View GLINT-GAMMA_892154.PDF datasheet online --- IC-ON-LINE

Datasheet File OCR Text:

1 (c) copyright 3dlabs 1997 glint gamma a 3d geometry and lighting processor for the pc glint gamma glint gamma a 3d geometry and lighting a 3d geometry and lighting processor for the pc processor for the pc neil trevett neil trevett vice president, marketing vice president, marketing nasdaq:tdddf nasdaq:tdddf www www . . 3dlabs 3dlabs . . com com .com .com .com 4 .com u datasheet
2 (c) copyright 3dlabs 1997 agenda agenda agenda ? a backgrounder to the 3d geometry pipeline ? system considerations for 3d geometry acceleration ? glint gamma architecture overview ? glint gamma detailed architecture ? board design examples using glint gamma .com .com .com .com 4 .com u datasheet
3 (c) copyright 3dlabs 1997 geometry is the pcs 3d bottleneck geometry is the pcs 3d bottleneck the last difference between 3d workstations and pcs the last difference between 3d workstations and pcs ? the fastest cpu cannot keep todays rasterization silicon saturated if running the geometry in software ? professional 3d is particularly demanding - many small polygons used for precise modeling of surfaces - viewperf uses polygons less than 1 pixel in size transforms lighting setup transforms transforms lighting lighting setup setup rasterization rasterization rasterization 100% of rasterization in silicon - typically 500k-1m polygons per second complete geometry pipeline in software = 350k polygons/sec on a pii/266 transform and lighting in software = 1 million polygons/sec application application application setup glint delta offloads setup processing = 100mflops cpu running application & geometry = 350k polygons/sec .com .com .com .com 4 .com u datasheet
4 (c) copyright 3dlabs 1997 what what is is a 3d object? a 3d object? ? surfaces of objects are described as a grid of polygons ? the vertices of the polygons are located in 3d coordinate space - x,y,z ? the objects making up a scene are held in a database x1,y1,z1 x2,y2,z2 x3,y3,z3 x y z a 3d object e.g. a sphere polygons describing the spheres surface .com .com .com .com 4 .com u datasheet
5 (c) copyright 3dlabs 1997 3d processing stages 3d processing stages the journey from database to screen the journey from database to screen traversal transforms lighting setup coverage color clip merge traversal traversal transforms transforms lighting lighting setup setup coverage coverage color color clip clip merge merge geometry rasterization what objects are in current scene? what objects are in current scene? what color are the polygons? what color are the polygons? what shape are they on the screen? what shape are they on the screen? which pixels are covered? which pixels are covered? what color is each pixel? what color is each pixel? which pixels are visible? which pixels are visible? write the pixels to the framebuffer write the pixels to the framebuffer operations operations on vertices on vertices operations operations on pixels on pixels where are the polygons? where are the polygons? .com .com .com .com 4 .com u datasheet
6 (c) copyright 3dlabs 1997 transform processing transform processing positioning objects in a scene positioning objects in a scene ? input = the vertices of all objects in the scene in a database ? output = list of visible vertices, correctly positioned ? transform for eye-point - 4x4 matrix multiplications - allows positioning with six degrees of freedom ? lots of floating point operations! 3d polygon 3d polygon eye-point eye-point x1,y1,z1 x1,y1,z1 x2,y2,z2 x2,y2,z2 x3,y3,z3 x3,y3,z3 .com .com .com .com 4 .com u datasheet
7 (c) copyright 3dlabs 1997 transform processing transform processing view volume clipping and the projection plane view volume clipping and the projection plane ? project onto projection plane - trigonometry ? clip against view volume - more trigonometry 3d polygon 3d polygon eye-point eye-point projection plane projection plane screen coordinate object screen coordinate object view view volume volume x1,y1,z1 x1,y1,z1 x2,y2,z2 x2,y2,z2 x3,y3,z3 x3,y3,z3 .com .com .com .com 4 .com u datasheet
8 (c) copyright 3dlabs 1997 geometry efficiencies view volume culling and backface culling view volume culling and backface culling ? view volume culling can eliminate the majority of polygons in a scene trivially - its behind you! ? the other big win is: backface culling - normally polygons are one-sided - polygons with normals pointing away from the eyepoint are invisible - the geometry pipeline can discard backfacing polygons - about 50% of the polygons in most real models are backfacing eye-point eye-point discard discard .com .com .com .com 4 .com u datasheet
9 (c) copyright 3dlabs 1997 lighting equation lighting equation calculating the color of each vertex calculating the color of each vertex ? objects are lit by one or more light sources - ambient light: non-directional, has a color - directional lights: with position, direction and color ? surfaces have reflective properties - diffuse reflection coefficient: independent of angle - specular reflection coefficient: depends on angle, gives glinting highlights ? color - ia.ka + id.kf + id.ks.cos? surface surface normal directional light source ? .com .com .com .com 4 .com u datasheet
10 (c) copyright 3dlabs 1997 setup calculations setup calculations preparing for rasterization preparing for rasterization ? take screen coordinates of vertices - in floating point format ? calculate slope and delta information - for each interpolant need to calculate starti, didx, didydom ? converts to sub-pixel accurate fixed point format x1, y1, z1, r1, g1, b1, a1, f1, s1, t1, q1 x2, y2, z2, r2, g2, b2, a2, f2, s2, t2, q2 x3, y3, z3, r3, g3, b3, a3, f3, s3, t3, q3 dominant edge worst case number of parameters, position, color, alpha, fog and texture .com .com .com .com 4 .com u datasheet
11 (c) copyright 3dlabs 1997 100% of the host cycles can process the application glint gamma glint gamma complete geometry processing in a chip complete geometry processing in a chip ? glint gamma integrates 100% of the geometry pipeline - at a price point suitable for pcs ? 3.3 million polygons/sec needs 2,000mflops - with full transformations and lighting enabled - 50% backface culled ? host is free to process the application - glint gamma will have a bigger effect on typical application performance than faster pixel fill-rates transforms lighting setup transforms transforms lighting lighting setup setup rasterization rasterization rasterization application application application glint gamma processes the complete transform, lighting, setup pipeline at 3.3 million polygons/sec feeds rasterization chip e.g. glint mx .com .com .com .com 4 .com u datasheet
12 (c) copyright 3dlabs 1997 system design using glint gamma hardware preprocessing for rasterization silicon system design using glint gamma system design using glint gamma hardware preprocessing for rasterization silicon hardware preprocessing for rasterization silicon ? glint gamma reads and processes polygon command stream - outputs command stream for a rasterizer processor e.g. glint mx ? 33mhz pci is not fast enough for host connection - full 33mhz pci bandwidth only supports 1.5 million polygons/sec - agp enables the full 3.3 million polygons/sec performance - need agps sideband addressing for optimized vertex fetching ? use 66mhz pci connection to rasterizer - can guarantee 66mhz on-board operation, and allows multiple devices pci or agp cpu cpu cpu glint gamma glint glint gamma gamma 66mhz pci to allow multiple devices video out glint mx glint glint mx mx framebuffer framebuffer framebuffer all pixel processing, e.g. texturing, shading, z buffering performed in rasterizer silicon only visible polygons - after clipping and culling all polygons in the scene. agp 1x allows 3.3 million polygons/sec .com .com .com .com 4 .com u datasheet
13 (c) copyright 3dlabs 1997 glint gamma functional pipeline glint gamma functional pipeline parameter fetch to fog parameter fetch to fog ? input dma controller fetches parameters - performs address translation ? opengl begin/end paradigm - points, lines, line loop, line strip, triangles, triangle strip, triangle fan, quads, quad strips ? full geometry transformation processing - vertex (view model and projection matrix) ? texture coordinate generation - spherical, object linear, eye linear ? fog calculations - linear, exponential and exponential squared parameter fetch primitive processing transforms texture coordinates fog calculations clipping lighting setup .com .com .com .com 4 .com u datasheet
14 (c) copyright 3dlabs 1997 glint gamma functional pipeline glint gamma functional pipeline clipping to setup clipping to setup ? full clipping - frustrum clipping - six user clip planes - backface culling ? full lighting calculations - up to 16 light sources - soft degrade to unlimited number of lights - point or directional lights - local lights, spot lights, attenuated lights - two-side lighting ? triangle, line and point set up - same as glint delta, but faster - aliased or anti-aliased - stippled or non-stippled parameter fetch primitive processing transforms texture coordinates fog calculations clipping lighting setup .com .com .com .com 4 .com u datasheet
15 (c) copyright 3dlabs 1997 glint gamma architecture glint gamma architecture geometry processing in a agp/pci bridge geometry processing in a agp/pci bridge agp/pci master & slave interface dma controller input fifo gamma geometry engine output fifo function 0 decode with vga/8514 function 1 decode pci 1 master interface and dma control primary agp or pci bus secondary pci bus 176 pin pqfp 3.3v power 5v i/o bypass path full geometry and lighting processor allows transparent use of vga and 8514 behind bridge provides setup-fetch overlap full bus master dma to glint mx .com .com .com .com 4 .com u datasheet
16 (c) copyright 3dlabs 1997 glint gamma pipeline glint gamma pipeline ? pipeline of function units ? message passing protocol between units ? each unit heavily-pipelined internally ? 150 total pipeline stages through the chip command unit vertex machine transform unit geometry unit normalization unit lighting unit material unit delta engine dma unit with page lookup for non-contiguous command buffers processes primitive types transform vertices, textures and normals frustrum and user plane clipping, backface culling, fog calculations normalizes and transforms normals lighting calculations for 16 lights combines material properties with lights, including specular textures rasterization setup .com .com .com .com 4 .com u datasheet
17 (c) copyright 3dlabs 1997 f i f o f i f o tag data tag data tag data tag data two stage fifo two stage fifo processing register storage control input stage output stage internal pipelining as required function unit structure function unit structure function unit structure ? the unit pipeline uses a message passing paradigm ? a message is made up of a tag field and a data field - 11 bit tag identifies the message type - 128 bit data holds four 32 bit floats = complete vertex eg rgba or xyzw ? the unit processes each message according to message type .com .com .com .com 4 .com u datasheet
18 (c) copyright 3dlabs 1997 glint gamma is hardwired maximum performance per $ glint gamma is hardwired glint gamma is hardwired maximum performance per $ maximum performance per $ ? 3d geometry is well-understood - dont need the luxury of programmability ? functionality encompasses 100% of all key apis - opengl 1.1, direct 3d, heidi and quickdraw 3d ? control in glint gamma is a vhdl state machine - no ram or rom for program storage (less gates) - no program sequencer or instruction set (less gates) - no program fetch (less memory bandwidth) ? data paths are inferred directly from vhdl - no general purpose routing costs ? 3dlabs design technology enables devices of this complexity - rapid conversion of floating point algorithms into hardwired silicon - design tools first used to develop glint delta .com .com .com .com 4 .com u datasheet
19 (c) copyright 3dlabs 1997 vertex 0 working data routing fmul fadd fdiv fdiv fconvert vertex 1 vertex 2 input vertex stores floating point operation units vhdl coded, inferred routing temp storage output fifo typical unit processing structure hardwired processing typical unit processing structure typical unit processing structure hardwired processing hardwired processing ? all operations are ieee compatible - nan handling not implemented ? type and number of operation units varies from unit to unit vertex 3 .com .com .com .com 4 .com u datasheet
20 (c) copyright 3dlabs 1997 total floating point units total floating point units total floating point units ? all units are independent ? all units can be used simultaneously - no restriction on mix of operations ? latency varies according to the unit # units type of units throughput (cycles/result) 15 multipliers 1 10 adders, 1 25 misc. conv. and comparators 1 4 divide 5 1 power 5 2 inverse square root 10 .com .com .com .com 4 .com u datasheet
21 (c) copyright 3dlabs 1997 glint gamma glint gamma physical and project characteristics physical and project characteristics ? ibm cmos 5s6 .35, 4 layer metal ? 176 pin pqfp ? 66mhz, 5 watts ? 8 man years design time ? 600 page design specification ? over 1 million gates - fp operators - 35% - registers - 15% - random logic - 50% .com .com .com .com 4 .com u datasheet
22 (c) copyright 3dlabs 1997 glint gamma as agp bridge multi-function agp boards glint gamma as agp bridge glint gamma as agp bridge multi-function agp boards multi-function agp boards ? agp is point-to-point connection only - only one slot and only one device on the agp bus - makes multi-functions on agp problematic ? glint gamma acts as a full agp bridge - multiple devices on an agp board can access main memory - via internal 66mhz pci bus ? enables multi-function boards - uses for high performance video devices agp core logic core core logic logic glint gamma glint glint gamma gamma glint mx glint glint mx mx video hardware video video hardware hardware 66mhz pci .com .com .com .com 4 .com u datasheet
23 (c) copyright 3dlabs 1997 twin glint mx design twin glint mx design a graphics pipeline well-matched with pentium ii a graphics pipeline well-matched with pentium ii ? glint mx is scalable from 2 to 8 processors - transparently increases fill-rate - interleaved scan-line rendering for effective load-balancing ? pentium ii -> glint gamma -> twin glint mx - well-balanced application, geometry, rasterization pipeline - 2 million visible, rendered pol yg ons with full g eometr y processin g - 66 million pixels/sec bilinear-mip-mapped filtered texture - 33 million pixels/sec trilinear-mip-mapped filtered texture agp, 33mhz or 66mhz pci ramdac ramdac glint gamma glint gamma 66mhz pci glint mx glint mx localbuffer localbuffer localbuffer localbuffer glint mx glint mx framebuffer framebuffer .com .com .com .com 4 .com u datasheet
24 (c) copyright 3dlabs 1997 summary summary summary ? 3d geometry is floating point intensive ? it is the current bottleneck for increased 3d pc performance ? 3d geometry functionality is well understood and encapsulated in standard apis ? it can be hard-wired for maximum performance per $ ? glint gamma implements the geometry and lighting for all key apis - including opengl and direct3d .com .com .com 4 .com u datasheet

▲Up To Search▲

Price & Availability of GLINT-GAMMA

	To Download GLINT-GAMMA Datasheet File
If you can't view the Datasheet, Please click here to try to view without PDF Reader .