PDA

View Full Version : Hardware Features and Video Cards



WILL
26-11-2005, 08:07 PM
Well this question stems from a long argument that a friend often brings up about computing power and video card features. In short he simply refuses to acknowlage that video card hardware acceleration features have enough of an impact on game performance on newer and more currently released games.

On my laptop, which is getting rather old and it's video card which is a few generations below the current generation of video card, has a ATI Rage Mobility M1 (2x) AGP (with 8MB of physical video RAM, AGP expandable). Now I'd call that ancient as far as video processing goes. :) And I cannot count the amount of missing hardware features for 3D graphics processing from todays cards and games.

So, to further my interestin the topic. What are the major 3D acceleration hardware features of today's video cards?

Sascha Willems
26-11-2005, 09:15 PM
Well, in terms of actually what is being accelerated by a graphics card (rather than the CPU) not very much has changed since the last big step, the GeForce-series. Since that, all graphics cards accelerate Transformation and Lighting (Hardware T&L), older cards (up to GeForce2 or 3) have a dedicated HW T&L-Unit (hardwirde), newer cards (that are capable of shaders) "emulate" HW T&L with their shaderunits.

So a shiny new GeForce 7800GTX won't accelerate more stuff than an old GeForce1 (which btw. was the most expensive card I've ever bought, over 700 DM back then, I stopped from buying high-end cards).

But what actually changed is programmability, due to the shaders. Whereas the hardwired T&L-unit of a GeForce1 does all the transformation and lighting for you, you had no way of changing how those are done. But with newer cards you have vertex- and fragmentshaders (also misleadingly called "pixelshader" under D3D) that allow you to actually tell the graphicscard what to do with your vertices and/or fragments.

So, modern graphics cards don't accelerate much more than i.e. the original GeForce cards from back 1999. In the end, they're just gone much much faster with smaller production techniques (like ATI's 90nm for current GPUs, compared to 220nm for a GeForce1), that allow the manufacturer to use much faster clocks for memory and the GPU, and also allow the manufacturer to add more pipelines (each pipeline calculates one texel at a time) and other stuff to the chip, that in the end do nothing than making the graphics card faster.
And the new features of modern graphics cards (like new shader models, etc.) are mostly for eye-candy. Shaders don't make things run faster, they make them easier to program (often making things possible that without them arent) and therefore add eye-candy, nothing more.

Just a quick comparison :

NVidia GeForce1 (NV10) from 1999 :
20 million transistors, 4 pixel pipelines, 15m triangles/s, 480 MPixels/s
Clocks : Core = 120 MHz, Memory = 166 MHz (DDR->333Mhz)

NVidia GeForce 7800 (G70) from 2005 :
300 million transistors, 24 pixel pipelines, ~300 triangles/s, 6880 MPixels/s
Clocks : Core = 430 MHz, Memory = 1.2 GHz

WILL
29-11-2005, 12:33 AM
Hmm... ok. Well I'm not even in the GeForce generation of video cards yet. :lol: I'm maybe, maybe within the TNT era though.


At any rate, I would still like to have a list of the hardware enhanced 3D features in a modern video card. I mean what effects are hardware accelerated that can be listed?

Sascha Willems
30-11-2005, 01:35 PM
At any rate, I would still like to have a list of the hardware enhanced 3D features in a modern video card. I mean what effects are hardware accelerated that can be listed?

Hmm...that's almost impossible to answer. Due to Shaders, you can nowadays "accelerate" any technique that does stuff on a per-fragment (or per-pixel, for all D3D-users) basis.
Some of those techniques are :
Bump Mapping
With all it's different implementations (DOT3, Normal mapping, Parallax Mapping).

Per-Pixel Lighting
With all lighting equations that one can imagine, since you can nowadays do your own calculations in the shader. Modern GPUs are even fast enough to do stuff like Dynamic Ambient Occlusion in realtime.

Post-Scene effects
Maybe one of the biggest things that is possible with modern cards. This includes stuff like Bloom, Glow, Heat-Haze, real HDR (High-Dynamic-Range Lighting), weird camera-effects (like fish-eye), Video-Effects, etc..

And one of the very recent features of modern cards I totally forgot are floating-point textures, which allow you to store much more precision into a texture. Those are mainly used for advanced HDR-effects, though they can be also used for other stuff. For HDR see this demo (http://www.daionet.gr.jp/~masa/rthdribl/) (Requires a PixelShader 2.0 card or better).

Many other effects, like dense grass-layers or dense foiliage we're already possible on old hardware, but that hardware was just too slow to do that.