According to Tech Times the chips have been sent from the TSMC fabrication plants to Nvidia's testing facilities in India.TSMC won the contract against Samsung in the bid to mass-produce the Pascal GPUs using the company's 16nm FinFET production process.
Nvidia's follow-up to Maxwell has some new technologies that include NVLink, HBM2 and Mixed Precision Support. At least four different Pascal GPUs are now being tested and seem to carry the same serial number of 699. These four different GPUs are the 699-2H403-0201-500, 699-1H400-0000-100, 699-1G411-0000-000, and the 699-12914-0071-100.
The first three are variants that belong to one basic board which has a 12914 label. ll of the GPUs will have similar per unit values, with the GP100 GPU being the largest and the most powerful among the Pascal graphics chips
The flagship GP100 GPU will also feature four 4-Hi HBM2 stacks and 16 GB of VRAM and 8-Hi stacks to reach up to 32 GB. This is clearly meant for top-of the-range professional SKUs.
The GP100 GPU apparently also has up to twice the performance per watt estimated improvement compared to Maxwell, DirectX 12 feature level 12_1 and higher, 17 billion transistors, 4096-bit memory bus interface, NVLink and half precision FP16 compute, which is two times the rate performed by a full precision FP32.
It would appear that the flagship GP100 GPU will be the successor to the GM200 GPU, which is under the bonnet of the GTX 980 Ti and the GTX Titan X.
More official details are expected to arrive at the company's GPU Technology Conference in early April.