NVIDIA has just pre-announced some new mobile GPUs. This is a way of preparing us for their actual release, which will probably be in September, but also a way of deflecting criticism from the technological ground lost by the GeForce range. The GPUs, GT215, GT216 and GT218 will be engraved at 40 nm by TSMC and will support GDDR5 and DirectX 10.1. NVIDIA has therefore finally consented to modify its architecture. We still don’t know if this will be based on a derivative of the GT200 or GeForce 8s as NVIDIA has not given much detail on these GPUs which have been designed to replace the current entry and mid levels of the range for mobiles.
The GT215 and its 96 processing units. It remains to be seen if they will be set up as 4x24 (like the GT200) or 6x16 (GeForce 8 architecture).
Moving over to 40 nm has pushed NVIDIA to support GDDR5 as the chips were too small for a large memory bus. Even for the largest of these chips, the GT215, NVIDIA has had to go for a 128 bit memory bus. They therefore had to go with the latest memory technology so as to compensate for this as AMD has done for the Radeon HD 4770.
Here are the specs for the 5 new mobile GPUs:
As we don’t know what architecture will be used yet we cannot compare them precisely with current solutions compared to which for example they may have less texturing power relative to their processing power. Basing our calculations on this however as well as the memory bandwidth, we can say for example that the GeForce GTS 250M corresponds to the GeForce 9800M GT.
The main innovation will not therefore be to increase performance, which will probably remain the same as on the old range, but rather to give much lower energy consumption. So, while the GeForce 9800M GT had a TDP of 65W, it is 28W for the GeForce GTS 250M.
Desktop versions of these GPUs should be launched in September, NVIDIA’s partners having told us that they are starting to develop their PCBs now and that they should be ready within a couple of months.