About GPUTejas
GPUTejas is a Java-based parallel GPGPU simulator built by the Srishti
research group, IIT Delhi. It is a fast trace
driven simulator, which uses relaxed synchronization, and non-blocking data structures to derive its speedups. Secondly, it
introduces a novel scheduling and partitioning scheme for
parallelizing a GPU simulator.
Getting started with GPUTejas is extremely easy. It is easy to install, and run the simulator. Additionally, GPUTejas has an extremely modular component-oriented design, and it is easy to add new features to it. The current version takes traces from NVIDIA's NVBit tracing engine.
We have evaluated the performance of our simulator with a set of Rodinia benchmarks and have demonstrated a mean speedup of 17.33x with 64 threads over sequential execution, and a speedup of 429X over the widely used simulator GPGPU-Sim. We have validated our timing and simulation model by comparing our results with a native system (NVIDIA Tesla M2070 and the latest Ampere processors).
For Download: click here
For more information: Click Here.
GPUTejas versions
GPUTejas 2.0 - Latest version. Tested on all versions of linux.
New Features in GPUTejas 2.0
- GPU Pipeline (Ampere architecture)
- Realistic streaming multiprocessor (SM) design
- CUDA core enhancements
- Memory hierarchy improvements
- Improved warp scheduling and execution pipelines
- Enhanced Instruction Set Architecture (ISA)
- Advanced NVBit-based trace generator
- Support for SASS traces
- Compatibility with the latest CUDA versions- Tested for CUDA 11
New Features in GPUTejas 1.1
- GPU Pipeline
- Operand Collector based Register File
- DRAM
- Electrical NoC
- BUS
- Ring
- Mesh
- Torus
- Fat Tree
- Butterfly
- Optical NoC
- Power Model - Tested for Pascal Architecture
- Parallel Thread Implementation - 1 Thread for each compute unit
Citing GPUTejas bibtex
Malhotra, G. and Goel, S. and Sarangi, S. R., "GPUTejas: A Parallel Simulator for GPU Architectures", HiPC 2014 [pdf]
