AMD speaking at an investors event organized by Deutsche Bank, announced that their next-generation “Vega” GPUs will be launched in the first half of 2017. These will be the first high-end parts from AMD in close to two years. Vega will bring with it significant performance/watt refinements over the current “Polaris” architecture.
AMD will start the Vega lineup with the Vega 10. It will be an upper-performance segment part that is positioned to sit between NVIDIA’s GP104 and GP102. The chip is expected to have 4096 stream processors, with up to 24 TFLOP/s 16-bit floating point performance. It will feature 8-16GB of HBM2 memory with up to 512GB/s memory bandwidth. The TDP rating should be somewhere around 225W.
Vega 11 will be a mid-range chip that will replace “Polaris 10” from the product stack. It will offer slightly higher performance and much better performance/watt.
Finally we will have Vega 20, which is a part we haven’t heard of before. It will be a die-shrink of Vega 10 down to the 7 nm GF9 process being developed by GlobalFoundries. It is expected to come much later. It will feature 4096 stream processors as well, but more than likely at higher clocks and up to 32GB of HBM2 memory cranking out 1 TB/s bandwidth. You will also get PCI-Express 4.0 support and typical board power of 150W.
AMD is expected to roll out the “Navi” architecture in 2019 so Vega will hold them out for the next two years or so. There is even talk of a dual-GPU Vega product rocking a pair of Vega 10 GPUs.