AMD reveals some details of Polaris, their next GPU architecture

AMD is getting some pre-CES buzz going with light details of their upcoming GPU architecture, Polaris. Please head on over to AnandTech for the full story where Ryan Smith talks about the information AMD released.

There’s not much in the way of hard info but there were a few tidbits of interest to SFF. The reporters at the event were not allowed to take pictures or even see it but RTG (Radeon Technologies Group) demonstrated a working Polaris card and had it installed in an Intel i7 system alongside an identical system with a GTX 950 installed. With Star Wars Battlefront running on both machines at 1080p and medium settings, the GTX 950 setup was pulling 150W according to a power meter and the system with the Polaris-based card was only pulling about 88W with both computers staying around the 60fps v-sync limit. Being able to get GTX 950 levels of performance in a setup where the entire machine draws less than 100W will make for some interesting ultra SFF possibilities.

While impressive, this is obviously not a flagship card and Ryan speculates that this is because the first Polaris cards to hit the market won’t be the high-end models due to the difficulties of manufacturing the larger dies on the new process. Speaking of manufacturing processes, these new cards will be built using 16nm/14nm FinFET, a welcome change from the 28nm plateau the GPU industry has been stuck at for several years now. Interestingly, RTG has confirmed they will be using both GlobalFoundries (14nm) and TSMC (16nm) for Polaris, though they didn’t specify exactly how that will work.

The slide deck emphasizes the benefit of the smaller process node for power efficiency and performance/watt but one thing I noticed is several different slides specifically call out SFF as a beneficiary of this, with phrases like “Enables incredible form factors”, “The GPU design for Small Form Factors”, and my favorite: “Discrete cards with less power connectors”. As someone using two splitters in order to get the necessary number of connectors to run my dual GPUs off a SFX PSU, that is welcome news indeed.

We will have to wait until actual cards start showing up but it looks like RTG remains committed to building physically smaller GPUs like the Fury X and Nano. Hopefully with this new-found power efficiency the AIB partners won’t find it necessary to slap humongous heatsinks on the new GPUs so the aftermarket cards can better work for the SFF community.

The presentation revealed a planned availability of the Polaris architecture in mid-2016 though no word on exactly which cards so like Ryan speculated, it could very well be that just the low to mid-range cards release at that point and the flagships follow at a later date but we should get more info closer to Computex.

Thoughts? Discuss them in the forum.