|
|
|||
 |
|||
| Erwin's Pages Central > Articles > Hardware Articles > Geforce 3 | |||
   
  
 |
 nVidia GeForce 3: Is it worth its price? It's been some time now since nVidia released it's new NV20 chipset, of GeForce 3. And since then (Spring 2001), the price of the Geforce 3 cards went down, but not sufficiently to justify it's, by home users' standards, high price (+- 450 € or 400 USD). What makes this chipset so special? Let's take a dive in the heart of the revolutionary technology behind nVidia's latest graphics powerhouse. Hard numbers Core clock speed: 200MHz Memory clock speed: 230MHz Memory type: 4.5 ns DDR SDRAM Memory bandwidth: 7.32GB/second Rendering engine: 4 pixel pipelines, 2 texture units each Fill rate: 800 million pixels per second; 3.2 billion samples per second Taking a first glimpse on these numbers reveil a specification not that different from the long-time GPU ruler, GeForce 2 Ultra. But it's the last number that elevates the Geforce 3. The sampling fill rate though, important for rendering crowded high detail scenes, is about double the of the NV15 or GeForce 2 GTS/Ultra. But it are not these hard numbers that will elevate this chipset from the current GPU's. It's the new technology and rendering approach that'll make the difference. Maxing current games Two technologies will run games today as fast as never before. "Lightspeed memory architecture". Sounds cool enough to get Star Wars and Star Trek fans jumping for joy. This new approach to alleviate the infamous "memory bottleneck" bypasses the need to place faster and more pricey memory modules. This is because for instance in the past small object would also take up this excessive amount of data space, while it actually doesn't need it. The idea is to split the 128-bit memory transfer pad into 4 smaller 32-bit transferring pads, making the stream of data between chipset and cpu more efficient. Geforce 2 and Voodoo 5 introduced hardware controlled anti-aliasing. Both videocards had a more or less same approach to cleaning up sawtooth edges on polygon objects. An extra amount of pixels is rendered, filled and mixed between the existing pixels. The amount of extra pixels are, in practice, the 2x and 4x AA modes. The problem wit this wonderful system though is apparent. More pixels than necessary are rendered, which made systems extremely stressed, particularly with the GeForce 2 MX, which had to perform the same tasks as it's bigger brother but with less memory and half the pixel pipelines. GeForce 3, as you can tell from the hard numbers, has the same rendering engines and memory bandwith as the GeForce 2 Ultra. But what GeForce 3 does is using pixels from a same texture and copying those pixels to the "gaps" required. So instead of calculating all anti-aliasing pixels, only a few samples are rendered and mixed. This method is of course much more efficient. The GeForce 3 also has a unique new anti-aliased method, using quincunx. While explaining this technique would be way to technical, the bottem line is that this method results in virtually no quality loss at an anti-aliasing rate similar to 2X sampling rate. In short, nVidia calls all these anti-aliasing methods "HRAA" or High Resolution Anti-Aliasing. Revolutionizing the way we look at a 3d game But of course, time doesn't stand still and the GeForce 3 is foremost the pioneer of a new graphcis' era. Well, this statement is perhaps overrated, but still major changes will take place. Again, 2 buzzwords come forward: nFinite FX and DirectX 8. It was about 1,5 years ago the GeForce bomb impacted the TNT2/G400 landscape. While we were all happy with the dazzling high resolutions and dito framerates on then contemporary games, GeForce vowed to revolutionize the graphics sector by claiming to take over geometrical and lighting effects from the cpu. T&L was the blinking neonlight shining loudly into our already blinded retinas. Software support though was somewhat slim, and that because of one distinct reason. The instrucionset of the T&L engine was set, and could not be altered by programmers. So, the software had to be manipulated to be able to be rendered by the T&L toolset. Of course, many programmers were reluctant to do so. nFinite FX though can be manipulated by software, and so being much more flexible. The effects generated are also much revolutionary than what the T&L engine offered. nFinite Fx is actually a grouping of two engines: vertex and pixel shaders. Hey, I heard some of that when I bought my Geforce 2 card! Indeed nothing seems to be new here, but let me stress again that these effects are now directly can be manipulated and that it indeed is an integral part of the GeForce 3 (rather than a nice extra like the T&L was on GF1 and GF2). Vertex shaders for example wil create CGI quality facial animations, giving programmers the possibility to create very fluent skeletal and skinning movements. Also, this technology will adhere to gradual deformation. For instance, the hot air coming from hot-running engines at the start of a race. Can you imagine a Grand Prix 3 with GeForce 3 graphics? I know I can... (drop of saliva hitting the ground now) Also, implementation of all these fabulous bells & whistles is virtually assured since Microsoft's new API DirectX 8 very much supports the GeForce 3. DirectX versions usually also laud in new graphical era's: Direct X 5 was Voodoo 1&2, Direct X 6 was TNT/TNT2/G400, Direct X 7 is GF1/GF2 and Direct X 8 is Geforce 3 and other new chipsets (like the new ATI Radeon, read more about later in this article). So, software support should be guaranteed since the hardware interaction (nFinite FX) and it's API (DX8) are almost made to make the GeForce 3 the industry standard. This last comment though also has a flipside for nVidia. The competiton and resulting pricewars With 3DFX out of the way by the almost overly pro-active takeover of the latter by nVidia itself, the only real rival with competing technology is the new ATI Radeon. ATI surprised many with it's fine Radeon card. Sharply priced in the zone of the GeForce 2 MX and boasting great performance, ATI finally shook off it's former "All-In-One" business card solutions and turned to the real graphics demanders, us gamers. The new Radeon is kept quite under the shades, but at least one word has been released. Truform. Truform is ATI's answer to the Geforce 3's nFiniteFX. In fact, it's even more revolutionary than it's rival. Truform is a technology that actually manipulates the triangular polygon forms themselves. Since the dawn of the "real 3d games" (Jedi Knight, Quake,..) 3d meant polygon models. These models were created from triangles. The more triangles, the smoother the resulting polyon. Problem is that perfectly round form are theoretically impossible. Quake 3 was the first pice of software that created "curved surfaces" with it's engine. The geometry detail determined the "curviness" of these surfaces. The lower the detail(=amount of polygons), the lower the curviness. Truform actually interprets all forms as being curved. This procedure is already put before the rendering engine, so every polygon data is a smooth form to the new ATI card. The method of this procedure is quite complicated (screw the mumbo-jumbo and just say what it does!) but suffice it to say that flat objects are actually bezier objects manipulateable along mutliple handles. Remember bezier curving in programs like Adobe Photoshop and Macromedia Flash? This is a straight line with 2 handles, which determine the curve. Truform allows multiple handles creating "true forms". The result is incredibly stunning facial animations and amazing rendering of any curved object. But since Truform is not an option but an essential part of the new Radeon, this mo will cause some possible problems with... you won't belive this...objects that actually have to be straight and pointy. Objects like guns, crates, etc. ATI says that this can indeed cause problems, but nothing a small registry patch can't fix. Because all this info is very preliminary, you should not be alerted though. Truform is an entirely new approach to rendering 3d scenes and promises to be a real nail on the coffin of nVidia's rain of new dazzling technologies. Furthermore, ATI is going to implement some GeForce 3 technologies since some of them are open. This is necessary if nVidia is to make any of an industry standard. So, this returns us to the GeForce 3. It's high price is justified, according to nVidia, because GeForce 3 really is a next-gen chip and is very hard to produce. Remember the stunning XBox graphics? It's actually powered by a GeForce 3 based engine! (Read more about XBox and consoles here). To make a comparison: GeForce 3 has more transistors than Intel's new cpu flagship: the Pentium 4. Get it now? But still, the current price, 450 € or 400 USD is very high for a video card. For the same amount of dough, you can get a hefty TBird (1.33) plus motherboard. The wait is for the release of GeForce 3 accelerated games and ATI new Radeon. Unreal 2, Soldier of Fortune 2, Doom 3,... games scheduled for Winter 2001 or 2002 will fully take advantage of GeForce 3's options. And because the new Radeon is scheduled for (call it coincidence) late 2001. Higher demand for better graphics cards because of the software and fierce competion will definitly lower prices. Don't you just love free trade? (Genua protestors please don't hack me ;-) One more thing, if you wish to know the performance of the GeForce 3 on current games in raw benchmark numbers, log on to Tom's Hardware. A great hardware site specialized in any piece of high-end software. (c) 2001 Erwin's Pages Erwin "Reaper" Husin Links > Tom's Hardware > nVidia > ATI |
      }  |
|
| (c) 2002 Erwin's Pages All Rights reserved. |
|
||