we are asked relative performance , but

D=1.2tcpi

D1=1.32tcpi

D2=.92tcpi

performance we won,t chech throuch CPI , if cpi is greater it means that processor is slow..

please correct me in case of wrong.

Dark Mode

4,693 views

28 votes

The floating point unit of a processor using a design $D$ takes $2t$ cycles compared to $t$ cycles taken by the fixed point unit. There are two more design suggestions $D_1$ and $D_2$. $D_1$ uses $30\%$ more cycles for fixed point unit but $30\%$ less cycles for floating point unit as compared to design $D$. $D_2$ uses $40\%$ less cycles for fixed point unit but $10\%$ more cycles for floating point unit as compared to design $D$. For a given program which has $80\%$ fixed point operations and $20\%$ floating point operations, which of the following ordering reflects the relative performances of three designs?

($D_i > D_j$ denotes that $D_i$ is faster than $D_j$)

- $D_1 > D > D_2$
- $D_2 > D > D_1$
- $D > D_2 > D_1$
- $D > D_1 > D_2$

–1

4

1

50 votes

Best answer

(B) is correct.

$T = 0.8 \times \text{time taken in fixed point} + 0.2 \times \text{time taken in floating point}$

$D = 0.8 \times t + 0.2 \times 2t = 1.2t$

$D_1 = 0.8\times 1.3 t + 0.2 \times 0.7 \times 2t = 1.04 t + .28t = 1.32t$

$D_2 = 0.8 \times 0.6 t + 0.2 \times 1.1 \times 2t = 0.48t + .44t = 0.92t$

So, $D_2$ is the best design for this given program followed by $D$ and then $D_1$. Option B.

1

usman11 No cycle and time are directly proportional (obviouslt cause time can be measured in multiple of cycles)

0

10 votes

Suppose t=1ns, there are total 100 operations, 80 fixed point operations and 20 floating point operations,

Time taken by design D = 80*1ns + 20*2ns = 80+40 =$120 ns$

Time taken by design D1 = 1.3*80 + 0.7*40 = 104 + 28 = $132 ns$

Time taken by design D2 = 0.6*80 + 1.1*40 = 48 + 44 = $92 ns$

Now, we can see that D2 is fastest, D is faster than D1.

**Hence (B) is correct option D2 > D > D1**