NVIDIA GeForce4 4200 Go: Bringing mobile gaming to new heights

The NV28M is the mobile variant to the NV28, just as the NV17M (GeForce4 440 Go) was the mobile variant to the NV17 (GeForce4 MX). What does being a mobile variant of the chip entail? In the case of the NV28M it means that the chip maintains 100% of the desktop NV28's hardware but incorporates power saving technology. The same 3D pipeline with dual vertex shaders; the same LMA II; the same AGP 8x support. Added to the core is PowerMizer and the ability to run at lower clock speeds and save power when possible, which we will discuss more in a minute. For now keep in mind that the GeForce4 4200 Go (NV28M) is based on the current generation highest performing desktop offering from a company, tge first time this has happened in the mobile graphics industry. Historically, mobile chips have been based off of mainstream parts as shown by the GeForce4 4x0 Go's base in the GeForce4 MX series and the ATI Mobility Radeon 9000's base in the Radeon 9000 cards. The GeForce4 4200 Go changes that trend and makes a huge statement as to where mobile graphics is going. There are some problems that arising from doing this, especially in the case of the NV28 specifically. Lets talk about some of the limitations of turning the NV28 into a mobile graphics processor.

Unlike the NV17 core found in the GeForce4 MX cards, the NV28 core was not designed specifically with a mobile variant in mind. As a result, the NV28M does not have the dedicated laptop functionality that the NV17 offered. The two big things that the NV28M is missing that the NV17M has are an integrated TV encoder (for TV-output) and an integrated LVDS (for driving the laptop display). This means that notebook manufacturers have to turn to at least an external LVDS and possibly an external TV encoder, making designing notebooks that use the GeForce4 4200 Go a bit more difficult.

The lack of integrated display hardware is not the only stumbling block in the GeForce4 4200 Go's way. As a desktop chip, emphasis on designing the NV28 did not lie in reducing heat and power consumption. Power saving hardware, such as PowerMizer, has the ability to reduce the bottom line power consumption of a chip but the peak power characteristics of a mobile variant remain fairly similar to that of the desktop one. This results in very high peak power consumption and thermal characteristics in the NV28M under situations when the chip is stressed, such as when rendering a 3D scene.

NVIDIA would not disclose the maximum power draw of the NV28M but they did mention that it was high. They were able to get the voltage down to 1.25 volts when at the maximum performance setting (the same as the Mobility Radeon 9000) but apparently the power draw of the chip is still desktop-like.

Another compromise that NVIDIA had to make when turning the desktop NV28 into the NV28M has to do with the chip packaging. The NV28 is composed of many more transistors than the NV17. As a result, both the chip and package of the NV28 is larger than that of the NV17. As a result, the GeForce4 440 Go and the GeForce4 4200 Go are different sizes. One characteristic of the entire GeForce4 4x0 Go line (with exception to the GeForce4 440 Go) is that all the chips are pin compatible. This means that all a manufacturer has to do to go from producing a notebook with a the high-end GeForce4 460 Go to the lower-end GeForce4 420 Go would be to pop out the old chip and pop in the new one. NVIDIA planed to expand the 31x31 BGA package size to future mobile NVIDIA products but obviously the NV28M marks a deviation from this path. Not a big deal for now (let's hope future mobile NVIDIA parts revert to this design), but it does make getting the GeForce4 4200 Go into a laptop a bit more difficult.

Likely due in large part to the GeForce4 4200 Go's thermal requirements, the NV28M will not come in a MAP package. You may recall from our GeForce4 4x0 Go review that the previous Go chips were available in discrete packages (which use external memory) as well as what NVIDIA called mobile AGP packages, or MAP packages (which include the GPU core on the same package). The way this was done in the past was by mounting the GPU core and the memory packages on the same substrate, thus requiring less physical space for the full video setup.

The lack of a MAP GeForce4 4200 Go solution turns out to actually be one of the more minor thorns in the chip's side. As it turns out, not many MAP solutions have made their way into market (this is true for ATI as well) as OEMs have been opting for cheaper external memory. In fact, the most appealing market for a GPU and memory combination chip is the thin-and-light and ultra-portable market, where thermal and power requirements keep the highest performing chips out.

Obviously, as a mobile part, the GeForce4 4200 Go runs at lower core and memory clock speeds compared to the desktop GeForce4 Ti 4200 AGP 8x. The desktop version of the chip (the NV28 in a 64MB configuration like the configuration of the GeForce4 4200 Go) runs at a 250MHz core and a 500MHz (250MHz DDR) memory bus. In contrast, the GeForce4 4200 Go has the ability to run at a maximum speed of 200MHz core and 400MHz (200MHz DDR) memory. Expect to find chips performing at this pace only in desknote type systems. The majority of notebooks, including the notebook version that we were given to test on, will be running at around the 175MHz core mark and a 400MHz (200MHz DDR) memory bus (ours ran at 175MHz/200MHz DDR). The reason for this: the GeForce4 4200 Go produces a lot of heat and draws a lot of power running at 200/200MHz DDR.