Lesson 14.2: Data-Level Parallelism

0.0(0)
Studied by 0 people
call kaiCall Kai
learnLearn
examPractice Test
spaced repetitionSpaced Repetition
heart puzzleMatch
flashcardsFlashcards
GameKnowt Play
Card Sorting

1/33

encourage image

There's no tags or description

Looks like no tags are added yet.

34 Terms

1
New cards

Single thread

CPU design is about making a _ _ run as fast as possible

2
New cards

Latency

pipeline stalls and memory accesses are expensive in terms of _

3
New cards

Large cache memories

use of _ _ _ to avoid memory misses

4
New cards

Computation Throughput

GPU design is about maximizing _ _

5
New cards

Extracting ILP

GPUs avoid much of the complex CPU pipeline logic for _ _

6
New cards

Parallelism

each thread executes on a relatively simple core with performance obtained through _

7
New cards

Memory and Pipeline Latencies

computations hides _ _ _ _

8
New cards

Bandwidth optimized memory systems

wide and fast _ _ _ _

9
New cards

Single Instruction, Multiple Threads

meaning of SIMT

10
New cards

SIMT

takes advantage of data-level parallelism (1)

11
New cards

Multithreading

SIMT, can be considered a constrained form of _

12
New cards

Operating on scalar register

SIMT, many threads each with their own state (1)

13
New cards

With their own local memory

SIMT, many threads each with their own state (2)

14
New cards

Single Instruction, Multiple Data

meaning of SIMD

15
New cards

SIMD

takes advantage of data-level parallelism (2)

16
New cards

Vector Processing

SIMD, can be considered a constrained form of _ _

17
New cards

One thread

SIMD; _ _ operating on vector registers

18
New cards

One processing element

a single thread runs on a simple processing element

19
New cards

One processing element

short pipeline

20
New cards

Thread state

the execution context consists of the _ _

21
New cards

Fetch and Decode

but _ _ _ are costly

22
New cards

Multiple Processing element

SIMT execution of threads

23
New cards

Multiple processing element

cost of fetch and decode spread across all threads in a work group

24
New cards

Cheap unit

one per thread for _ _, multiple functional units for parallelism (1)

25
New cards

Fewer expensive

_ _ units than threads, multiple functional units for parallelism (2)

26
New cards

Stalls

however, _ are costly

27
New cards

Minimizing Stalls

include support for multiple work groups

28
New cards

Independent

minimizing stalls, each is _ of all others

29
New cards

Scheduler

a _ chooses work groups to run

30
New cards

Ready Work Groups

minimizing stalls, scheduler, maintains a list of _ _ _

31
New cards

Shader Core

minimizing stalls is sometimes called a

32
New cards

Scaling out

multiple instances of each shader core provided together

33
New cards

Scaling out

massively increases parallelism

34
New cards

Scaling out

memory hierarchy provided, too