Back to Study Material GATE Exam

Last Minute Notes Computer Organization

For last-minute COA revision, start with number representation because it underpins nearly every other topic. Signed integers use two's complement in virtu...

Last Minute Notes Computer Organization

Quick Revision: Number Systems and ALU Operations

For last-minute COA revision, start with number representation because it underpins nearly every other topic. Signed integers use two's complement in virtually all modern architectures. To negate a number in two's complement, flip all bits and add one. This eliminates the dual representation of zero present in sign-magnitude and simplifies hardware subtraction since A minus B equals A plus the two's complement of B.

IEEE 754 single-precision floating point uses 1 sign bit, 8 exponent bits with a bias of 127, and 23 mantissa bits with an implicit leading 1 for normalized numbers. The exponent field value of all zeros denotes special cases like zero or subnormal numbers; all ones denotes infinity or NaN. GATE has tested denormalized number representation, so know that denormals have an exponent field of zero and no implicit leading 1.

The ALU performs arithmetic and logic operations. Carry lookahead adders reduce addition latency by computing carry bits in parallel rather than rippling through each stage. A four-bit carry lookahead adder reduces worst-case carry delay from four gate delays (ripple carry) to two gate delays. Know the generate (G = A AND B) and propagate (P = A XOR B) signals: Ci plus 1 equals Gi plus Pi times Ci.

Booth's algorithm performs two's complement multiplication efficiently by encoding runs of ones in the multiplier as a subtraction at the start and an addition at the end of the run. For the sequence 0111, instead of three additions you do one addition and one subtraction, saving cycles. GATE may ask you to trace Booth's algorithm for a small example.

Restoring and non-restoring division algorithms are less commonly tested but appear occasionally. In restoring division, if a trial subtraction produces a negative partial remainder, you restore the original value by adding the divisor back before shifting. In non-restoring, you skip the restore step and adjust in the next iteration instead.

Quick Revision: Cache, Memory, and Bus Concepts

Average memory access time (AMAT) is the single most important cache formula for last-minute review. AMAT equals hit time plus miss rate times miss penalty. For a two-level hierarchy, AMAT equals L1 hit time plus L1 miss rate times (L2 hit time plus L2 miss rate times main memory access time). Ensure you can extend this to three levels if needed.

Cache replacement policies determine which line is evicted on a miss when the set is full. LRU (Least Recently Used) evicts the line not accessed for the longest time — optimal in practice but requires tracking access order. FIFO evicts the oldest-loaded line regardless of usage. Random replacement is simple to implement in hardware. GATE typically asks which policy performs best for a given access sequence.

Valid and dirty bits accompany each cache line. The valid bit indicates whether the line contains meaningful data (cleared at startup or on flush). The dirty bit (used with write-back) indicates the line has been modified and must be written to main memory before eviction. A line that is valid but not dirty can be silently evicted. A dirty line must trigger a write-back, adding memory traffic.

Bus protocols govern how devices share the communication channel. Synchronous buses use a shared clock; every operation takes a fixed number of clock cycles. Asynchronous buses use handshake signals (request and acknowledge); each device operates at its own speed. GATE questions may ask you to compute bus bandwidth given bus width, clock speed, and number of overhead cycles per transfer.

Virtual memory maps logical addresses to physical addresses through a page table. The page number is the higher-order bits of the logical address and indexes into the page table to find the frame number. The frame number concatenated with the page offset forms the physical address. A TLB caches recent page-table lookups. Effective memory access time with TLB equals TLB hit rate times (TLB access time plus memory access time) plus TLB miss rate times (TLB access time plus two memory accesses for page table and data).

Quick Revision: I/O, Interrupts, and Key Formulas

For the final hours before GATE, consolidate all I/O, interrupt, and performance formulas on a single mental checklist. This section gives you a dense, exam-ready summary of the most tested points.

I/O transfer rate for DMA: if a disk transfers data at rate R bytes per second and DMA is used with block size B bytes, the number of interrupts per second is R divided by B. CPU overhead per second equals (R divided by B) times ISR time. For programmed I/O, CPU overhead per second is R times poll time (one poll per byte or word depending on granularity).

Interrupt latency is the time from when an interrupt is raised to when the ISR begins executing. It includes the time to finish the current instruction, save the program counter and status register, and fetch the ISR address from the interrupt vector. GATE may ask you to identify which of these steps contributes most to latency.

For pipeline speedup with hazards, remember: effective CPI equals 1 plus sum over all hazard types of (fraction of instructions causing that hazard times stall cycles per hazard). Speedup of pipelined over non-pipelined equals non-pipelined CPI divided by pipelined effective CPI, but you must also account for the clock cycle difference if the pipeline changes the clock period.

Amdahl's Law applies when only a fraction f of the program can be sped up by factor S: overall speedup equals 1 divided by ((1 minus f) plus (f divided by S)). GATE uses this to test whether students understand why speeding up a small fraction of a program yields diminishing returns.

Key mnemonics: RISC has more registers, fewer addressing modes, fixed instruction length, and relies on compiler optimization. CISC has fewer registers, more addressing modes, variable instruction length, and complex microcode. Direct-mapped cache: one location per block. Fully associative: any location. Set-associative: any location within a set. Write-through: update memory on every write. Write-back: update memory on eviction of a dirty line only.