1(a) Why do you think that
parallel processing with large number of nodes has been still in infancy as far
as general purpose computing is concerned? (4 Marks)

(b) Give two reasons why Amdahl’s
law should hold? Is it possible that there are cases where the law may fail? If
yes, explain. (6 Marks)

(c) Do you find any machine that
does not fit into Shore’s classification scheme? Explain. (6 Marks)

(d) A processor had no floating
point unit in earlier version but was added to it later. The speedup for
floating point operations is 500 compared to software implementation of
floating point routines. Find the speedup for programs that has floating point
operations in the original machine consuming 10%, 20% and 90% of time. (4
Marks)

2(a) Using a neat sketch explain
electronic implementation of a pipeline. (6 Marks)

(b) Write a note on optimal pipe
segmentation. (6 marks)

(c) Distinguish between fixed
point addition pipeline and floating point add pipeline using neat sketches. (8
Marks)

3(a) Control of a pipeline
instruction processor poses a number of problems. Explain. (6 Marks)

(b) Using a neat figure, explain
the instruction processing pipeline in superscalar processor. (8 Marks)

(c)Given 3-stage multiplier pipe
and 2-stage adder pipe and delays, draw a schematic of a chained pipeline for doing
the following vector computation on vectors A,B,C and Y. Y

_{i}=(A_{i-2 }+ B_{i})^{*}C_{i-3}+ D_{i+1}(6 Marks)
4(a) Give the schematic design of
a scheduler that implements following latency cycle for dynamically configured
pipe running for operations A, B, C and D. Give the number of initiations per
clock (IPC). Latency cycle = <2AD 3ABC 1BD 2AC> (8 Marks)

(b) Write a note on static branch
prediction. (4 Marks)

(c) Prove that the average
latency of any greedy cycle (simple) is less than or equal to the number of 1’s
in the collision vector (d

_{0}is included) of the reservation table. (8 Marks)
5(a) Work out an algorithm to
multiply two 4x4 matrices on a hypercube machine having 16 nodes. (6 Marks)

(b) A data item is to be
distributed to all the PEs. Find out the data routing steps to carry out the
job for hypercube connected machine. (6 Marks)

(c) A ring connected SIMD
parallel computer is to add n numbers. Work out a parallel algorithm to add these
n-node rings. How many routing steps will be required? What is the time
complexity to perform the n number addition on n-node ring machine? (8 Marks)

6(a) Write a note on ILLIAC IV
Computer. (4 Marks)

(b) Draw and explain different
communication ports of PE in a mesh connected computer. (4 Marks)

(c) Assume a 64 node mesh
connected machine having the 64 records of 64 students. It is desired to find
all students having more than 90% marks. Work out the program schematic for the
problem. What is the speedup obtained? (6 Marks)

(d) A two dimensional matrix of 8x8
contains real numbers. It stands for the values of pixels of same image. A
simple smoothing algorithm requires to smoothen out the image so that every
element of the matrix is replaced by the average value of the four neighbors.
Formulate the parallel algorithm and code it for the machine discussed. (6
Marks)

7(a) Illustrate the data routing
from node 6 to 14 in a single stage shuffle exchange network. (6 Marks)

(b) How many steps shall be
required for routing a data item from one node to another in a 32 node Omega
network. Explain. (4 Marks)

(c) Illustrate the routing
algorithm for the 8x8 Benes network with permutation p = [1 3 5 7 6 8 9 0 4 2 11 10] (6 Marks)

(d) Design an 8x8 cross bar using
digital logic elements. (4 Marks)

8(a) What is the effect of low
and high value of τ on SLT network? (4 Marks)

(b) Can 2 input XORs be
implemented using 3 neurons? If yes, show the neural network with its weights. (6
Marks)

(c) Take a 3 PML node 2 input 3
class neural network and illustrate the training algorithm for classification
for 2 classes. (6 Marks)

(d) State and explain Hebb’s
learning (Delta rule). (4 Marks)

## No comments:

## Post a Comment