Skip to content
Pritiks23Public

About

No description, website, or topics provided.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Latest commit

Β 

History

27 Commits

Folders and files

NameName
Last commit message
Last commit date
Β 
Β 

Repository files navigation

Hello, I'm Pritika πŸš€

GPU Systems Engineer | CUDA β€’ PyTorch Distributed Training β€’ GPU Optimization

I build and benchmark GPU workloads focused on understanding how hardware, memory, and distributed computing impact modern AI systems.

My work focuses on:

Focus Area Technologies
GPU Programming CUDA, CUDA Memory Management, GPU Optimization
Distributed Training PyTorch DDP, NCCL, DistributedSampler
Model Scaling Tensor Parallelism, Transformer Parallelization
Deep Learning Systems PyTorch, Mixed Precision Training
Languages Python, C++, C

πŸš€ Featured GPU Projects

🟒 GPU Graph Memory Lab

CUDA benchmarking suite analyzing how GPU memory allocation strategies impact graph analytics performance.

Workload

  • Algorithm: PageRank
  • Dataset: SNAP citation graph
  • Representation: CSR Graph Format
  • Hardware: NVIDIA RTX 3090

Memory Strategies Tested

Strategy Result
cudaMalloc Baseline allocation
cudaMallocAsync πŸ₯‡ Best runtime
Memory Pool Improved reuse, higher memory usage
Unified Memory Higher overhead

Key Finding

cudaMallocAsync achieved the best performance by reducing allocation overhead for repeated GPU memory operations.

πŸ”— View Repository

🟣 EuroSAT GPU Training Pipeline

Screen Shot 2026-08-06 at 1 25 45 PM

End-to-end GPU accelerated computer vision training pipeline built with PyTorch.

Implemented

βœ… CUDA accelerated preprocessing
βœ… Mixed precision training (AMP)
βœ… GPU data loading optimization
βœ… Training performance analysis
βœ… Model evaluation and visualization

Results

Epoch Validation Accuracy
Epoch 1 73.6%
Epoch 20 94.1%

Goal

Analyze how GPU utilization, preprocessing acceleration, and training pipeline optimizations affect deep learning throughput.

πŸ”— View Repository

πŸ”΄ GEMM Kernel Optimization Study

                GEMM Benchmark
                       |
    ----------------------------------------
    |                  |                   |
    v                  v                   v

CPU Baseline      GPU Library        Custom CUDA Kernels
    |                  |                   |
    |                  |          --------------------
    |                  |          |                  |
    v                  v          v                  v

CPU NumPy       PyTorch CUDA   Naive CUDA       Tiled CUDA
                                Global Memory    Shared Memory

CUDA benchmarking study analyzing how different matrix multiplication implementations impact GPU performance.

Workload

  • Operation: Matrix Multiplication (C = A Γ— B)
  • Matrix Size: 4096 x 4096
  • Precision: FP32

GEMM Strategies Tested

Strategy Runtime
CPU NumPy 221.75 ms
PyTorch CUDA GEMM 2.98 ms
Naive CUDA Kernel 38.97 ms
Tiled CUDA Kernel 21.18 ms

Key Finding

GPU GEMM achieved ~74.5x speedup over CPU execution.

Shared memory tiling improved custom CUDA kernel performance by 45.6% through improved data reuse.

πŸ”— View Repository

Distributed AI Systems

πŸ”΅ PyTorch DistributedDataParallel

Multi-GPU training implementation using PyTorch distributed systems.

Architecture

              Model
                |
     -------------------------
     |           |           |
   GPU 0       GPU 1       GPU 2

          NCCL AllReduce

Implemented

  • PyTorch DistributedDataParallel
  • NCCL communication
  • DistributedSampler
  • Mixed precision training

Results

  • Multi-GPU RTX 3090 training
  • Improved throughput through distributed execution

🟠 Tensor Parallel Transformer

Explored model parallelism techniques for transformer architectures.

Architecture

          Transformer Layer

              Linear

        ----------------
        |              |
      GPU 0          GPU 1

     Partitioned computation
              |
        Combined output

Implemented

  • Tensor parallel linear layers
  • Distributed model execution
  • GPU workload partitioning

🧠 GPU Engineering Focus

CUDA Programming
        ↓
GPU Memory Optimization
        ↓
Distributed Training
        ↓
Efficient AI Infrastructure

Currently exploring:

  • GPU performance optimization
  • CUDA-based acceleration
  • Distributed deep learning systems
  • Efficient AI workloads

πŸ›  Technical Stack

Category Tools
GPU Computing CUDA, CUDA Toolkit, GPU Memory Management
AI Frameworks PyTorch
Distributed Systems DDP, NCCL, Tensor Parallelism
Programming Python, C++, C
Development Linux, Docker, Git

πŸ“Š Engineering Highlights

Area Highlights
GPU Computing CUDA memory allocation benchmarks, GPU optimization experiments
Distributed Training PyTorch DDP with NCCL multi-GPU communication
Model Parallelism Tensor Parallel Transformer implementation
Computer Vision GPU accelerated EuroSAT training pipeline
Performance Engineering Runtime analysis, memory profiling, workload benchmarking

πŸ§ͺ Current Focus

πŸ“« Connect

LinkedIn: https://www.linkedin.com/in/pritika-vipin/

About

No description, website, or topics provided.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors