[Paper Review] 3DES ECB Optimized for Massively Parallel CUDA GPU Architecture

Łukasz Świerczewski|arXiv (Cornell University)|May 19, 2013

Chaos-based Image/Signal Encryption4 references3 citations

TL;DR

This paper presents a highly optimized CUDA kernel for 3DES encryption in ECB mode, leveraging the massive parallelism of GPU architectures. By mapping independent 64-bit blocks to individual CUDA threads, the implementation achieves significant performance gains over CPU-based encryption, demonstrating the feasibility of GPU acceleration for legacy symmetric ciphers in compute-intensive workloads.

ABSTRACT

Modern computers have graphics cards with much higher theoretical efficiency than conventional CPU. The paper presents application possibilities GPU CUDA acceleration for encryption of data using the new architecture tailored to the 3DES algorithm, characterized by increased security compared to the normal DES. The algorithm used in ECB mode (Electronic Codebook), in which 64-bit data blocks are encrypted independently by stream processors (CUDA cores).

Motivation & Objective

• To explore the feasibility of accelerating 3DES encryption using GPU architectures.
• To optimize the 3DES ECB mode for execution on massively parallel CUDA GPU hardware.
• To evaluate performance improvements over traditional CPU-based 3DES implementations.
• To demonstrate the effectiveness of thread-level parallelism in processing independent 64-bit blocks.

Proposed method

• The 3DES algorithm is implemented in CUDA C++ using a kernel where each thread processes one 64-bit block independently.
• ECB mode is exploited, as it allows independent encryption of each block, making it highly suitable for GPU parallelization.
• The implementation uses standard DES S-boxes and key scheduling logic, adapted for GPU execution with coalesced memory access patterns.
• Kernel launch configurations are tuned to maximize occupancy and hide memory latency on the GPU.
• The design minimizes thread divergence and maximizes throughput by ensuring all threads execute identical instruction paths.
• Performance is measured in encryptions per second, comparing GPU throughput to sequential CPU execution.

Experimental results

Research questions

• RQ1 Can 3DES ECB be efficiently parallelized on a GPU using CUDA?
• RQ2 How does GPU-based 3DES performance compare to CPU-based implementations?
• RQ3 What is the maximum throughput achievable using a massively parallel GPU architecture?
• RQ4 What are the key optimization strategies for minimizing latency and maximizing occupancy in GPU 3DES kernels?

Key findings

• The GPU-optimized 3DES ECB kernel achieves significantly higher throughput than CPU-based implementations due to massive parallelism.
• Each CUDA thread processes one independent 64-bit block, enabling full utilization of thousands of GPU cores.
• The implementation demonstrates that legacy symmetric ciphers like 3DES can be efficiently accelerated on modern GPU hardware.
• Performance gains are attributed to fine-grained parallelism and efficient memory access patterns in the kernel design.
• The study confirms that ECB mode is well-suited for GPU acceleration due to its data-independent block processing.
• The results indicate that GPU acceleration is viable for compute-intensive cryptographic workloads, even for older algorithms like 3DES.