Deck 19: General-Purpose Graphic Processing Units

Question 1

All but one set of GPU processor cores will be idle,while one SM is bearing the full processing load.

Answer 1

True

Question 2

The total number of threads defined is typically in the thousands to maximize the utilization of the GPU processor cores as well as maximize the available speedup.

Answer 2

True

Question 3

Because the GPU and the CPU are designed and optimized for two significantly different types of applications,their architectures differ significantly.

Answer 3

True

Question 4

GPGPU is a computing platform and programming model created by NVIDIA.

Question 5

The GPU is most efficient when it is processing as many warps as possible to keep the CUDA cores maximally utilized.

Question 6

For about $200 you can purchase a GPU with 960 parallel processor cores for a workstation.

Question 7

The Fermi architecture upgraded from the IEEE 754-1985 floating-point arithmetic standard to the IEEE 754-2008 standard.

Question 8

In embedded systems the GPU is composed of only a single-digit number of cores,and are typically combined with a number of conventional cores,referred to as _________.

A)arithmetic logic units
B)control units
C)central processing units
D)graphic processing units

Question 9

It is not important for the programmer to understand the nuances of the various GPU memories.

Question 10

The largest GPUs are found in embedded systems.

Question 11

The equivalent GPU hardware component for a block is the CUDA multiprocessor (SM).

Question 12

CUDA was created by __________ .

A)Amdahl
B)NVIDIA
C)the U.S.Government
D)Herbert Moore

Question 13

GPUs can be found in almost all of today's workstations,laptops, tablets,and smartphones.

Question 14

A kernel typically will have few to no branching statements.

Question 15

A SoC product architect can create product families or a specific product within a family by placing a single slice or multiple slices on a SoC chip.

Question 16

The __________ is designed specifically to be optimized for fast three-dimensional (3D)graphics rendering and video processing.

A)CPU
B)GPU
C)CU
D)ALU

Question 17

A group of threads assigned to a particular SM is a __________ .

A)block
B)grid
C)unit
D)kernel

Question 18

An instance of the kernel on the GPU is a ___________ .

A)thread
B)warp
C)grid
D)block

Question 19

The grid and the block need to have the same dimensions.

Question 20

CUDA C is a C / C++ based language.

Question 21

The data-parallel code to be run on the GPU is called a ___________ .

Question 22

The entire Gen8 compute architecture interfaces to the rest of the SoC components via a dedicated unit called the ____________ .

Question 23

A __________ program can be divided into three general sections: code to be run on the device,code to be run on the host,and the code related to the transfer of data between the host and the device.

Question 24

Threads are uniformly bundled in _________ .

Question 25

The dual warp scheduler will break up each thread block it is processing into _______ .

A)kernels
B)warps
C)grids
D)all of the above

Question 26

The GPU has found its way into massively parallel programming environments for a wide range of applications,which is where the term __________ is derived from.

Question 27

A _________ is a bundle of 32 threads that start at the same starting address and their thread IDs are consecutive.

A)warp
B)grid
C)block
D)grouping

Question 28

The first NVIDIA GPU with added GPGPU support hardware was the _________ .

Question 29

The parallel code in the form of a function to be run on GPU is the ________ .

A)grid
B)thread
C)kernel
D)none of the above

Question 30

The _________ performs transcendental operations,such as cosine,sine,reciprocal,and square root,in a single clock cycle.

A)SM
B)SIMD
C)SFU
D)FMA

Question 31

To enhance performance,a technique known as __________ is used for the shared L3 data cache.

A)cache banking
B)thread blocking
C)streaming
D)warping

Question 32

A subslice includes a unit called the _________,which is used for sampling texture and image surfaces.

A)stride
B)sampler
C)EU
D)floating-point

Question 33

A _________ is a single instance of the kernel function.

Question 34

In 2006 NVIDIA facilitated the use of its new GPGPU language,________ .

A)GPU / GP
B)SIMD
C)CUDA
D)NVIDIA C

Question 35

The EU can issue up to ________ different instructions simultaneously from different threads.

A)four
B)five
C)six
D)seven

Question 36

__________ is a parallel computing platform and programming model created by NVIDIA and implemented by the GPUs that they produce.

Question 37

In the CPU the control logic and __________ make up the majority of the CPU's real estate.

Question 38

The number of blocks per kernel launch is called a __________ .

Question 39

__________ are caused by limited SFUs,double-precision multiplication,and branching.

A)Structural hazards
B)RAW data hazards
C)Vertical hazards
D)Latency hazards

Question 40

___________ is a GPU processing technology.

A)Fermi
B)Kepler
C)Maxwell
D)All of the above

Question 41

The _________ GPU has a total of 16 SMs x 32 CUDA cores / SM,or 512 CUDA cores.

Question 42

A GPU uses a massively parallel ________ architecture to perform mainly mathematical operations.

Question 43

The _________ scheduler breaks up each thread block it is processing into warps.

Question 44

The _________ global scheduler unit on the GPU chip distributes the thread blocks to the SMs.

Question 45

The fundamental building block of the Gen8 architecture is the ________ unit.