• On TechRepublic: Why VISTA HATERS will love Windows 7
advertisement

GeForce4 Go: power and glory
By Jon L. Jacobi
(2/13/02)

Yes, I know I told you in my last column that I'd be describing my 802.11b home-networking experience this week. What can I say? A more timely topic arose: Nvidia announced its GeForce4 Go mobile GPU last week, and I just couldn't 4-go the pleasure of taking you on a tour.

Now that your groans have subsided, there are two reasons why notebook users will be excited about the GeForce4 Go: graphics quality and power savings. The first is actually a by-product of vastly increased speed; Nvidia's new chip can render high-resolution, antialiased images at game-playable frame rates--a first for a notebook GPU. And while you wouldn't expect a more powerful GPU to save battery life, it's also supposed to do just that by running more efficiently than its predecessor.

Speed to burn
The GeForce4 Go comes in flavors ranging from a low-end G410 with 16MB of memory and 1.6GB per second of bandwidth to a high-end G440 with 64MB of memory and 8GB per second of bandwidth. I've seen only the G440 in action (in the newly announced Toshiba Satellite 5005-S507), so I can't say how much of a performance hit you'll take with lesser implementations. Notebook gamers should be delighted with the G440, however, because it can propel Quake at frame rates as high as 140.

Now for the bad news: Most of that power's wasted on a notebook because the fastest LCDs have slothful response times of 25ms or so. Obviously, when you divide that into a second, you get a screen redraw rate of no more than 40 per second. Forget those 60Hz or 70Hz ratings you see in the specs for LCDs; that's the type of signal they can sync to (another great marketing practice from the industry that brought you 19-inch displays with an 18-inch viewable area). The bottom line: You'll never see those 140 frames per second (fps) except with an external CRT.

Although the GeForce4 Go can't realize its full potential on a notebook LCD, the GPU's increased horsepower still allows it to render antialiased images at much better frame rates than before: 70fps or so on Quake with the less sophisticated 2X or Quincunx antialiasing enabled and about 50fps with full-on 4X antialiasing. If you read my last column about Microsoft's ClearType, you might remember that antialiasing smoothes jagged diagonal edges by shading the adjoining pixels in tones midway between the two divergent colors that are creating the problem. To smooth an entire screen takes incredible amounts of computation, but the results are worth it; antialiased games appear much more realistic and pleasing to the eye.

Power play
Nvidia has also broadened the implementation of its PowerMizer power-saving technology (and, thus, caught up with ATI's Mobility Radeon, which was the first to implement hardware-level power-savings features). Although the GeForce4 Go has more transistors (29 million vs. 17 million on the GeForce2 Go) and faster core clocks (a 220MHz GPU and a 220MHz memory interface, compared to the GeForce2 Go's 143MHz and 166MHz), Nvidia claims the GeForce4 Go sips less power. With the GeForce2 Go, PowerMizer simply lowered the chip's clock speed when less graphics-processing power was needed. On the GeForce4 Go, entire sections of the chip can be shut down. If you aren't using the 3D capabilities, they're shut off; if you aren't playing a DVD or outputting to a second display, the MPEG-2 decoding and the TwinView sections take a nap.

Offloading MPEG-2 decoding tasks (for DVD playback) from the CPU helps save power as well. The GeForce4 Go takes over YUV-RGB, motion compensation, and IDCT in the same way that rival ATI's chips do. In addition, the GeForce4 Go handles inverse quantization.

Easy does it
Nvidia has also made the GeForce4 Go significantly easier for notebook designers to implement; the new GPU is fabricated on a .15-micron (the distance between transistors) process, so it's much smaller than the .18-micron GeForce2 Go--another reasons it draws less power. The company has taken advantage of the die shrink to offer both the GPU and the memory on a mobile 4X AGP package (MAP) that measures the same 31mm by 31mm in size as the GeForce2 Go Performance measured on its own (the GeForce2 Go 200 and 100 were 23mm by 23mm). Nvidia says that it will offer the same MAP for future GPUs, so vendors won't have to redesign their notebooks every time Nvidia releases a new chip.

The GeForce4 Go's advances in speed, power savings, and design efficiency should mean faster, longer-lasting, and cheaper notebooks that still pack plenty of punch. As always, I'll reserve final judgment until I see it in a shipping system. Meanwhile, in my next column, I'll be relating my observations of PalmSource, last week's Palm PDA conference in San Jose. Then--I promise--will come my tale of 802.11b.



Next steps


Jon L. Jacobi is a contributing editor for CNET Hardware and a freelance writer and programmer based in San Francisco.