



Study with the several resources on Docsity
Earn points by helping other students or get them with a premium plan
Prepare for your exams
Study with the several resources on Docsity
Earn points to download
Earn points by helping other students or get them with a premium plan
Community
Ask the community for help and clear up your study doubts
Discover the best universities in your country according to Docsity users
Free resources
Download our free guides on studying techniques, anxiety management strategies, and thesis advice from Docsity tutors
A closed-note midterm exam for the computer architecture and engineering course offered by the cs division at the university of california, berkeley. The exam covers topics such as the design of directed mapped, two-way set associative, and fully associative caches, as well as memory system design and block diagrams.
Typology: Exams
1 / 7
This page cannot be seen from the preview
Don't miss anything!
University of California at Berkeley College of Engineering Department of Electrical Engineering and Computer Sciences Computer Science Division
CS 152 J. Wawrzynek Spring 1994 Handout #
Consider the design of three separate caches, a directed mapp ed version (DM), a two-way set asso ciative version (SA), and a fully asso ciative version (FA). The total cache size in data words for each is 1K words and cache blo cks (lines) contain four words of data. Assuming a 16-bit memory address, ll in the table b elow to indicate how many bits of the address to assign to each eld.
For each cache version, illustrate the breakdown of the address into elds. Lab el each eld using the names in parenthesis in from the table ab ove.
(a) [1 p oints] What is the maximum theoretical sp eed-up for a n-stage pip eline pro cessor over a \single-cycle" pro cessor implementation?
(b) [2 p oints] List two reasons why in (a) the sp eedup will actually b e less in practice.
(c) [1 p oints] For a standard MIPS implementation, approximately what p ercentage of branch delay slots would you exp ect could b e lled?
(d) [2 p oints] Data forwarding can solve some data hazards in pro cessor pip elines. Brie y, how do es forwarding e ect the pro cessor cost?
(e) [1 p oints] True or False. A two-bit branch prediction scheme will always miss-predict less than a one-bit scheme.
(f ) [1 p oints] Rank order the three cache organizations according to miss rate.
(g) [1 p oints] Rank order the three cache organizations according to implementation cost.
(h) [2 p oints] In what way do es a system with a write-back cache achieve a p erformance advantage over a write-through cache?
(i) [1 p oints] In what way is a write-back cache more complex than a write-through cache?
(j) [1 p oints] True or False. Because of its simplicity, one-transistor Dynamic RAM cells are usually used for register le implementations.
(k) [3 p oints] What is the approximate data bandwidth out of a frame bu er to feed a 1K X 1K pixel color display device?
(l) [2 p oints] What are three factors that contribute to the time needed to access a blo ck from a magnetic disk?
(m) [1 p oints] True or False. Unlike ethernet, token-ring networks use no arbitration scheme.
(n) [2 p oints] What is the primary advantage of asynchronous bus signalling over syn- chronous?
(o) [1 p oints] True or False. The primary disadvantage of a vector pro cessor over a sup er-pip elin ed pro cessor is that the vector pro cessor requires multiple execution units.
(p) [2 p oints] List two limiters of sup er-scalar pro cessor p erformance.
(q) [2 p oints] List two limiters of sup er-pip eline d pro cessor p erformance.
(r) [1 p oints] Name one commercial SIMD parallel computer system.
(s) [2 p oints] True or False. The theoretically lowest diameter network for multipro cessor is a hyp ercub e top ology.
(t) [1 p oints] True or False. All massively parallel pro cessors use a message passing mo del for inter-pro cessor communication.
(d) Would you exp ect that the average CPI would go up or down with this change to the pip eline? Justify your answer.