You are on page 1of 12

Presented by 1)Y.

Navanaeeth Krishnan 1st sem CSE 2)Vipin Dwyedi 1st sem CSE

The reach of Information technology is very wide. From a hand held blackberry to a datacentre .Every element provides us an element to change .The effects of this change may be in the near future or in a couple of years IT touches everyone's life today .Not even a single day passes without us relying on a computer. It is estimated that if the amount of power consumed per person for his IT needs is reduced by 15% it is equivalent to removing one car from the road

In a data center the computers, storage devices and air conditioners work in an inefficient manner Many server computers run at less than 30% of their server capacity Usually algorithms are run faster in a computer on the cost of cache memory. The umpteen number of registers and cache memories in a computer cause it to consume a lot of power .This not only causes a large amount of heat to be produced it increases the resistance of the electrical pathways leading to the processor

CPU

GPU

CPUs have a lower clock speed when compared to GPUs. CPUs require 100% usage of processing powers to do a computationally intensive calculation.Gpu remains idle during the execution of this calculation. Therefore the power consumed by the computer = Power consumed by CPU+Power consumed by GPU

GPUs can never be shut off or put in low power mode while the Monitor is on ,but the CPU can be put into low power state even when the monitor is on. The GPU has a processing speed greater than that of a CPU because it executes a program in parallel. Therefore computationally intensive parts carried out on the GPU takes less time as well as energy . Since the GPU has a very high processing capacity there will be no change in its usage and IT WILL STILL REMAIN IN THE IDLE MODE during the processing of a computationally intensive program

We have implemented the running of a computationally intensive application (multiplication of 2 full matrices of 1024 order ) in OpenCl. We have used the GPU to perform the computationally intensive part while the sequential part is done by the CPU. Therefore by running the computationally intensive parts on GPUs the CPU is in a low power mode because its cache memory and registers are not used at all .Thus saving power .

2016 MATRIX MULTIPLICATION IN CPU

2016 MATRIX MULTIPLICATION IN GPU

DEVICE

MATRIX A

MATRIX B

TIME(S)

Kernel time (S)

CPU GPU CPU GPU

1024 1024 2016 2016

1024 1024 2016 2016

2.220 1.529 6.804 2.036

1.908 0.413 6.514 0.639

GPUs cant perform sequential operations like input and output like a cpu . GPUs generate a lot of heat GPUs consume more power when both graphics and calculations are required like games.

You might also like