S - U

S

• S3TC
  An acronym for S3 Texture Compression. This is S3's trademarked term for their texture compression technique. See DirectX Texture Compression.
  
• Sampling (Audio)
  The conversion of an analogue signal to a stream of digital values (or back again) requires the signal to be analysed, called sampling, at fixed intervals. The more sampling that is done over a given amount of time (sampling rate), the more accurate the digital form of the signal.
  
• Sampling (Graphics)
  This term is used to describe the process of looking up texel values from a texture map, or pixel values from a frame buffer.
  
• Scanline Rendering
  This is the general method of rendering a 3D scene in desktop graphics adapter and games. It works around the principle that the camera views the scene through a "window" (viewport) and calculates what the objects will look like by using an imaginary line from the viewer, through each pixel in the viewport, into the scene. The technique then proceeds one pixel at a time, going across the viewport in lines (scanlines) until the scene is complete.
  Scanline rendering is really only feasible on a desktop PC thanks to the z-buffer; if there wasn't some way of deciding what objects are visible in the scene, the application would be forced to draw every pixel in every object! As it is, the choice of ordering used to render the scene can still have a large impact on the performance.
  
• Sector
  A sector is a sub-division of a hard disc track. Filing systems such as NTFS (Windows NT, 2000 and XP) take a single sector as being the smallest possible unit of storage on the medium, whereas FAT systems (Windows 9x) use groups of sectors called clusters.
  
• Serial
  A term used to describe processes that run consecutively, i.e. one after another.
  
• Serial AGP
  This is the old name for a 16 channel PCI Express interface; so named as it was obviously designed to replace the AGP architecture.
  
• SGRAM
  Synchronous Graphics RAM. This is a slightly modified version of SDRAM in that it allows multiple memory operations to be performed in one go and it allows larger blocks of data to read/write in one operation. It was briefly popular on graphics cards, replacing the likes of EDO RAM and WRAM, but the cheaper and mass-produced SDRAM pushed it out of favour.
  
• SGSS
  Skewed Grid Supersampling. This type of supersample anti-aliasing uses a sampling kernel (the grid or pattern of the sample points around the pixel being tested) that has the sample points shifted around; this helps to increase the appearance of lines at certain angles. Skewed grids are often confused with rotated grids because they look similar.
  
• Shader
  A term used in many different ways but in DirectX, it refers to a set of small pieces of instruction code that are sent to the graphics adapter for it to perform vertex or pixel operations. Vertex shaders apply in the vertex transform and lighting stage of the pipeline, whereas pixel shaders apply during the texture mapping phase.
  
• Shading
  This is a more general name for the lighting stage within the 3D graphics rendering pipeline - it is not to be confused with shaders. To accurately model how an object appears in a 3D scene, one must account for where the light sources are in the scene, what type of material the object is constructed from and which the camera is facing.
  There are many shading procedures, each one suited to a particular type of lighting, but the 3 of the most well known are:
  • Flat Shading: The object has the same light intensity all over it. The process calculates this intensity per vertex and then applies it to the whole polygon. Although unrealistic, the process is very quick.
  • Gouraud Shading: This lighting process takes into account the direction of the light source and viewer, but assumes that the object has a dull or rough surface. This method is used to calculate the light intensity at each vertex of a polygon and then interpolate it across the rest of the surface, allowing for smooth colour gradients. This process is obviously more computationally intensive than the ambient lighting model, and is therefore slower. However, when applied to objects with lots of polygons, it can look very realistic.
    Note that Gouraud Shading can be used in conjunction with other techniques to produce specular lighting, such as using lightmaps.
    
  • Phong Shading: This shading method produces the best looking results but at the highest cost to the rendering performance. Accounting for everything (surface material, position of surface to the light source and camera, etc), it also calculates the colour value per pixel and not per vertex. DirectX uses a simplified version of the Phong lighting model to produce specular highlights.
  The total lighting effects of any model are worked out by accumulating the light properties of the object; in other words, final light = ambient + diffuse + specular + emission.
  
• SDR
  An "incorrect" term used to describe normal SDRAM (it is supposed to be an acronym for single data rate, concerning the fact that DDR-SDRAM can send data twice as often as SDRAM).
  
• Sideband Addressing (SBA)
  See AGP Sideband Addressing.
  
• Skinning
  Despite its name, this process has nothing to do with skin. It is just another name for vertex blending.
  
• SLI
  Scanline Interleave. A process whereby two or more rasterizers are used to render a frame; each chip handles so many of the scanlines from the frame. The process increases fill rate but at a cost of manufacturing complexity and design.
  
• Sound Blaster
  This is the trademark name for a range of audio adapters by Creative Technology Ltd. The original card was very popular and it set many of the standards still used today, making the "Sound Blaster" the minimum requirements for lots of games.
  
• Sound Card
  This is a plug-in card that contains the audio adapter as well as the necessary input and output sockets; most sound cards use a PCI slot.
  
• Southbridge
  This is the name for the chip in most desktop PC motherboards that handles the data transfers for devices that use the PCI, ISA, and IDE buses.
  
• S/PDIF
  Sony/Philips Digital Interface. This is actually a format used during the transfer of audio files from one digital device to another. Since there is no conversion to and from an analogue signal in the transfer, there is virtually no degradation in quality. Devices that support the S/PDIF format often use the format for fibre optic connections.
  
• Specular Lighting
  See Shading. A dull, rough object will have a dull, rough colour - ambient or diffuse lighting is sufficient to accurately model how the object will appear. Shiny or smooth objects though will produce specular highlights and therefore require a different lighting model.
  Specular lighting is more realistic than diffuse lighting but it is much slower as it requires details of the direction of the light source (or sources), plus the viewpoint of the camera.
  
• Speculative Read/Write
  When a processing chip issues a read/write instruction to a memory controller, it actually sends two pieces of information: the read/write command, plus the address location of where the data is or is going to be. A speculative read/write involves the controller passing on the read/write command before it has finished working with the address. This helps to reduce the overall delay in the data flow and thereby boosts performance.
  
• Spherical Environment Mapping
  See Environment Mapping. This method of texture mapping a surface uses, as the name suggests, a spherical environment map. Since this requires the use of just one texture map, it is more commonly used than cubic mapping (which requires 6). Having just one map also means that the actual texture mapping process is faster too although one disadvantage of sphere maps is the "wasted" texture buffer space - texture maps are square/rectangular, so parts of the map will be blank.
  Another problem is that a graphics adapter will sample the sphere map linearly - it does not take into account that the points on the map are separated by arcs and not lines. Therefore, samples taken from the edges of the sphere map will appear distorted on the object unless some form of correction (or different texture sampling method) is applied.
  
• Sprite
  Sprites, in a modern 3D game, are typically 2-dimensional images textured on a simple rectangular primitive. They are used for lots of different effects - flames, smoke, billboards, grass, leaves, particles, etc - as they are quick and ease to produce.
  
• SRAM
  Static Random Access Memory. This form of memory differs to the more common DRAM (dynamic RAM) in that the data is "stored" using a sequence of transistors, rather than the capacitor cells in DRAM. This results in greater power consumption but without the need for constant refreshing. SRAM memory chips can be accessed much quicker than DRAM chips but since they are physically larger and more expensive to produce, SRAM is predominately used for L2 cache in CPUs.
  
• SSE
  Streaming SIMD (Single Instruction, Multiple Data) Extensions. This is Intel's marketing term for a set of instructions that access special registers within the CPU (if it has the SSE feature). The SSE registers are larger than the MMX registers, and they can store floating-point values unlike the integer ones that the MMX registers are restricted too. Applications that require many floating-point calculations to be performed, e.g. 3D games, can gain a boost in performance by using SSE instructions, as they are custom-built for such work.
  
• Stencil Buffer
  This is a portion of the local memory used hold to a "stencil" value for each pixel in the frame. During a stencil test, the graphics adapter checks the contents of the buffer for each pixel against another value, to see if some form of action is required. The stencil buffer can be used for a variety of effects or procedures; it can be used to mask off sections of a frame or to produce realistic looking shadows.
  
• Stencil Shading
  This is an incorrect term, generally used with anything involving rendering with, or from, a stencil buffer.
  
• Stencil Shadows
  This is a rendering technique used to produce realistic looking shadows. A light source will cause a shadow to be projected away from an object; the process of making a stencil shadow requires the object to be rendered several times, with at least once into a stencil buffer.
  The projected shape in the buffer acts like a mask - successive renderings of the scene around the object is checked against the buffer and any pixels that fall into the area marked out by the mask are given a very dark or plain black value. Although the process is relatively straight forward, the large number of passes (and read/writes of the stencil buffer) means that a scene with lots of shadows or light sources will be very demanding to render.
  
• Stencil Test
  If used, this test is performed near the end of the rendering pipeline in Direct3D - after the alpha test but before the depth test. The process involves the graphics adapter comparing a pixel's stencil buffer value against another; a test pass results in some form of action taking place (e.g. the value in the stencil buffer is replaced or the pixel is declared visible).
  
• Supersampling
  Supersampling is one method of doing full-scene antialiasing. In its simplest form, the process involves the graphics adapter rendering a frame at a higher resolution, than the one to be displayed, into the local memory. It is then sampled using a grid of points set on the lower resolution, with each pixel in the final displayed frame being a blend of the samples taken.
  
• Surround Sound
  As the name suggests, this is sound system that uses multiple audio channels and speakers to "surround" the listener. Games supporting this demanding audio arrangement can produce greater realism when moving sound sources around the scene.
  
• Synchronisation
  The process whereby two or more systems are timed so that events occur in time with each other.
  
• Swizzling
  This is a process whereby data is reorganised or replicated to improve performance and efficiency. Texture swizzling, for example, can involve rearranging the texture map layout, the texture address coordinates or even copying individual pixel components to increase the efficiency of texture cache.
  
  

  T

• T&L (TnL) Transforms are the spatial calculations that are executed on vertices (to move them about, etc); lighting is the process of gauging the colour of the vertex and the rest of the polygon due a light source; clipping involves checking which vertices are visible within the camera's field of view and removing them if necessary.
  Such procedures are all computationally very intensive, as they involve long floating-point calculations. Normally such work is done by the CPU, sending the results to the graphics adapter to be rasterised. Virtually all new graphics adapters released onto the market today can perform all T&L work in hardware, i.e. the graphics processor does it and not the CPU. Since these devices are custom-built to carry out these calculations, the tasks are typically done much faster.
  
• Tearing
  See Vsync. Tearing is a visual effect that is due to a lack of synchronisation between the monitor screen refreshes and the front/back buffer swaps of the graphics adapter. It can look like a constant flickering, running up and down the screen, or simply like a frame cut into 2 separate halves.
  
• Tessellation
  This is the process of increasing the number of triangles within a simple mesh by the vertex processing part of the graphics adapter. This is more preferable to sending lots of triangles to the card, as it keeps the amount of data to be sent across the AGP or PCI bus to a minimum - all you need to send is the basic mesh and the amount of tessellation to be performed.
  
• Tessellator
  The part of the graphics adapter or rendering pipeline that performs tessellation.
  
• Texel
  A texture element, i.e. a pixel from the surface of a texture (note that this does not necessarily correspond to a pixel on the monitor).
  
• Texture
  A texture is a 2-dimensional image that is "mapped" onto a polygon or several polygons, to produce the required object in a scene.
  
• Texture Blending
  See Blending. This is the process of multiply or adding the texel values from two textures maps together in order to produce an overall effect - e.g. lightmaps are multiplied with the base texture to give the impression of lighting.
  
• Texture Buffer
  This is the portion of the local memory that is used to store texture maps. If there is insufficient space within the memory to store the size of buffer required, a number of texture maps will be kept in the AGP aperture or system memory until required.
  
• Texture Clamping
  This procedure locks the range of texture coordinates available to a certain range (usually 0 to 1). If a vertex contains a texture coordinate outside this range, then the texel value given will be the same one as from the edge of the texture. This function is useful if you do not want a particular texture map effect to be repeated - for example, a specular highlight lightmap obviously only needs to appear once on the object.
  
• Texture Compression
  A term used to describe a process that reduces the file size of a texture map (Direct3D uses a compression format called DXTn). The advantages of compressing a texture are that it allows for higher resolution (and therefore greater detail) texture maps to be stored without using higher amounts of memory and it can also reduce the load on the local memory bandwidth as the textures are copied about. The disadvantage is that the compression used in Direct3D is lossy.
  
• Texture Coordinates
  These values are the locations of each pixel in a texture map.
  
• Texture Filtering
  This term is used to describe a procedure done when apply texture maps to objects. Filtering is needed when there are more or less pixels in the texture map than there are in the object to be mapped - for example, the object may be in the distance slightly and so not all the detail in the texture will be visible, and therefore not necessary.
  The process mainly involves taking the surround texture pixels (or texels for short) to the one that is to be applied to the object pixel and then blending them together, to produce one single texel. There are various filtering techniques, each one using different amounts of blended texels and calculation methods but the standard ones in DirectX are:
  • Nearest-point: This method forces the graphics chip to choose a texel from the texture map that is nearest in value to the one defined by the texture coordinate - the latter may be non-integer in value but the texel will always be a fixed integer. While this process is fast, it produces poor quality textures.
  • Linear: Like nearest point, this method uses the texel nearest to that defined by the texture coordinate but then takes a weighted average of the surrounding pixels in a "North-South-East-West" pattern. In other words, 4 texels are used to produce one screen pixel. The more common name for this method is bilinear filtering.
    An extension to this procedure is trilinear filtering where the process is done over two successive mipmaps, rather than just one. This helps reduce the visible boundaries between mipmaps on large objects, although at a cost to the rendering performance. This is because 8 texels are used to produce one screen pixel.
  • Anisotropic: This filtering method involves additional calculations of the texture map taking place before any texel sampling is done. The anisotropy of a texture map is a measure of how angled it is in relation to the camera; the greater the angle, the higher the level of anisotropy. Although graphics adapters use different means to employ anisotropic filtering, the underlying principle is the same - take more samples, and adjust the sampling pattern, for higher levels of anisotropy.
    Anisotropic filtering can have a significant effect on the rendering performance compared to linear or nearest-point, but it produces better quality texture maps.
  Texture filtering is so common today that many manufacturers will quote fill rate values while using bilinear filtering.
  
• Texture Layer
  A term that usually refers to the texture map being applied to an object. For example, a wall might have 2 layers applied to it: a base layer, such as bricks, with the second layer being a lightmap. The phrase is slowly being replaced by texture stage though as it makes more sense to describe the texturing ability of a graphics adapter in terms of how makes stages it can do per pass, rather than how many layers.
  
• Texture Map
  A more commonly used (and more complete) name for a texture image. The procedure of apply texture maps to an object, in the rendering process, is called texture mapping.
  
• Texture Mirroring
  This procedure forces a texture map to be repeated across the surface of an object at regular intervals, but with each repeat being a mirror image of the one next to it (in both directions). The effect is similar to texture wrapping but the alternating pattern of textures is useful for simulating surfaces such as wood or grass.
  
• Texture Palette
  This is a look-up table associated with a texture map; paletted textures have a number of benefits: (1) reduced file size by storing the actual texel colours in the palette, so less bits are needed to describe each texel; (2) simple texture map animations can be performed by looking up the palette and modifying the original texture map.
  
• Texture Stage
  This is the section of the complete texture mapping process - graphics adapters can perform one or more texture stages per pass (more being obviously better!). Typically, the first stage of mapping process is to apply the base texture layer but it could equally be a pixel shading process such as lighting.
  
• Texture Thrashing
  The texel values that are used by a graphics processing chip need to be stored in the memory "closest" to the chip, to ensure optimal performance - in order of preference, the locations should be the texture cache first, then the texture buffer, then the AGP aperture and finally the system memory.
  Texture thrashing occurs when the required texels are not in the nearest source, resulting in wasted time while the data is copied across. Thrashing between the cache and buffer occurs quite frequently but given the high performance of the memory system found on most graphics adapters today, the problem is an acceptable one. Thrashing between the buffer and aperture/system memory is far more significant, as the bandwidth of the AGP bus is far less than even the weakest of graphics cards.
  Texture thrashing will also occur if there is insufficient local memory for the texture buffer, causing some texture maps to be stored in the aperture or system memory. If they are ever required, then the rendering process must wait until they have been copied across.
  
• Texture Wrapping
  This procedure forces a texture map to be repeated across the surface of an object at regular intervals, with each repeat being identical to the other. The effect is similar to texture wrapping but the fixed pattern of textures is useful for simulating surfaces such as metal.
  
• Thermal compound
  This is a substance that is designed to seep into and fill the microscopic gaps in a surface. This improves the transfer of heat energy between two connecting surfaces by increasing the amount of contact area. The best thermal compounds conduct heat very well but with virtually no capacitance or electrical conductivity.
  
• TIFF
  Tagged Image File Format. A bitmap image format, the specification for which is owned by Adobe Systems, that is platform independent - this makes it a popular choice for the storage of images. The TIFF format supports both lossy and lossless forms of compression (such as JPEG and ZIP).
  
• Tile-based Rendering
  Sometimes called bucket rendering. The standard method of rendering a 3D scene involves sending all of the data in the frame directly to the graphics adapter and let it piece together all the objects one-by-one. This process is rather inefficient as a single frame will contain huge amounts of data - at a resolution of 1024 x 768, a 24-bit z-buffer alone would take up 2.25MB of memory - that will need to be accessed frequently. One solution would be to have RAM built into the chip that could be read and written to very easily; unfortunately, the cost would be beyond the range of the desktop graphics adapter.
  Tile-based rendering is a hardware solution to the problem. The graphics adapter has small amount of memory built into the processing chip - not enough to store the entire frame data but enough to contain a small fraction of it. Therefore, before being sent to the adapter, the frame is split up into a grid of "tiles" with each one contain the necessary data for that piece. The adapter then processes the frame, one tile at a time, but since all the tile data can fit into the chip, the only time the local memory needs to be accessed is during texturing and the final buffer write.
  The effective use of the system resources means that tile-based rendering is far less likely to have fill rate problems due to a lack of memory bandwidth. Desktop graphics adapters that support this method are rare but the Kyro family of cards have been very popular.
  
• TMU
  Texture Mapping (or management) Unit. Originally, this part of a graphics processing chip was responsible for performing the job of requesting texture information from the local memory, blending the requested values together (see filtering) and placing the value into registers within the pixel pipeline. The very latest graphics adapters have TMUs so deeply integrated into the pixel processing units that they no longer have such a specific role - however, original function of the TMU is still applied.
  It is generally considered that the acceptable minimum ability of any TMU is to request and receive 4 texel samples, blend them together and send the result to a register within one clock cycle of the pixel pipeline - in other words, map one bilinearly filtered texel in one "tick". Graphics adapters with more than one TMU per pipeline can obviously apply more than one texel to a pixel per clock cycle - however, this means that the local memory will be accessed more often and so the potential texturing performance may not be fully achieved.
  The current trend is to avoid using multiple TMUs per pipeline and instead allow more texture stages to be executed per pass.
  
• Track
  The line on a storage medium that contains the data (through various means, such as pits on a CD-ROM track or magnetic substrates on a hard disc). Depending on the medium used, the track may be completely circular (e.g. hard discs) or spiral (e.g. CD-ROM disc). All data is stored sequentially along the length of the track.
  
• Transform (Transformations)
  In 3D graphics, these are the mathematical functions that are performed on vertices in order to change the position of an object, location of a camera, direction of view, the perspective of the scene and so on. The primary functions used in transform are matrix operations.
  
• Transistor
  An electronic component that can act as a switch, but without moving or mechanical parts. Specific sequences of transistors can perform logic functions; millions of them together form the basis of all modern processing chips.
  
• Transparency
  See Alpha. The ability to make surfaces look transparent in a 3D scene is vital - how else could one produce glass or plastic materials, or recreate the surface of a lake?
  
• Triangle Fans
  A triangle fan is a group of connected triangles stuck together such that they all share one vertex in the middle of the fan; this helps to increase the triangle-to-vertex ratio, which reduces the amount of work for the vertex processing stage of the pipeline.
  
• Triangle Lists
  As the name suggests, a triangle list is a catalogue of unconnected triangles. Since each triangle is individual, the list must contain at least 3 vertices and the total number must be an integer number of 3. Triangle lists are useful for managing large groups of separated triangles.
  
• Triangle Strips
  A triangle strip is a group of connected triangles stuck together such that they share one or more vertices. For example, a simple square could consist of 4 triangles but only 6 vertices would be needed to describe the shape, a saving 6 if the triangles were modelled separately. This helps to increase the triangle-to-vertex ratio, which reduces the amount of work for the vertex processing stage of the pipeline.
  
• Triangle-to-vertex Ratio
  A single triangle consists of 3 vertices; if all of the triangles in a frame from something like the Battle of Proxycon were individually set out, the number of vertices needed could be as high as 750,000! Different methods are employed by developers to reduce the total number of vertices required by having polygons share vertices - there are various methods from triangle strips to index buffers, but they all increase the triangle-to-vertex ratio. The greater this value is, the more efficient the vertex processing will be.
  
• Trilinear Filtering
  See Texture Filtering. This method of texel filtering performs bilinear filtering on two successive mipmaps, before blending the results together for the final texel. Trilinear filtering helps to reduce the noticeable "steps" between mipmap levels within a 3D scene but at a cost to performance. Each texel applied to a polygon requires 8 samples to be taken from the texture map in question; trilinear filtering therefore costs twice the memory bandwidth usage than bilinear filtering.
  
• Triangle
  A triangle is a shaped formed by three vertices (points). All desktop graphics adapters produce 3D objects by rendering groups of triangles as these shapes require the least number of vertices to describe and connecting triangles can share vertices too.
  
• Triple Buffering
  See Buffering. Triple buffering is essentially the same as double buffering but an additional "back" buffer is used; the graphics chip alternately renders successive frames into the two back buffers. The main advantage with using triple buffering is when vsync is enabled, as it allows the frame rate to remain far more stable. The disadvantage is the increase in the amount of the local memory used, although this is not much of an issue with modern graphics adapters.
  
• Tris
  Abbreviation for "triangles". See Triangle.
  
• Tweening
  An abbreviation of the phrase in-betweening, the term refers to a process whereby two separate stages are blended together to simulate a smooth change from one to the other. Vertex tweening in Direct3D, also known as vertex morphing or keyframe interpolation, blends the positions or normals from two vertex streams to give a smooth motion between two frames without having to manually produce the animation.
  
  

  U

• USWC
  An acronym for Uncacheable Speculative Write Combining. This is a setting that some motherboards have in the BIOS - when enabled; the data from the CPU for the graphics adapter is not cached and is written to the local memory in a write burst (see Write Combining). Not all graphics cards support this function and the small potential increase in performance is seen by many people to be not worth the typical decrease in stability.
  
• UV coordinates
  These are the equivalent x-y coordinates of a texture map.