Thanks to our well informed and reliable friends, Fudzilla has learned that you can expect as much as 24 TFlops of half precision performance and 12 TFlops of single precision performance. Both half precision and single precision numbers are higher than Nvidia scores with Pascal P100, its highest end Tesla chip.
Earlier we learned that HBM 2 won’t be really available before the end of the year and this is one of the main reasons why the Vega 10 had to be pushed to 2017. Nvidia announced the Tesla P100 but it always claimed that it plans to ship it in Q4 2016.
Bear in mind that the Titan X uses GDDR5X memory and scores significantly lower than the Tesla P100 and Titan X uses the GP102 chip that packs 12 billion transitions, while the P100 has 15.3 billion.
It is safe to say that Vega 10 will be rather a big chip and it gives you the idea that in case that Volta is not ready to launch in time to compete with Vega 10, Nvidia could have another Titan card based on the P100 with HBM 2 memory up its sleeves.
AMD is working hard to return on all fronts. The Polaris generation is a decent competitor in the mainstream market but it cannot really touch Nvidia in the high end sector. This might change with Vega 10 and it will make AMD / RTG a bit more competitive.
The Vega 10 comes with 16GB HBM 2 memory.