Academic Integrity: tutoring, explanations, and feedback — we don’t complete graded work or submit on a student’s behalf.

1. [20 marks] We investigate the increase in CPI (clocks per instruction) due to

ID: 3806395 • Letter: 1

Question

 1. [20 marks] We investigate the increase in CPI (clocks per instruction) due to cache misses that occur during memory references.  For simplicity, we pretend that instruction fetches never miss.  a) [10 marks] Suppose the processor takes an average of 1.5 clock cycles to execute an instruction when there are no cache misses.  Assume that the miss penalty is 8 cycles and that there is an average of 1 memory reference per 3 instructions.  This base CPI of 1.5 cycles includes the cache hit time. Suppose the miss rate is 5%.  Using the formula t_ave = ht + mr * mp, what is the CPI when cache misses are taken into account?  b) [10 marks] Consider the same processor with a two-level cache.  The hit rates for the L1$ and the L2$ are 95% and 80%, respectively.  The local miss penalties are 8 cycles and 60 cycles, respectively.  Assume the same density of memory references.  If the CPI is 1.5 cycles when there are no cache misses, what is the CPI when cache misses are taken into account? Hint: Apply the formula recursively to find the effective miss penalty of the L1$. 

Explanation / Answer

CPI=base CPI+ Icache

base CPI=1.5

Icache=miss rate*miss penalty

Icache=0.05*8

0.4

CPI= 1.5+0.4

=1.9 cycles per instruction

2) CPI= CPI+L1 miss rate*L1 miss penalty+ L2 miss rate* L2 miss penalty

L1 miss rate=(100-95)=5%

L1 miss penalty=8

L2 miss rate=(100-80)=20

L2 miss penalty=60

CPI=1.5+(.05*8)+(.20*60)

cpi=13.9 cycles per instruction