« August 2010 | Main | December 2010 »
September 3, 2010
What are discrete graphics, and why would you need them?
If you follow the mobile computing beat, you've probably come across the term "discrete graphics." What that generally means is a computer's graphics display capabilities that are a separate sub-system and not part of the motherboard or, more recently, processor. Why should you care?
Because as with almost everything else in life, one-size-fits-all only applies to a certain extent. Most computers take the one-size-fits-all approach, offering a set of features and performance that is good enough for most intended applications. Most, but not necessarily all. In graphics, that means that your standard mobile computer can handle all the usual functions such as communications, browsing, office apps and most media. However, the one-size-fits-all graphics capabilities that come with a system may struggle with more demanding applications such as advanced 3D graphics, CAD, GIS or other graphics-intensive tasks.
Discrete graphics, while uncommon in mobile systems, are a standard part of almost all desktop and many notebook computers. Starting with the earliest IBM PC, computers had separate graphics cards that handled the moving of pixels on a display. When users wanted more resolution on their IBM PCs to run Lotus 123 than standard 640 x 200 CGA, they popped in a Hercules graphics card that boosted res up all the way to 720 x 348 pixel and made charting faster and more impressive.
Over time, graphics "cards" became increasingly powerful graphics subsystems that provided a very significant boost in capabilities and performance. Top of the line graphics "cards" can cost as much or more than the rest of the computer combined, and they often have their own cooling sinks and fans. Graphics subsystems in notebooks are usually less conspicuous and they can even be integrated into boards, but they are still often differentiators between low-end and high-end versions of the same computer.
Now who needs "discrete" graphics? Not everyone. In the past, separate graphics subsystems or cards often offered higher resolution than standard built-in graphics. That's because old-style CRTs were able to support multiple resolutions. LCDs are different in that they are designed as a matrix of so and so many pixels, and that is the "native" resolution that results in the crispest picture. Most integrated graphics are more than capable of running a LCD in its native resolution, and since the LCD doesn't support higher resolutions, there is no need for a graphics card that can drive more pixels.
However, resolution isn't everything. Over time, computer graphics have evolved into a science with numerous standards and technologies. That's especially true in the areas of shading, rendering and manipulating 3D objects. This goes way beyond simply making pixels appear on the screen in a certain color and brightness. Games, for example, can require huge amounts of graphics computing power to make objects and 3D action look as lifelike as possible. 3D modeling and visualization likewise can require vast amounts of graphics computing power.
How does all of this affect mobile computing? Well, mobile systems cannot possibly accommodate top-of-the-line graphics for the same reason that they cannot provide top-of-the-line desktop performance: power and heat. So just like mobile systems must have a carefully designed balance between performance, weight, size, cost and battery life in their choice of processors, the same goes for their graphic sub-systems. Up to now, for the most part, the processor handled processing and its complementing chipset handled graphics. And the graphics part of those chipsets came from third parties that specialize on graphics, such as nVidia or ATI.
Up to recently, the situation was that most mobile systems made do with the "integrated" graphics capabilities inherent in their chipsets. These designs share system RAM and have some limitations. Some higher end or specialized devices had more powerful graphics to speed up certain applications.
With the advent of Intel's 2010 Core processors, the game changed somewhat because Intel integrated the GPU (graphics processing unit) right into the CPU package. Intel claims this improves efficiency, speed and stability while graphics chipmakers probably view it as an Intel land grab designed to assert even greater control without, however, being able to provide the graphics performance some customers require. Both sides have their points, but one thing hasn't changed: a separate "discrete" graphics sub-system will still outperform one-size-fits-all integrated graphics, and may also provide graphics functionality not included in a standard integrated system.
But what does it all mean in the real world?
RuggedPCReview.com recently had a chance to benchmark test two mobile computers that offered discrete graphics on top of whatever came integrated into the chipsets. One of them was the General Dynamics Itronix Tadpole Topaz, a high-end "COTS" (Commercial Off The Shelf) notebook designed primarily for military applications. It came with an nVidia GeForce 8600M GT graphics sub-system. The other was a Panasonic Toughbook 31 equipped with discrete ATI Radeon HD5650 graphics. Both machines ran Intel chips at a clock speed of 2.56GHz. However, while the Topaz uses an Intel Core 2 Duo T9400 processor without integrated graphics, the Toughbook employs an Intel Core i5-540M with integrated graphics that can either be turned on or off via BIOS settings.
The two machines are not directly comparable as they address somewhat different markets. However, when comparing the GD-Itronix Topaz with a GD-Itronix GD6000 that runs the same processor but does not have discrete graphics, the Topaz substantially outperformed the 6000 both in 2D and 3D graphics benchmarks, and absolutely blew it away in an OpenGL benchmark, by about a factor of 12:1. Now, OpenGL (Open Graphics Library) refers to a cross-language, cross-platform API for 2D and 3D graphics and is widely used in CAD, simulations and visualizations. If a customer has applications that use OpenGL code, then having OpenGL optimized graphics is absolutely mandatory.
While the Topaz used its nVidia graphics full-time, the Panasonic's discrete ATI graphics can be switched on and off. Why would one want to switch off presumably superior graphics? For the same reason why in a vehicle you wouldn't want four-wheel-drive or a turbo engaged all the time when you don't really need it. Such performance boosters for special purposes can have a very negative impact on fuel mileage, and that, for now, is no different with discrete graphics. Panasonic quotes up to 11 hours of battery life with the discrete ATI graphics off, but only 5 hours with them on. That is a big difference.
So what do discrete graphics get you in a modern Core i5 machine like the Toughbook 31? Not surprisingly, in day-to-day use, you probably would hardly ever notice the difference. But as soon as you get into 3D graphics and such, the ATI boosted performance by about a third, a very noticeable difference. The real payoff, again, comes in with OpenGL, where things happen more than four times as fast. That's the difference between barely tolerable and actual, real work.
Bottom line? For now at least, if your application requires speedy 3D graphics or includes a lot of OpenGL code, discrete graphics is almost a must. It's a bit of a dilemma as Intel is clearly trying to eliminate third party separate graphics and probably doesn't pay much more than lip service to easy integration of external GPUs. This uneasy relationship may or may not contribute to the steep drop in battery life with discrete graphics engaged, but if battery life is an issue, it's certainly good to be able to engage discrete graphics only when needed, or when the machine is plugged in.
Posted by conradb212 at 5:10 PM