After what seems like an eternity on 28nm, Both Nvidia and AMD are moving to new process nodes. Nvidia has already shown their hand with their Pascal GTX 1080 on 16nm FinFETs, offering a compelling upgrade over the GTX 980Ti. On the red team, AMD is readying their Polaris 10 GPUs on 14nm FinFETs as well. Today, we have what appears to be a Polaris 10 test that shows the card running at 1266 Mhz.
On 28nm, Nvidia cards tended to clock higher than their AMD counterparts. This was due the architectural differences between the two firms and AMD made up for it on the performance side. On 14/16nm, we may have the same situation occur again. It’s important to keep in mind that these might not be the final clock speeds for Polaris 10 just yet. As we all know, clock speed is also only one measure of performance and with the changes to GCN 4.0, AMD might not even have to clock that high to reach their performance targets.
We also see that Polaris 10 is using 8GB of 7.6Ghz DDR3 over a 256bit bus. This is likely an error, meaning the choice of either GDDR5 or GDDR5X is still up in the air. I expect that the 256bit bus is correct but the 7.6Ghz speed for the ram might not be entirely accurate just yet. 7.6Ghz GDDR5 over a 256bit bus would only give 243GB/s of bandwidth which seems a bit low but still possible.
So far, we have the 67DF:C7 and 67DF:C4 which are both Polaris 10 dies, likely one full and the other cut down. Unlike Nvidia’s staged release, we may see AMD launch their entire Polaris 10 portfolio at once during Computex. Hopefully, we’ll get the cards soon and see how they fare against the GTX 1080 and GTX 1070.
