It used to be that GPU chips moved to new process nodes pretty frequently, previously as often as annually. That is up until 2011. That was the year that 28nm GPU’s were unveiled. Since then there has been a long pause. Now in the wake of the 2016 CES both Nvidia, with its previously announced Pascal, and AMD, with the just announced Polaris, are both on record as moving from 28nm planar to FinFET nodes in 2016. This will make 2016 a big year for GPU’s. While Nvidia is going to be fabricating on TSMC’s 16FF+, AMD will most likely be using Globalfoundries’ 14nm FinFET process for the first of the Pascal family.
In their announcement, AMD is touting huge power savings in their design. They demonstrated it using Star Wars Battle Front running at 60fps on 1080p monitors to compare their power consumption against Nvidia’s current 28nm offering. The AMD consumed ~85W while the Nvidia chip consumed ~150W. To be fair it is worth pointing out that is more of an illustration of the performance to power ratio between 28nm planar and FinFET. We’ll have to wait for Nvidia’s FinFET chip to do an apples-to-apples comparison between the two companies’ comparable offerings.
AMD says that they have made major architectural changes to improve performance. On top of this they spent “months” tuning clock gating and power gating. FinFET of course offers big reductions in leakage power. AMD’s Polaris will support HDMI 2.0a and DisplayPort 1.3. It also will include H.265 Main 10 decoding at 4K. The first Polaris design will not use HBM, opting instead to use conventional GDDR5 memory for now. This makes sense for lower cost consumer products. But the option to switch to HBM is always available for subsequent high end offerings.
At this point the biggest question is when will each competitor get to volume production and ship finished products. AMD has other big plans for 2016 as well. They are going to be rolling out their latest Zen architecture for CPU’s. AMD has undergone big changes over the years. But they are fundamentally doing battle with Nvidia and Intel at the same time. While they no longer have a captive foundry, they are working with Globalfoundries at 14nm on the new Polaris based GPU’s.
It looks as though AMD is pitching Polaris as a way to offer higher performance graphics in chassis that had thermal dissipation issues such as laptops, and smaller form factor desktop computers. Also it will likely simplify power connections to discrete graphics cards. It’s good to see GPU’s moving forward after being stalled out at 28nm for an unprecedented 5 years. It seemed that while CPU’s and FPGA’s moved ahead, GPU consumers have had to wait for the architectural gains that a node change offers.
Share this post via:
Build a 100% Python-based Design environment for Large SoC Designs