Here it is, the Next Generation Mobile GPU, or just NV17M, a new child of NVIDIA. The most of the aspects which will be addressed today will concern not only the mobile solution but also a budget one, NV17, which is to be announced a bit later. Let me outline the current situation around the NV17M.

1. The new firmware technology Powermizer is designed to provide more effective power consumption, as well as a possibility to adjust a performance/power ratio of the graphics accelerator integrated into a notebook. If you manually decrease a limit of the peak chip's performance (available settings are some 1/2 and 1/3 of the maximum value) the power consumed by the chip at the highest load (e.g. in a game) will be almost halved. Here is how the new tab looks like in the drivers:
   
   Besides, the chip is able to decrease the frequency dynamically, depending on the instantaneous value of the load (even down to 16 MHz in the Idle mode of Windows) and maintain the respective supply voltage. Consumed power is linearly dependent on the frequency and is proportional to the voltage squared, that is why it is quite a baffling problem to make a chip being able to change both values. At last, the chip can "freeze" for a short time if there is "nothing to do" (the frequency is equal to 0 and the power consumed is also almost zero).
2. MPEG2 (DVD) Hardware Decoding. It is a good response to the hardware IDCT conversion which has been living in ATI products for a long time already. The NV17M implements IDCT and IQ (Inverse Quantization -ATI can't fulfill this initial stage of decoding) on the hardware level, as well as Motion Compensation and Color Space Conversion. It is stated that the utilization of the processor decoding a DVD video stream is 3 times lower now (from 45% to 15%), which is a large commitment to the overall power consumption of a notebook.
3. I'm glad there is an integrated TV-Out controller and other outputs: 1*CRT + 1*DVI + 1*TFT controller for an LCD display of a notebook.
4. The NV17M combines the technologies of two chip generations (!):
   • 0.13 fab process - NV25
   • MSAA (including Qincunx) - NV20/25
   • ZCompress and ZCull HSR - NV20/NV25
   • Hardware DVD reproduction - NV25
   • T&L Second Generation - NV11/NV15 (it is exactly GeForce 2 *, as vertex shaders, probably, consume too much).
5. 250 MHz clock speed and a normal 128bit DDR memory bus are promised.
6. 3 versions of the chips are expected:
   • Usual version with an external memory bus (128 and 64bit memory bus optionally)
   • MAP (Mobile AGP Package) - with 16 or 32 MBytes integrated DDR (64bit bus). A die of the accelerator's core and two memory dies are mounted on the chip case, over the microboard.
     
   • MAP version with 64 MBytes integrated DDR (128bit bus). A die of the accelerator's core (below the microboard) and four memory dies are mounted on the chip case.
     
7. The declared memory bandwidth is 8 GBytes/s, and the effective one is 9.6 GB/s. It means that there are no compromises as in the NV11: we have a normal 128bit DDR bus and a normal internal memory controller on the chip.
8. The most interesting thing is a chip configuration. Reportedly, it has some 4 Gigapixel effective AA fillrate; i.e. taking into account 4X MSAA the maximum fillrate is 1 Gigapixel, hence 4 pixel pipelines. But now the site shows different figures (and I consider them more credible) - 500 Mpixel/s of a true maximum fillrate. It means that like in the NV11 we have two pixel pipelines with two texture modules on each. Will it be possible to accumulate the results for combination of 4 textures per clock (like in the NV20)? We will do our best to find out it.
   An interesting marketing argument is that a powerful, T&L equipped chip is able to save energy when implementing the tasks which earlier made the central processor consume a great deal of power. This is a work on the main errors of the GeForce 2 Go. The performance gain is impressing: a 1.5 or even double lead on the closest competitor from ATI (M7) in the majority of tests. I have an ocean of guesses about parameters of the future non-mobile NV17 family. And now there is no information concerning real effectiveness of energy-saving technologies and typical operating time of notebooks working in intensive 3D applications. We shall see what we shall see...

Write a comment below. No registration needed!