Preparing for the Next-Gen GPUs
Nvidia’s next GPU architecture is likely coming in August
FALL IS FAST APPROACHING, and we should see new GPUs from Nvidia in the near future. The company has been tight-lipped as usual, but I do know a few things, and can make some reasonable guesses at the rest. Codenamed Turing, the GTX 1180 (possibly 2080) will be the first commercial use of GDDR6, and it will be manufactured on TSMC’s 12nm FinFET process. Nvidia has big plans for Gamescom, which means hardware should be in our hands shortly before you read this.
Nvidia looks set to transition to a split architecture this round. Pascal had professional features, such as half-speed FP64 and HBM2, on certain chips (GP100 versus GP104, for example), and it looks like Volta will be exclusively the domain of supercomputing and professional solutions. Turing will likely skip the Tensor cores and have 1/32-speed FP64, but the double-- speed FP16 should be present. GDDR6 will move from 11GT/s for the 1080 Ti to 14–18GT/s for Turing, probably 16GT/s, for a nearly 50 percent generational increase in bandwidth.
I’m looking forward to seeing a single graphics card that can break 60fps at 4K ultra settings in most games. Now that 4K 120Hz displays are available, the need for faster graphics cards is more pressing than ever. If the 1180 follows past patterns, it will be 20–35 percent faster than the 1080 Ti, which means games where the 1080 Ti comes up a bit short should finally move over the 60fps barrier—or I can just drop settings a notch, and hit 90–120fps for ultrasmooth high-res gaming. I can’t wait!
As for AMD, Vega 7nm should arrive by the end of 2018, but will it be for gamers? I hope so, as there’s nothing else on the roadmaps.
Whatever the next GeForce is called, it will be fast.