Silicon Investor (SI) -- The First Internet Community

STOCKTALK

We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor. We ask that you disable ad blocking while on Silicon Investor in the best interests of our community. If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.

Technology Stocks : 3DFX -- Ignore unavailable to you. Want to Upgrade?

To: Jeffrey P who wrote (13489)	6/25/1999 6:42:00 PM
From: Obewon	Read Replies (1) \| Respond to of 16960

I found this courtesy of AGN3D.com website. Note that in the very first paragraph the author reveals his biases. Of course, as a 3dfx shareholder and card user, I was interested to see that 3dfx is struggling in the market with its Voodoo3 card. In fact, the price reductions 3dfx has found necessary to put in place (ie. none!) and the rebates that they are offering (again, none!) in order to stay competitive with TNT2 based cards is quite revealing. (Can you tell I'm worried?!??) Obewon Original webpage: kachardware.dk Riva TNT3/NV10 Author: Kristian H. Andersen Date: 06-25-1999 Manufacturer: Nvidia Nvidia has through its first years always seemed to fall a step short of 3Dfx. Their first products failed to hit the consumers and their marketing plans never reached their target group. The entire picture turned when the company this summer released their TwiN Texel 2 (TNT2) video chipset: 3Dfx saw oneself beaten by their smallest competitor and are now struggling to survive with their Voodoo3 cards. Nvidia has taken the leading position in the accelerator market and TNT2 was just the top of the iceberg; their TNT3/NV10 (codename NV10) will hit 3Dfx with full strength in the end of the year and the specifications leaked out from Nvidia reveals that 3Dfx probably will be toppled. TNT3/NV10 already? Nvidia's latest roadmap informs their TNT3/NV10 will hit the market in late 1999. TNT2 is experiencing a tremendous success and 3Dfx's market share has crash to the ground, so it can seem quite peculiar already wave plans about releasing a new chipset. On the other hand 3Dfx experienced the same development a few years ago and they did nothing else than resting on the laurels and then hit the ground when their monopoly status was broken, so it is quite possible Nvidia is trying to avoid the same tragic fate. Nvidia has always observed their planned release dates so we should expect their next generation of video cards to hit the stores at Christmas 1999. Of course TNT3/NV10 is still in the planning stage and much can change this summer, but it should be quite sure TNT3/NV10 will not suffer the same sad fate as Command and Conquer Tibirian Sun; TNT3/NV10 will be available in late 1999 or at most early 2000. Photo-realistic graphics Lately there has been much talk about photo-realistic graphic in games and applications but none of the major 3D accelerators on market have yet achieved this feature. Nevertheless Nvidia has revealed TNT3/NV10 will have support for transform- and lightning acceleration - geometry acceleration - and thereby should photo-realistic graphic be possible on even the modest computers. To understand the geometry acceleration technology a basic knowledge of the 3D Graphics Pipeline is required. The 3D Graphics Pipeline is the term assigned to the different step application data must go through to display 3D-graphics on a 2D-monitor. The process consists of two different stages - the geometry stage and the rendering stage - with each stage consisting of several steps. The first stage is the geometry stage where floating-point mathematical calculations are translated into polygons - usually triangles - and arranged in a way giving us the feeling we are seeing 3D-objects on a 2D-monitor (however these figures have no indication of being realistic as they have no colour or anything else than the shape of triangles). The arranging of these polygons may involve clipping or discarding of polygons, which does not lie within the user's viewpoint. Next the calculations determining lighting characteristics of the scene, including defining whether light sources are ambient or directional or if the surfaces are diffusive or reflective, are performed. Finally a definition of the polygon view in all three dimensions are calculated and stored for use in the rendering stage. The movement through the geometry pipeline depends upon maximum utilisation of available processing power and must be done with as little waiting as possible. Extensive use of FIFO (First In First Out) insures that no stages of the pipeline keeps any other stage waiting. In order to prevent the geometry calculations from becoming a bottleneck in the 3D Graphics Pipeline a processor with a strong Floating-Point Unit (FPU) is required. The second stage is the rendering stage where the polygon information is actually converted to pixels for display. This procedure may involve shading, texture maps and special effects to the polygon information provided by the geometry stage. These calculations complete the image generated in the geometry stage and give it a high degree of realism. In contrast to the geometry stage this procedure is memory-intensive and thereby dependent on how much available memory you have (texturing buffers). When both stages are completed data can be shown on screen. Even though it may seem like a long and difficult procedure it is done in less than a second: When we are talking about frame rates in games it is actually a measuring of how many times per second each frame is updated through the geometry and the rendering stage per second. The 3D graphics pipeline used to be handled by the CPU alone and that created, as the processor also has to control all hardware devices, operating system and application, poor and slow graphics. When the first 3D accelerators hit the market the video card accelerated the rendering stage in hardware and the CPU handled geometry stage. This way of acceleration has through the last five years been the best option for showing graphics. However games and applications have become more and more detailed and run in higher resolutions so the processor has become a bottleneck in the system hence that the geometry stage is very processor depending. To relieve the pressure on the processor several accelerators for the professional market have support for hardware acceleration in both the geometry and rendering stage. Because the processor has been locked-down to only run hardware devices, operating system and applications the extra power available can be used to helping the video card creating smoother and more detailed graphics. The result; photo-realistic graphics generated on modest processors thanks to an unbelievable fast and technical sophisticated video accelerator. Geometry acceleration can be used in applications and games demanding high performance for on-screen dynamics such as 3D animations. The technology can be used in A/E/C, MCAD, Visual Simulation, Digital Content Creation and 3D games. The results you can achieve with geometry acceleration are incredible. The 3D benchmark suite Viewperf is specially optimised to take advantage of geometry acceleration and the results speak for themselves (higher is better): Viewperf Benchmark for RealiZm II 3D Normal acceleration Geometry acceleration Light-01 1.279 2.198 DX-03 9.32 21.04 AWadva-01 12.98 30.38 DRV-04 10.02 16.06 CDRS-03 71 135 The grand question is though; will Nvidia be able to integrate both geometry and rendering acceleration in the one and same product? 3Dfx has said they will use the same technology but rumours claim that it turned out not to become a reality; it was much to complex a process. More and faster RAM RAM on 3D accelerators is only used for the rendering stage when polygon information is stored in texture buffers for display. Ordinary accelerators, like Voodoo3 and TNT2, comes with 32MB of RAM while professional CAD-accelerators, like 3Dlabs Oxygen, comes with up to 256MB. Generally speaking the more realistic and advanced graphics you want rendered the larger texture buffers you need and since TNT3/NV10 will have photo-realistic rendering 64MB will probably be the minimum if not a 16/32MB OEM-version is released. Like Voodoo3 TNT3/NV10 will presumably be released in four different versions; an OEM-version with 16/32MB of RAM, a standard with 64MB, a standard ultra with 128MB, a professional with 192MB and an professional Special Edition with 256MB of RAM. Still the models are just guessing and could be so very wrong - on the other hand 3Dfx's Voodoo4 will come with 128MB of RAM and it is doubly Nvidia will not try to beat this level. Rumours claim that Nvidia will use a new and faster type of RAM than the latest 133MHz SD-RAM modules. Still it is impossible to know what the new type will be but RAMBUS, DRAM and 166MHz SD-RAM modules could be possible candidates. 15,000,000 transistors Considering TNT3/NV10 will have geometry acceleration a strong processor, in both its speed (MHz) and its mathematical calculations (Floating-Point Unit), is required to achieve photo-realistic image-quality. According to Nvidia's roadmap TNT3/NV10 will have 15,000,000 transistors on the same chip; the same amount as Intel's planned 800MHz Merced processor. Hence the massive numbers of transistors the micron will surely drop from 0.25 to 0.18 or even 0.15 allowing much higher clock rates. The 0.18-micron processors (Advanced Micro Device's K7) on today's market run at 650MHz but it is very doubtful that a video card could reach this thrilling clock rate. What seems more accurate would be a clock rate at 250MHz or even 300MHz - about 40% more than the fastest 3D-accelerators (TNT2 ultra can go up 183MHz) have today. Many transistors, low micron rate and high clock speed; all in all these generates a significant amount of heat and TNT3/NV10 will probably have both heatsinks and coolers not to overheat. It is doubtful that the next generation of video cards will be as overclocking-friendly as their progenitor owing to their technology. Conclusion The specifications available today indicates TNT3/NV10 will shake the market when it is released. All the specifications are of course still very thin and can be changed every second. However it is doubtful they actually will change that much as the technology is moving in that direction. Even though TNT3/NV10 should come with less RAM, lower clock speed, higher micron, less cooling and so on it is still a very interesting product, which deserves as much attention as possible.