First Look at Nvidia Pascal GP106 GPU: Die Size b/w 170-200mm2, 5.4B Transistors and 1280 CUDAs
At CES 2016, Nvidia CEO Jen-Hsun Huang presented the Drive PX 2, a Tegra-based autonomous driving oriented board with deep learning capabilities. The new package was said to be equipped with two Pascal GPUs; however, some of the eagle-eyed journalists out there took a closer look at the Drive PX 2 Huang was holding, and realized it had Maxwell chips on board instead of Pascal.
Fast forward three months, Huang took the stage at GTC 2016 and this time showcased the “real” Drive PX 2 module, which was definitely equipped with new Pascal GPUs. The module was however shown for a brief moment and that is at an angle that many journalists couldn’t take better pictures of it.
As it turns out, the area of this die is somewhere between 170-200mm2. The chip appears to be Pascal GP106 GPU as it is definitely smaller than GM206, which measures at 227mm2, but also bigger than ‘typical’ size for entry-level GPU.
Further, the rectangular nature of this GPU suggests that it could feature two graphical processing clusters right next to each other. If both of them have 5 streaming multiprocessors, then we’re looking at 5.4 billion transistors, 2 GPCs and 1280 CUDA cores – indicating a smaller bump in performance over GM206 chip.
As for the availability, Nvidia is expected to launch new Pascal based gaming and enthusiast graphics cards at Computex 2016 which runs from May 31 until June 4.