The Game Developers Conference in San Jose California turns out as a jewel of a news source this week. The Microsoft DirectX Day is held during the GDC and is intended as a presentation and discussion platform for developers. In a recent lecture held by Charles Boyd, personal business group product manager @ Microsoft, it became clear that DirectX 9 will cause another stir in the 3D hardware scene and that surprisingly Matrox will be among the first to have new technologies implemented into their products. Read more...
According to Boyle's DirectX 9 presentation users can expect the following innovations (please note this is a rough translation from japanese): "Displacement Mapping" opens up the territory of the new photo reality 3D graphics The territory of the new photo reality 3D graphics which "displacement mapping" opens up as for being released first with demonstration of DirectX 9. With it (displacement mappping) was the image due to the technology which is called [...]. Concretely the character picture of several hundred polygons, by the fact that displacement mapping is applied, becomes something which has [a] very complicated form with real [...]. With it was something which is said. When you say, the simple character real becomes form, the automatic polygon interpolation mechanism "TRUFORM" which the ATI RADEON 8500 series supports (N-PATCH) and the like the person who is associateen they probably are many things. If you say this directly, in order the rectilinear character model to be formed with living thing curved line, it is something which rendering is done. With completely another approach from this, to convert to the form which becomes independent rather vis-a-vis the geometry which the character model has it is possible displacement mapping. There is the meaning: Displacement= displacement and conversion, but displacement mapping being what ones? Considerably, if rough expression is done, it can extended edition of bump mapping. Bump mapping is called the height map, it is something of the image that it sticks unevenness to plane surface, unevenness was expressed (or housed that releasing line map) making use of texture, but as for displacement mapping the unevenness information the polygon data processing (conversion) is done on the basis of [...]. With it becomes the image which is said. At the highest only the dynamic range of 8 - 16bit integer there was unevenness information of bump mapping, but because the unevenness information which can be applied with displacement mapping is the 32bit floating point, it means to be able to express rather complicated form. With demonstration, the numbers of polygons being few, it made of the form and the like of the rib bone come out which comes to the surface in the chest board of the character which the っ ぺ り is done, the numbers of apexes were few and harasssed and in order the cord head to modify in the form which had the smooth contour line showed. Actualization method of concrete form modification seems that differs every GPU, but when actualizing the N-PATCH, it is thought the ecad tessellation the way (polygon division) the mechanism is utilized. That character being in the distance, when being visible small and the like, dropping the precision of displacement mapping automatically, it seems that also the mechanism of the dynamic LOD and the like which makes the operational quantity is offered. Furthermore, with DirectX 9 as for applying plural displacement mapping to one object simultaneously you say it is not possible. The kind of mechanism which synthesizes two forms seems that becomes support after the DirectX 10. Pixel Shader Version 2.0 The programmable apex ti (the PVS) [...], [...] (the PPS) it is pulled up together to version 2.0 concerning the version of the [...] which is one of the highest interest things where it confronts DirectX 9. With DirectX 8.1 former 1.1, latter 1.4 calls imbalance had become relationship, but version number is standardized and becomes the shape which will be been clear. Furthermore, the DirectX originally, while " being different the hardware absorption ", the expectation which is born as API which operates to high speed, the fee char which with DirectX 8.1 is supported in NVIDIA system and ATI system to polarization became the feeling which is converted. With DirectX 9 including reflection point to this, for the singing/stating " the full compatible of the time before the PVS1.1 and the time before the PPS1.4 makes プログラマブルシェーダ 2.0 support indispensable ", it was declared. Concerning the PVS2.0 it mentions later, but because as for the PPS2.0 as for the PPS1.4 becoming the base it is clear, the GPU " R300 " which is succession of the ATI RADEON system which full corresponds in the PPS1.4 as for the possibility the implement of doing the PPS2.0 is high. Branch instruction subroutine call is supported in the PVS, the GPU evolves more and more to CPU like With the PVS2.0 large-scale order addition was done, how, divergence, was decided with that subroutine / call instruction is supported. As for divergence receiving the relative order result, binary condition (the TRUE/FALSE) divergence is possible. However, it is limited to lower direction. Also subroutine order is similar specification. The fact that it is the subroutine, is the case that it is stack operation, but jump outside the stipulation which operated stack and recurrent structure is not supported. Also construction of the loop structure which used the loop control register being possible in branch instruction, maximum of 16 heavily the loop is possible, but as for passing through from the loop there is the restriction that it is not possible. With say, while the PVS which is limited to the logic where the dependency is little, with the relationship which had become the instruction set where the degree of freedom where the dependency is high at a stroke is high, reached the point where it can actualize complicated apex flow, arithmetic unit of the PVS (the ALU) as for the number of transistors it is thought it increases. Because the dual apex シェーダ of the GeForce4 Ti the dependency of the apex program is low, very it is the case that performance of 2 times occurs, but if with the GPU of DirectX 9 correspondence unit is increased simply, throughput probably will stop limiting rises with. In such sense, more and more you can call the GPU it has become CPU like. In regard to the PPS the point which comes to the point of being permitted you can list to maximum depth 4 dependence reading as the strengthening point. It forms index value from a certain operational result, texture (with does not say from that value and as for being the picture, it is the case that the numeric data have entered but) to refer to, it removes new index value and it means that...... with the control was said is permitted up to depth 4. It was not shown to concrete use example, but so far being soft and high-level above, it suggested that complicated shading becomes possible. Furthermore, with the PPS2.0 branch instruction is not supported unlike the PVS2.0. The fact that divergence is supported with the PPS must be when it is the prospect that it becomes the PPS3.0. When with the PPS3.0 furthermore the index register is supported, with thing, this and, it becomes the program execution system where the dependency is high. In order the implement to do the PPS3.0, this and it probably means that the number of suitable transistors becomes necessary. In multiple display environment real correspondence to the 3D graphics In addition, it seems that can do functional strengthening in the small point. We will have decided to keep seeing one each. DirectX 9 API object model. You observe most to the DMA & the Tess of block on in the figure. As for Tess abbreviation of tessellation. Namely as for the GPU of DirectX 9 correspondence you can ask that it designates that it has the mechanism which the [...] it is possible polygon division as prerequisite. This is utilized even in actualization of displacement mapping. It starts in the DualHead and the HydraVision and the like, multiple display environment has generalized recently in the TwinView, but even unexpectedly the 3D [...] as for the GPU which functioning is done is little at the time of secondary displaying. Under present conditions it probably is about the GeForce4ti and the RADEON8500. This mechanism is supported with DirectX 9 formally. It means that concretely one Direct3d device (the GPU), offers the mechanism which drives plural display outputs. Indication when it is limited to full screen mode, the [...] being done, video memory and the like share just, reaches the point where it is used. And new the 40bit color is supported in picture mode. This is color expression system of the RGB each 10bit. So far, it is the case that 1,677 ten thousand color colors of the RGB each color 8bit are most significant color number, but with the 40bit color it becomes 10 hundred million colors. Until high-level shading with the [...] becomes possible, final lust information is calculated, when frequency of operation becomes many, with the 8bit based color channel error accumulating, it has known that finally the ideal result is not produced. This seems like those which are devised because it corresponds to the problem such as that. Simply, when you say whether the video card which finally supports the indicatory color 40bit color steadily appears it seems that is not such reason. Is because with the notion that where you say it does not have the ability the present most display devices to indicate the 40bit color. After all, it does lighting operation with 10bit color channel system, makes round final indicatory color in the 8bit color channel...... probably becomes the realistic solution. Furthermore, with DirectX 9 other than the RGB each 10bit, are supported to the 16bit, the 16bit floating point and the 32bit floating point as a surface format. Furthermore Boyd expresses the pixel where each color of the RGB is expressed with the floating point the " Float Pixels ", as for the especially 16bit FloatPixels width of the dynamic range and convenient of the data quantity it is estimate the empty, 3d graphic system use frequency becomes high from with this. DirectX 9 corresponding video card first as for feature from MATROX? The G800 coming finally? Well according to the article above the presentation of Mr. Boyd from Microsoft was pleasantly interrrupted by a MATROX technician who showed the attendents a working presentation of "displacement mapping". A beta test for DirectX 9 is planned for summer 2002 and the release is set for early 2003. Probably in conjunction Windows XP Second Edition which will incorporate all updates and additions from SP1 (Internet Explorer 7, DirectX9?, Bluetooth, Mira and other enhancements). Source: PC Watch Japan
According to Boyle's DirectX 9 presentation users can expect the following innovations (please note this is a rough translation from japanese): "Displacement Mapping" opens up the territory of the new photo reality 3D graphics The territory of the new photo reality 3D graphics which "displacement mapping" opens up as for being released first with demonstration of DirectX 9. With it (displacement mappping) was the image due to the technology which is called [...]. Concretely the character picture of several hundred polygons, by the fact that displacement mapping is applied, becomes something which has [a] very complicated form with real [...]. With it was something which is said. When you say, the simple character real becomes form, the automatic polygon interpolation mechanism "TRUFORM" which the ATI RADEON 8500 series supports (N-PATCH) and the like the person who is associateen they probably are many things. If you say this directly, in order the rectilinear character model to be formed with living thing curved line, it is something which rendering is done. With completely another approach from this, to convert to the form which becomes independent rather vis-a-vis the geometry which the character model has it is possible displacement mapping. There is the meaning: Displacement= displacement and conversion, but displacement mapping being what ones? Considerably, if rough expression is done, it can extended edition of bump mapping. Bump mapping is called the height map, it is something of the image that it sticks unevenness to plane surface, unevenness was expressed (or housed that releasing line map) making use of texture, but as for displacement mapping the unevenness information the polygon data processing (conversion) is done on the basis of [...]. With it becomes the image which is said. At the highest only the dynamic range of 8 - 16bit integer there was unevenness information of bump mapping, but because the unevenness information which can be applied with displacement mapping is the 32bit floating point, it means to be able to express rather complicated form. With demonstration, the numbers of polygons being few, it made of the form and the like of the rib bone come out which comes to the surface in the chest board of the character which the っ ぺ り is done, the numbers of apexes were few and harasssed and in order the cord head to modify in the form which had the smooth contour line showed. Actualization method of concrete form modification seems that differs every GPU, but when actualizing the N-PATCH, it is thought the ecad tessellation the way (polygon division) the mechanism is utilized. That character being in the distance, when being visible small and the like, dropping the precision of displacement mapping automatically, it seems that also the mechanism of the dynamic LOD and the like which makes the operational quantity is offered. Furthermore, with DirectX 9 as for applying plural displacement mapping to one object simultaneously you say it is not possible. The kind of mechanism which synthesizes two forms seems that becomes support after the DirectX 10. Pixel Shader Version 2.0 The programmable apex ti (the PVS) [...], [...] (the PPS) it is pulled up together to version 2.0 concerning the version of the [...] which is one of the highest interest things where it confronts DirectX 9. With DirectX 8.1 former 1.1, latter 1.4 calls imbalance had become relationship, but version number is standardized and becomes the shape which will be been clear. Furthermore, the DirectX originally, while " being different the hardware absorption ", the expectation which is born as API which operates to high speed, the fee char which with DirectX 8.1 is supported in NVIDIA system and ATI system to polarization became the feeling which is converted. With DirectX 9 including reflection point to this, for the singing/stating " the full compatible of the time before the PVS1.1 and the time before the PPS1.4 makes プログラマブルシェーダ 2.0 support indispensable ", it was declared. Concerning the PVS2.0 it mentions later, but because as for the PPS2.0 as for the PPS1.4 becoming the base it is clear, the GPU " R300 " which is succession of the ATI RADEON system which full corresponds in the PPS1.4 as for the possibility the implement of doing the PPS2.0 is high. Branch instruction subroutine call is supported in the PVS, the GPU evolves more and more to CPU like With the PVS2.0 large-scale order addition was done, how, divergence, was decided with that subroutine / call instruction is supported. As for divergence receiving the relative order result, binary condition (the TRUE/FALSE) divergence is possible. However, it is limited to lower direction. Also subroutine order is similar specification. The fact that it is the subroutine, is the case that it is stack operation, but jump outside the stipulation which operated stack and recurrent structure is not supported. Also construction of the loop structure which used the loop control register being possible in branch instruction, maximum of 16 heavily the loop is possible, but as for passing through from the loop there is the restriction that it is not possible. With say, while the PVS which is limited to the logic where the dependency is little, with the relationship which had become the instruction set where the degree of freedom where the dependency is high at a stroke is high, reached the point where it can actualize complicated apex flow, arithmetic unit of the PVS (the ALU) as for the number of transistors it is thought it increases. Because the dual apex シェーダ of the GeForce4 Ti the dependency of the apex program is low, very it is the case that performance of 2 times occurs, but if with the GPU of DirectX 9 correspondence unit is increased simply, throughput probably will stop limiting rises with. In such sense, more and more you can call the GPU it has become CPU like. In regard to the PPS the point which comes to the point of being permitted you can list to maximum depth 4 dependence reading as the strengthening point. It forms index value from a certain operational result, texture (with does not say from that value and as for being the picture, it is the case that the numeric data have entered but) to refer to, it removes new index value and it means that...... with the control was said is permitted up to depth 4. It was not shown to concrete use example, but so far being soft and high-level above, it suggested that complicated shading becomes possible. Furthermore, with the PPS2.0 branch instruction is not supported unlike the PVS2.0. The fact that divergence is supported with the PPS must be when it is the prospect that it becomes the PPS3.0. When with the PPS3.0 furthermore the index register is supported, with thing, this and, it becomes the program execution system where the dependency is high. In order the implement to do the PPS3.0, this and it probably means that the number of suitable transistors becomes necessary. In multiple display environment real correspondence to the 3D graphics In addition, it seems that can do functional strengthening in the small point. We will have decided to keep seeing one each. DirectX 9 API object model. You observe most to the DMA & the Tess of block on in the figure. As for Tess abbreviation of tessellation. Namely as for the GPU of DirectX 9 correspondence you can ask that it designates that it has the mechanism which the [...] it is possible polygon division as prerequisite. This is utilized even in actualization of displacement mapping. It starts in the DualHead and the HydraVision and the like, multiple display environment has generalized recently in the TwinView, but even unexpectedly the 3D [...] as for the GPU which functioning is done is little at the time of secondary displaying. Under present conditions it probably is about the GeForce4ti and the RADEON8500. This mechanism is supported with DirectX 9 formally. It means that concretely one Direct3d device (the GPU), offers the mechanism which drives plural display outputs. Indication when it is limited to full screen mode, the [...] being done, video memory and the like share just, reaches the point where it is used. And new the 40bit color is supported in picture mode. This is color expression system of the RGB each 10bit. So far, it is the case that 1,677 ten thousand color colors of the RGB each color 8bit are most significant color number, but with the 40bit color it becomes 10 hundred million colors. Until high-level shading with the [...] becomes possible, final lust information is calculated, when frequency of operation becomes many, with the 8bit based color channel error accumulating, it has known that finally the ideal result is not produced. This seems like those which are devised because it corresponds to the problem such as that. Simply, when you say whether the video card which finally supports the indicatory color 40bit color steadily appears it seems that is not such reason. Is because with the notion that where you say it does not have the ability the present most display devices to indicate the 40bit color. After all, it does lighting operation with 10bit color channel system, makes round final indicatory color in the 8bit color channel...... probably becomes the realistic solution. Furthermore, with DirectX 9 other than the RGB each 10bit, are supported to the 16bit, the 16bit floating point and the 32bit floating point as a surface format. Furthermore Boyd expresses the pixel where each color of the RGB is expressed with the floating point the " Float Pixels ", as for the especially 16bit FloatPixels width of the dynamic range and convenient of the data quantity it is estimate the empty, 3d graphic system use frequency becomes high from with this. DirectX 9 corresponding video card first as for feature from MATROX? The G800 coming finally? Well according to the article above the presentation of Mr. Boyd from Microsoft was pleasantly interrrupted by a MATROX technician who showed the attendents a working presentation of "displacement mapping". A beta test for DirectX 9 is planned for summer 2002 and the release is set for early 2003. Probably in conjunction Windows XP Second Edition which will incorporate all updates and additions from SP1 (Internet Explorer 7, DirectX9?, Bluetooth, Mira and other enhancements). Source: PC Watch Japan
