AI Notes

AI Deep Dive

Tag: cuda

  • Tensor Reduction (Sum) with PyTorch and CUDA

    Tensor Reduction (Sum) with PyTorch and CUDA Tensor Reduction operations involve aggregating the values in a tensor across one or more dimensions to produce a tensor with a smaller number of dimensions (or a scalar). The sum reduction operation computes the sum of all elements (or elements along specified dimensions) of a tensor. CUDA significantly… Read more

  • Tensor Reshaping with PyTorch and CUDA

    Tensor Reshaping with PyTorch and CUDA Tensor Reshaping involves changing the shape of a tensor without altering its underlying data. This operation is frequently used to prepare tensors for different operations in neural networks and other numerical computations. While the reshaping operation itself is typically not computationally intensive, performing it on a GPU using CUDA… Read more

  • Matrix Multiplication with PyTorch and CUDA

    Matrix Multiplication with PyTorch and CUDA Matrix Multiplication is a fundamental operation in linear algebra and is crucial in many machine learning algorithms, especially in the layers of neural networks. CUDA significantly accelerates this operation by parallelizing the numerous multiply-accumulate operations involved. Code Example with PyTorch and CUDA import torch # Check if CUDA is… Read more

  • Tensor Multiplication (Element-wise) with PyTorch and CUDA

    Tensor Multiplication (Element-wise) with PyTorch and CUDA Element-wise Tensor Multiplication, also known as Hadamard product, involves multiplying corresponding elements of two tensors that have the same shape. Utilizing CUDA on a GPU significantly accelerates this operation through parallel processing. Code Example with PyTorch and CUDA import torch # Check if CUDA is available and set… Read more

  • Tensor Addition with PyTorch and CUDA

    Tensor Addition with PyTorch and CUDA Tensor Addition is a fundamental operation in tensor algebra. It involves adding corresponding elements of two tensors that have the same shape, resulting in a new tensor of the same shape where each element is the sum of the corresponding elements of the input tensors. When performed on a… Read more

  • Accelerating Image Classification with CUDA

    Image Classification using CUDA CUDA (Compute Unified Device Architecture) significantly accelerates image classification tasks by leveraging the parallel processing power of NVIDIA GPUs. Deep learning models, which are commonly used for image classification, involve numerous matrix operations that are highly parallelizable and thus benefit greatly from GPU acceleration via CUDA. How CUDA Accelerates Image Classification… Read more

  • CUDA vs. ROCm for LLM Training

    CUDA vs. ROCm CUDA (Compute Unified Device Architecture) and ROCm (Radeon Open Compute) are the two primary software platforms for General-Purpose computing on Graphics Processing Units (GPGPU) used in accelerating computationally intensive tasks, including the training of Large Language Models (LLMs). CUDA is developed by NVIDIA and is designed for their GPUs, while ROCm is… Read more

  • How CUDA Solves Transcendental Functions

    How CUDA Solves Transcendental Functions CUDA leverages the parallel processing power of NVIDIA GPUs to efficiently compute transcendental functions (like sine, cosine, logarithm, exponential, etc.). It achieves this through a combination of dedicated hardware units and optimized software implementations within its math libraries. 1. Special Function Units (SFUs) Modern NVIDIA GPUs include Special Function Units… Read more

  • Exploring CUDA (Compute Unified Device Architecture)

    Exploring CUDA CUDA is a parallel computing platform and programming model developed by NVIDIA for use with their GPUs. It allows software developers to leverage the massive parallel processing power of NVIDIA GPUs for general-purpose computing tasks, significantly accelerating applications beyond traditional CPU-bound processing. 1. CUDA Architecture: The Hardware Foundation NVIDIA GPUs are designed with… Read more

  • Can AMD GPUs Train LLMs?

    Can AMD GPUs Train LLMs? AMD GPUs can be used to train Large Language Models (LLMs). While NVIDIA GPUs, particularly those with CUDA architecture, have historically dominated the LLM training landscape, AMD has been making significant strides in this area with its ROCm (Radeon Open Compute) platform. 1. ROCm Platform ROCm is AMD’s open-source software… Read more

  • AMD GPUs vs. NVIDIA GPUs for LLM Training

    AMD GPUs vs. NVIDIA GPUs for LLM Training Here we dive into how AMD GPUs can be used for LLM training, and compare them directly with the dominant player in this field: NVIDIA GPUs. Comparison: AMD vs. NVIDIA GPUs for LLM Training Feature NVIDIA GPUs AMD GPUs Dominant Architecture/Platform CUDA (Compute Unified Device Architecture) –… Read more

  • Competition Between NVIDIA and Broadcom Offerings

    NVIDIA vs. Broadcom: Competition (April 2025) Historical Differentiation NVIDIA: Pioneered & dominates the general-purpose GPU market, with a strong foothold in AI, gaming, & professional visualization. Their CUDA platform is a significant barrier to entry. Broadcom: Traditionally a leader in custom ASICs for networking & communication infrastructure. Their entry into custom AI silicon leverages their… Read more

  • NVIDIA vs. Broadcom: Future Direction

    NVIDIA vs. Broadcom: Future Directions (April 2025) NVIDIA: Future Direction NVIDIA’s future strategy is deeply rooted in its leadership in accelerated computing, aiming to be the foundational platform for the AI era across diverse industries. Their vision extends beyond just selling chips to providing a comprehensive ecosystem of hardware and software. Continued Advancement in AI… Read more

  • RAG with locally running LLM

    Sample code to enable running the LLM locally. This will involve using a local LLM instead of OpenAI. Key Changes: To run this code with a local LLM: Important Considerations: Read more

  • Tensor

    PyTorch‘s fundamental data structure is the Tensor. It’s the central object for numerical computation in PyTorch, analogous to NumPy’s ndarray but with added capabilities for GPU acceleration and automatic differentiation (crucial for deep learning). Here’s a breakdown of PyTorch’s data structure landscape, with the Tensor at the core: 1. Tensors (torch.Tensor) 2. NumPy Arrays (numpy.ndarray)… Read more