Introduction
It's time for the annual graphics card upgrade. Nvidia today officially announced and released its new graphics card generation. Once again, Nvidia rings in the season with a faster chip that delivers more realistic graphics for enthusiasts. The real news, however, is the fact that graphics cards carrying the GeForce 7800 GTX label are available from the start.
Once a year, Nvidia and ATI renew their flagship product lines, offering especially the gaming crowd, a new way to improve their gaming experience. Nvidia is first this year with the GeForce 7800 GTX series, which the company claims is developed from the ground up.
From a performance perspective, the most visible characteristic of the new chip is that Nvidia achieved a significant boost in speed without going overboard with clock speed. The 7800 GTX received a modest increase over the GeForce 6800 Ultra, ending up at 430 MHz. Much of the performance increase instead is attributed to more pipelines which now stand at eight vertex shaders and 24 pixel pipes. As the previous generation, GTX boards come with a 256 Mbit GDDR3 memory interface for a total of 256 MB of memory that is now clocked at 600 MHz. The cards will be available only in PCI Express versions, offer SLI capability and outputs for DVI and HDTV. Power consumption remains at about 100 watts, according to Nvidia.
All this data ends up in significantly more graphics performance, giving Nvidia - at least temporarily - the desktop graphics crown, at least until ATI launches its R520 chip in the coming weeks. We'll have more on the actual characteristics along with plenty of our benchmark tests posted on our site tomorrow. Check for updates.
"Graphics is inherently a massively parallel application," said Nvidia's vice president of technical marketing, Tony Tamasi, in an interview with Tom's Hardware Guide. Graphics processor architecture, he explained, has the advantage of scaling up rendering performance linearly in correspondence to the number of pipelines added, because the process of rendering a scene can theoretically be delegated among an arbitrary number of pipes.
For CPU architects to improve their processors' performance, said Tamasi, "the most common "knob" they have is frequency. That's very expensive, from a power perspective. We don't have to do that. We can use architectural advancement. We've re-architected the pipelines to do more work per clock cycle. You will see roughly 2x from this generation to the last, and yet our frequency is approximately the same, and we have more pipelines."
One of the more subtle advances which Tamasi pointed out is called transparency anti-aliasing or "transparent supersampling." Here, components of a scene that you would not normally consider "objects"- little side details, such as blades of grass or chain-link fences - are rendered more sharply by applying a technique similar to the way ClearType makes anti-aliased text look sharper against a multi-color backdrop. These elements of detail and scenery, Tamasi explained, are rendered by taking essentially a 2D image and mapping it against a pair of adjoined triangles - a quad. The quad is essentially a square, but to keep all blades of grass and leaves on trees from looking square, selected pixels in the quad are masked out with transparency. But up to now, pixels in-between the edges of scenery items and the transparency has been jagged or stair-stepped. "Classic graphics processors don't anti-alias that intersection or that edge," said Tamasi, "which is why if you look in Far Cry or any number of games that go [outdoors], and you look at the grass or leaves or at a chain-link fence in Half-Life 2, when you get close to it, the edges have a jagged shape to it.
