Out of 60 possible
Nvidia confirmed that there are 56 streaming multiprocessor (SM) units each having 64 CUDA cores enabled on the GP100 GPU. This results in 3,584 cores on the Tesla P100 GPU. However, this leaves 4 SM cores disabled as the total number of SM’s is 60.
Shipping values priced between $608 and $810 USD per unit
Any time a new batch of GPUs are ready to be sent from a manufacturing facility overseas to a supply partner for testing and metrics, there are always debugging and testing parts sent ahead of time to ensure that silicon is quickly brought to market without suffering silicon respin and unforeseen manufacturing delays.
Updatad: The next generation
Micron has promised to launch a new kind of memory that is twice the speed of mainstreams GDDR5 . This will be released during 2016 and will be company's answer to HBM.
Will add an Intel LTE-A modem
LG has been trying to make and launch a decent inhouse SoC. The first attempt was a limited success and now the company plans a successor – the LG Nuclun 2 to launch in 2016.
More compact process
TSMC has quietly rolled out a compact, low-power version of its 16nm FinFET fabrication process.
The future of graphics
High Bandwidth Memory (HBM) will define the future of graphics and it looks like Samsung will be ready to release products using the technology soon.
16FF coming soon, 10nm in late 2016
TSMC has shed more light on its FinFET plans, saying its 16nm and 10nm nodes are on track.
Moore’s Law alive and kicking
TSMC is planning to qualify its 10nm node by the end of the year, with volume production slated for 2017.
16nm process might be a better choice than Samsung’s 14nm
TSMC is reportedly getting the majority of Apple A9 orders, which would be a big coup for the company.
To boost 16nm FinFet and beyond
Reuters carried a rather interesting announcement about TSMC, the maker of most SoCs and GPUs we care about.