Lecture 12 Graphics continued

1
Lecture 12 Graphics continued

• In this lecture we will cover
• Graphics card operation
• Speed up
• Bus developments
• Memory
• Graphics co-processors
• Graphical data compression

2
Schematic of simple graphics card
3
Simple graphics card components and operation

• Simple graphics card is responsible for
  converting data that represents image into
  signals that are sent to display device
• Essential components of simple graphics card
  consists of
• 1. A video memory frame buffer to hold the data
  that represents all the pixels for a frame

4

• 2. Three Digital to Analog Converters (DAC) one
  for each primary colour
• DACs convert a discrete numerical value into a
  quantity of something that can vary continuously
  in our case it is a voltage level that will
  produce the appropriate Red/Green/Blue colour
  intensity on the image screen
• 3. Palette registers that hold colour values if a
  palette approach is being used to map imag data
  into RGB colour intensity numbers

5

• 4. Dot clock that synchronises operations on
  board, in particular the rate at which pixel RGB
  intensity values are sent to screen
• Operation
• Video image data is processed by CPU and sent to
  image frame buffer
• RGB numerical values are sent to DAC (possibly
  via Pallette registers) which converts values to
  appropriate voltage levels and sent to display
  device synchronised to dot clock

6
Speed/memory improvements

• As we have seen the data that needs to be
  processed to update a screen display is very
  large so cards improved to deal with this
• Variety of on card improvements
• Size of frame buffer
• Double buffering
• Internal Bus width
• Dual ported and fast memory
• Co-processors
• Also external bus speed from CPU
• Data compression

7

• Size of frame buffer needs to be large enough to
  hold a frame of data as colour depth increased
  from 4 bits to 8, 16 and 24 bits and as
  resolution increased from 320x200 to typical
  modern values of 1024x768, or 1280x1024, etc.,
  then size of memory required grew
• So frame buffer memory increased from 256K to 8MB

8

• With single frame buffer you have to construct
  the frame and then display the frame, or try and
  construct frame while it is being accessed for
  display
• So double buffering uses 2 frame buffers one
  where the image is being output to the screen and
  the other which is being constructed and
  manipulated by the processor
• So memory requirements double 16MB needed in
  place of 8MB to support higher resolution levels

9

• Dot clock frequency requires very high speed
  access to memory conventional RAM memory can
  not keep pace with demand
• Helped reduce frequency of memory accesses by
  increaseing size of internal bus from 16/32
  bits to 256 bits more pixel data per memory
  access
• Use faster RAM technology in particular use of
  dual ported RAM which can be read from and
  written to at the same time

10

• Originally CPU was responsible for performing the
  calculations that are needed to produce pixel
  colour values to represent the image
• The pixel value data is then sent over the
  external bus to the graphics card frame buffers
• This is very time consuming for CPU and CPU has
  other demands on its processing time, so this can
  slow speed with which appropriate pixel colour
  values are sent to graphics cards

11

• Also high volume pixel data needs to be sent over
  external bus to graphics card
• So co-processors (graphics accelerators)
  introduced onto graphics cards which are powerful
  and fast CPUs in their own right that are
  dedicated to performing the pixel colour
  calculations
• Main CPU still has to specify information about
  the nature of the image to be produced but this
  si a high level specification of image rather
  than specific pixel colour values

12

• High level specification of image requires much
  less data to be sent from CPU over external bus
  plus there being no competition for main CPU
  processing - gives improved throughput of image
  information from CPU
• Co-processor performs actual generation of
  geometric shapes, rendering their surfaces, etc.
  to produce the pixel colour values that form the
  image to do this well requires a lot more
  memory e.g. 512MB on card memory
• Typically 2 separate co-processors one for
  handling 2D images and 3D images because the
  demands of processing 2D and 3D images are quite
  different

13

• External bus speed also increased over the years
  from old standard ISA bus that could only moved
  8MB/sec to PCI bus that could move 266MB/sec
• Then AGP (Advanced Graphics Port) a special
  purpose bus dedicated to sending data from main
  CPU unto graphics card so no competition for
  data transfer could move 528MB/sec (AGP1) up to
  2GB/sec AGP4
• PCI Express bus can handle even higher data
  transfer rates up to 8GB/sec

14
Schematic of modern graphics card
15
Data compression

• Another way in which you can improve the through
  put of images in the system is to reduce the data
  transfer and memory requirements for given image
  output by reducing the amount of data required to
  send/store an image
• Data compression are algorithms and techniques by
  which data size is reduced while still retaining
  an acceptable level of quality

16

• 2 types of compression - lossless and lossy
• Lossless is where compression reduces size of
  data file but original values can be fully
  recovered when data is decompressed
• Lossy is where compression reduces size of data
  but at expense of some degradation of quality
  when image is decompressed

17
Run Length Encoding

• Simple bitmapped image is a simple set of array
  of pixel values no compression
• Simple example of compression Run Length
  Encoding (RLE)
• In an image often the case that pixel values that
  lie next to each other have the same colour
  hence the same data value would represent that
  colour

18

• So we often have sequences of pixel values that
  are the same the number of pixels in a sequence
  is the run length,
• RLE uses a key byte
• If top bit is 1 then 7 remaining bits of byte
  gives number of next bytes which do not repeat
• If top bit is 0, then remaining 7 bits of byte
  gives number of next bytes are the same so
  replaces them with just one byte that has
  repeated value

19

• Other compression schemes
• jpeg compression
• mpeg compression
• jpeg divides image into small blocks e.g. 8x8
  array
• It applies a transform called Discrete Cosine
  Transform (DCT) to the pixels in that block to
  produce a representation in which the block is
  seen as being composed of a combination of a
  variety of components (frequency components
  actually)
• depending upon degree of compression required
  some of components thrown away and some accuracy
  is lost

20

• mpeg uses an approach similar to RLE
• from one frame to next, in a movie sequence a
  large number of pixels are the same
• so mpeg completely compresses some frames using
  DCT but for intermediate frames it only encodes
  changes from the fully encoded frame
• it encodes just information about those pixels
  that have changed