TEAM RED PIXEL PUNCHERS
Where have we been and where are we headed?
2016 has been a year of graphical revolution. We finally transcended the confines of 28nm down to 16nm and 14nm. Big wins have been made for both Team Red and Team Green, as performance-per-watt figures spiked in celebration of this new transistor size. But, it hasn’t been a bed of roses for the guys at AMD. With an early entry into the mid-range battlefield, AMD clawed market share back from its green rival, but seven months on from the launch of the priceto-performance dominator that was Polaris and the RX 480, we’ve still seen nothing in regard to a true endgame from AMD. Nvidia is left to reap the rewards, dominating the field with the GeForce GTX 1080 and 1070, able to price itself wherever it pleases. So where are we heading?
The answer is Vega. A plethora of rumors surround the card, but all we know for sure is that it will be debuting in the first quarter of 2017. With it likely to be revealed at CES, by the time you read this, it should have been announced.
At a guess, we expect this new GPU architecture to still be based on GlobalFoundries’ 14nm FinFET manufacturing process, featuring twice as many compute units and shader cores as the RX 470—around 4,096 cores in total. In terms of memory, it’s a widely held belief that these cards should come packing HBM 2.0, and the delay is due to AMD trying to work the bugs out of the manufacturing process. With 16GB of HBM 2.0 being the top end, we can expect somewhere around 512Gb/s total memory bandwidth.
Alongside Vega 10, we’re expecting a cutdown variant of the new architecture, known as Vega 11. This should be the spiritual successor to the RX 480 and the original Polaris chips. Based on the same architecture as Vega 10, we expect it to come with around 30 percent fewer shaders and compute cores, and perhaps even a drop down into HBM 1.0 and GDDR5 memory, hopefully with a TDP of less than 75W on the lowest end of that spectrum.
This will likely be followed by a rebadging of the Polaris architecture at the bottom end. It’s hard to say what the nomenclature will be, but our money’s on AMD dropping the RX 400 series, and opting for the RX 500 series instead.
BEYOND THE BEYOND
With AMD and Nvidia announcing their product timelines publicly, we can comment on a few items that we know for certain are coming over the next few years. We know that Vega 10 and 11 should be dropping in 2017, to be followed by AMD’s transistor shrink in 2018 from 14nm to 7nm, thanks to the help of GlobalFoundries. This means we could expect to see a dual-cored, 7nm card in only two years. Vega 20 (all hypothetical from this point onward) should come packing a phenomenal 32GB of HBM 2.0 for a mind-boggling 1TB/s of total memory bandwidth, and a scarily huge 8,192 shader cores. That’s around nine times more transistors than found in a single RX 470—45.6 billion on a single GPU.
Vega 20 sounds insane, and for now is just rumor. But when you consider that alignment with GloFo’s 2018 goals, and the fact that AMD will launch its new Navi architecture in 2019, it’s a lot more reasonable to expect Team Red to produce a 7nm dual-cored test subject, before dropping a new architecture on top of a die shrink simultaneously.