2797337
Description
Flashcards by alanischelsea, updated more than 1 year ago
|
Created by alanischelsea
almost 9 years ago
|
|
| Question | Answer |
| conception about bigger space before? | more advisable to have bigger space before because that meant you can put more. but that concept was changed. |
| What is parallel programming? | small weak computers can be configured with processors/special software that would improve their performance; they are grped together as clusters/grid; massively parallel system |
| types of multiprocessors? | mainframe and server |
| mainframe? | perform concurrent simple tasks |
| supercomputer? | aimed to solve few computationally intensive tasks |
| what make supercomp and mainframe similar? | both have many processors. They can do what each other can do provided that they are given the same software. Their performance depends on the software installed to them. |
| Changes in technology | Performance- must double every 1.5 years Cost- depreciation; people are smarter now in manufacturing; "learning curve" Function- ie. phone that can be a camera, video recorder blabla |
| perception on size of a device | - size as dictate of design not performance - small/large device can be weak/powerful - performance not based on size anymore |
| graphical representation of the relationship between learning and experience | learning curve |
| Moore's Law | performance doubles every 1.5 years |
| Why does Moore's Law's meaning never change? | before, the number of transistors are parallel to the performance. |
| What is a transistor? | a microscopic switch with on/off states; allow/not allow electricity to pass through; more transistors mean more ability to answer |
| What does a computer system do? | it breaks down complex questions into simple ones that can be answered by yes/no (transistor's task) ; system is not smart, but fast because you have to explicitly state its course of decisions. Unlike for man, you can come into conclusion w/o explicitly saying the steps you took to arrive at such conclusion. |
| Why does tech improve exponentially? | because of other technology; the improvement of tech is brought about by an existing one; ex. iphone from ipod + phone; not really new in terms of products, but new in the sense of procedures |
| transistor density? die size? disk? | - how many transistors put together - square that transforms itself into CPU - optimal size that we can handle |
| task of computer architecture & factors to consider when recommending | - be able to recommend computer/needs - Application: saan gagamitin yung computer? - History: what worked/didn't work in the company - Prog Language- C, C#, Java ba? Technology- web-based ba or......... - budget!!!! - maximize performance |
| yield? | percentage of good products over total products; increase in yield means increase in LC which affects the cost |
| formula for cost of die | = wafer cost / (dies per wafer * die yield) |
| IC cost formula | (die cost + packaging cost + testing cost) / (final test yield) |
| cost when finding out the good/bad die wafers | testing cost |
| 2000 die wafers, 50% die yield, 25% IC yield answer? | = 2000 * 0.5 * 0.25 = 250 IC |
| die in the beginning.............. | has no circuit then the good ones are transformed into IC |
| direct cost? gross margin? ave discount? | - coming together of components; labor - aka indirect cost; R&D, advertising cost - profit; ito yung range na binabawasan ng retailers |
| time it takes to finish one task | latency/execution time/response time |
| how many tasks that can be performed at a given time | throughput/bandwidth; measured as rate |
| execution time & performance?? | inversely proportional - if lower exec time, better performance |
| Amdahl's Law | improving 1 computer means improving the part/component that you use the most in order to feel the change |
| Fraction enhanced and Speedup enhanced | - how often is the component used? ; should be less than or equal to 1 (percentage) - how much faster is the new improvement? ; greater than 1; if equal to 1, then there's no improvement at all |
| formula of amdahl's law | extime old / extime new = 1 / ((1-fe) + (fe/se)) - no unit; "___ times better than old performance" |
| CPU clock | - have a particular cycle that dictates computer operations - comp operations are implemented in accordance with the rate of the cycle |
| What is a cycle? | the interval between the start of the burst and the start of the next burst ; electricity is present during burst; measured by Hertz; 3 cycles = 3 Hertz |
| CPU and floating numbers | - floating number: ridiculous number of decimal pts - CPU that can compute floating pt operations is more ideal/more precise - such performance is significant when computing pixels for gaming |
| Locality of reference | - CPU: the brain - main memory or RAM: primary memory; what we upgrade is really the RAM - secondary memory: large memory space that is relatively slow; hard drive; storage |
| interaction among CPU, RAM and HD | Analogy - the secondary memory is like a cabinet with all files; RAM is the study table that contains address; CPU gets the data from memory then works on the RAM or respective address - highest priority is given to the data you use now |
| temporal locality? spatial locality? | - across time, there's data with high probability of referencing; for example, alam na ng waiter yung i-oorder mo kung yun yung lagi mong inoorder - more on space; high probability of referencing of data that are juxtaposed with each other |
| - what is a benchmark? - types of benchmark? | - benchmark is the act of running a computer program, a set of programs, or other operations, in order to assess the relative performance of an object, normally by running a number of standard tests and trials against it. --- types--- - synthetic: fake program, can be cheated, artificially changing results to match with benchmarks - Toy: literally toy; hindi gaanong ginagamit - kernel: "isolated performance"; come from real program; real prog into pieces |
| Computer hardware - what are the parts of the computer? | - processor: CU + datapath - Memory: Main memory / secondary memory - Input/Output |
| what is a control unit? what is a datapath? | CPU- has control unit inside control unit- issues signals to the rest of the CPU - datapath is a collection of functional units, as arithmetic logic units or multipliers, that perform data processing operations, registers, and buses. - all other parts won't function unless given a signal by the CU |
| Von Neumann Machine vs Harvard Archi | VN - applied the concept of alan turing; programs are executed sequentially; parameters can be changed; bottleneck H.A.- made 4 pathways for data and instruction & their respective addresses; solved the bottleneck; can be accessed simultaneously http://v-codes.blogspot.com/2010/09/difference-between-harward-and-von.html |
| What is the memory hierarchy | - registry is the fastest followed by cache, memory, secondary memory - access time is the time it takes to locate the data and give data - smaller size means better performance kasi mas konti yung space na hahanapan - tapes are the slowest because it's sequential; longer process |
| - boss inside the CPU - fed with instructions - without its signals, other parts of CPU will not function | CONTROL UNIT |
| parts of this have specific roles that are non-flexible; one primary example is register; general purpose- can hold general data and special purpose- limited to data specified to it (adobo analogy) | DATAPATH |
| higher vs lower programming language | HIGH LEVEL PL - closer representation to the user; compiler tells RAM to make a space for 'x';various addresses contain x's; compiler binds value of x in the RAM's address - garbage collection in higher level PL - higher PL = more generic - ease of implementation advantage LOW LEVEL PL - no garbage collection; explicitly states the need to have one - closer to the machine; ex: C prog lang - greater hardware detail = low level - therefore, computer archi is more on low level - control address of the hardware advantage - more connected to hardware so you can specify the address |
| it performs the mapping - sets the representation - allows the operation to happen | instruction set architecture |
| instruction cycle | Fetch instruction Decode instruction Operand Fetch Execution Store- store in the left operand |
| memory address register memory data register instruction register | - contains the address to which the data is fetched from - temporarily stores the data after it was fetched - where the instructions permanently resides |
| signal issued by CU to fetch the instruction | MEM read |
| difference between risc and cisc | complex instruction set computers - new processors are more powerful over older models to maintain compatibility - use of microcode - richer/more complex instruction sets simplify compiler design - more powerful instruction means fewer instructions needed - the more complex, the bigger the required control store is Reduced ISC - perform simpler ops and use simple addressing modes - complex ops can be done by using several simpler ops - 80/20 rule: if most used instructions was sped up, then the performance would be greater - basic hardware ; easier to compile |
| set of machine level instructions for a particular type of computer or hardware | instruction set |
| the sequence of ops involved in processing an instruction | instruction cycle |
| instruction cycle..... | fetch instruction decode instruction operand fetch execution store |
| this happens when the memory informs the cpu that it has completed read/write request; memory sends mfc - memory function complete signal - situation in which memory & cpu are synch | cpu-memory handshake |
| types of control | - hardwired - microprogrammed |
| considerations to hardwired control | - fast: everything done thru hardware - permanent: difficult to modify - difficult to implement in large instruction set |
| microprogrammed consideration | - flexible because u can just modify the control store if needed - slower and may require more hardware due to increasing number of instructions |
| considerations on microprogrammed | - flexible because u can just modify the control store if needed - slower and may require more hardware due to increasing number of instructions |
| areas of devt for cpu design | system bus speed internal and external clock freq casing cooling system instruction set wire material |
| what's the end result of these areas of devt in cpu design? | enhance the speed of the cpu and system in general |
| conduit for moving data between the processor and other sys components | system bus main memory --> mar MM --> mdr MM --> control |
| speed of data processing inside the cpu | internal clock freq |
| speed of data transfer to and from the cpu via the system bus | external clock freq |
| faster? internal or external? | internal!!!!!!!!!!!!!!! external * multiplier = internal |
| how long will moore's law last? | 40 years!!! |
| what's the purpose of micron tech? what is it in the first place? | - 1 millionth of a meter - objective is to have thinner wires - allow cpu to operate at lower voltage - generating less heat and operating at higher speeds |
| material used for wire | copper - from MHz to GHz real quick |
| what about this system on chip? | - many components bundled in one chip this results to replacing dozen or separate chips |
| impact of system on a chip | - longer battery life - less heat dissipation - smaller devices |
| fc-lga | - for casing of the IC - lower voltage used and less heat dissipation - the pins are found in the socket and not in the chip itself |
| going beyond the recommended clock frequency; done by - system bus freq - cpu freq multiplier - change both | overclocking |
| exceeds the normal speeds depending on processing reqs | dynamic overclocking |
| cooling sys | make sure that cpu won't overheat as it tends to get hotter as it gets faster |
| multimedia processing | - gpu - cpu that can compute floating pts operations faster is more ideal - kung mabilis yung pixel calculation, mas okay daw for gaming |
| how to handle multimedia | - speed up cpu - use high-end 3d graphics cards - pipeline implementation, multithreading (multiple processes at the same time) - multimedia specific instructions |
| single-cpu innovation | - internal: how many bits can the cpu process simultaneously? - external: how many bits can the cpu receive simultaneously for processing? |
| superscalar | Superscalar processor executes multiple independent instructions in parallel. • Common instructions (arithmetic, load/store etc) can be initiated simultaneously and executed independently |
| current innovation today | multicore multithreading |
| intel cpu lineup | pentium celeron atom core i3 5 7 xeon itanium |
| amd cpu lineup | desktop - amd fx - phenom - athlon - sempron tablet - a series server - opteron |
| combining several components in a single chip | accelerated processing unit cpu gpu memory controller video encoder/decoder |
| intel cpu innovation | - hyperthreading - multicore - quickpath interconnect |
| amd cpu innovation advanced micro devices | hypertransport multicore direct connect archi |
| POWER7+ 8 cores 4.42 Ghz 4 threads per core 32 nm process introduced TrueNorth- prototype chip that mimics 1 million human neurons and 256 million synapses | ibm |
| SPARC M6 Processor -12 cores 8 threads per core 3.6 Ghz 48 MB L3 cache 28 nm process | ORACLE |
Want to create your own Flashcards for free with GoConqr? Learn more.