
- Any issues, mistakes and or hallucination, please open a [ticket].
- These stats will be manually checked, best efforts.