I am planning to work on Multi GPU system based on CUDA. But I only have single GPU machine. So is there any way to simulate GPU cluster to get results? Also suggest if there is any way that i can combine normal laptops with CUDA GPUs to create cluster/
Dear Parth Shah, You can use GPUs with MATLAB through Parallel Computing Toolbox, which supports: CUDA-enabled NVIDIA GPUs with compute capability 2.0 or higher. For releases 14a and earlier, compute capability 1.3 is sufficient.
GPU use directly from MATLABGPU-enabled MATLAB functions such as fft, filter, and several linear algebra operations
GPU-enabled functions in toolboxes: Image Processing Toolbox, Communications System Toolbox, Statistics and Machine Learning Toolbox, Neural Network Toolbox, Phased Array Systems Toolbox, and Signal Processing Toolbox (Learn more about GPU support for signal processing algorithms)
CUDA kernel integration in MATLAB applications, using only a single line of MATLAB code. Multiple GPUs on the desktop and computer clusters using MATLAB workers in Parallel Computing Toolbox and MATLAB Distributed Computing Server.
Christo Ananth I am already running my application using GPU but now i want to check GPU performance in case of multi GPU but i don't have multiple GPUs so i want simulator where i can prove effect of multiple GPUs using simulation.
I think it would be such a hassle to get a simulator to work, it wouldn't judge performance correctly, and would probably cost something anyways. Your best bet is probably to pick up a cheaper GPU.
Chris Rackauckas I can get 3-4 basic gpu enable machine, how can i combine them to create cluster?
In order to run in a cluster, you have to check if the simulator has support for a distributed computing paradigm like MPI, and calculate the overhead according the amount of communication between the different threads.
You could use profiler tools like nvprof, to guess the performance and scalability of the problem.
Although if the problem is not suitable for a distributed paradigm, running in a cluster will slow it down due the communication overhead. For multi-GPU solutions is more common to run in a single machine with several GPUs
- Badges
- Science topic
- Similar topics
- Engineering
- Computer Engineering