Its seems as if NVIDIA’s next GPU launch will be much like what they’ve done in previous generations with the second biggest chip launching first. This will be a well-priced “enthusiast” SKU that will outperform the previous generation enthusiast product. Then they will launch the bigger chip later and the high-end enthusiast product. Now the second biggest chip coming from NVIDIA is “GP104” which is part of the “Pascal” architecture.
As the report mentions NVIDIA will be using the GeForce GTX 1080 naming for the “GP104” product. This is not confirmed so we do not know if NVIDIA will be using the 1000 series naming scheme or changing it to something more simple. We do know that NVIDIA is designing the chip to keep costs low enough to score big in price-performance and and potential price-war with AMD. To keep things very cost effective NVIDIA is likely to skip out on HBM2 with this chip. Instead with “GP104” NVIDIA will make use of either GDDR5 or GDDR5X. It will more than likely be GDDR5X as the new GDDR5X standard which offers double the density of GDDR5 and faster speeds as well. HBM2 would be reserved for NVIDIA’s big Pascal chip which would more than likely make up the next TITAN product.
GP104-based cards should ship with a single 8-pin PCI-Express power connector. This will be a nice upgrade from the current dual 6-pin approach. A single 6-pin connector is 75W while the 8-pin is 150W. Cards should also have two DisplayPort, one HDMI, and one DVI output connections.
The “GP104” is going to be built on the 16nm FinFET process and be produced by TSMC. NVIDIA will unveil the first GP104-based products by April at their Graphics Technology Conference (GTC) and then make them available to market in May. This could be a late-May launch right before Computex. This could also give NVIDIA’s add-in-board partners a great chance to show off their custom cards at Computex.
