Compute Unit Counts Of AMD Polaris 10 & 11, and Vega 10 GPUs Surface
Updated Apr 11, 2016 @ 14:49: Added compute unit count for AMD Vega 10, as well.
AMD is all set to give a tough competition to Nvidia by releasing its next generation Polaris and Vega GPUs. The chip maker will release GPUs based on its Polaris architecture this year, while GPUs based on Vega will be released sometime in 2017, and will have double the performance compared to Polaris.
Assuming that the compute unit of the Polaris architecture consists of 64 stream processors, then do the math, and it could be possible that, Polaris 11 could feature 1024 stream processors over 16 CUs, while Polaris 10 code named “Ellesmere” could feature 2304 stream processors across 36 CUs, and Vega 10 has up to 4096 stream processors, across 64 CUs.
Polaris 11 is the successor to this generation`s “Curacao” silicon, which is used in mid-ranged graphics card. It is expected that it would be capable of having 4GB of GDDR5 memory, and 128 bit GDDR5 memory interface.
While Polaris 10 is the successor of current generation`s Tonga silicon, which powers the high-end graphics cards, and could feature up to 8GB of GDDR5(X) memory.
AMD is already claiming 2.5 times more performance per watt increase over the Polaris architecture. The above mentioned Polaris GPUs could be released in mid-2016.
Vega 10 is expected to succeed Fiji, and will take advantage of next gen High Bandwidth memory (HBM2), and feature up to 16GB of GDDR5(X) memory. Vega architecture is set to be released in sometime 2017, while
So what are your thoughts on this? Do you think AMD could give Nvidia a tough time, through its “Polaris” and “Vega” GPUs? Do tell us in the comments section below.