According to a new rumor floating around it looks like NVIDIA will be changing up the naming for their upcoming GeForce graphics cards. Most of us expected the GeForce GTX 1000 series (that’s what comes after 900, of course), but apparently that has too many digits. NVIDIA will instead call its next-generation high-end graphics card series the GeForce X80 series. This series will be based on the performance-segment “GP104” and high-end “GP100” chips. The GeForce X80 series will be made up of the GeForce X80 (performance), GeForce X80 Ti (high-end), and GeForce X80 Titan (enthusiast).
The “GP104” silicon, which will be based on the “Pascal” architecture is expected to feature as many as 4096 CUDA cores, 256 TMUs, and 128 ROPs. It will have a GDDR5X memory interface with 384 GB/s memory bandwidth. 6 GB could be the standard amount of memory. Its texture and and pixel fill rates are said to be 33% higher than those of the GM200-based GeForce GTX Titan X. The “GP104” chip will have a TDP of 175W and is built on the 16 nm FinFET process.
It seems like “GP100” will be an entirely different beast all together. While it is built on the same 16 nm FinFET process as “GP104” it will have a TDP of 225W. This chip will have unique memory controllers, which will support both GDDR5X and HBM2 interfaces. So there could be two packages for the GP100 silicon, one for each memory type. The GDDR5X package would be smaller with a large pin-count to wire out to the external memory chips. The HBM2 package would be larger and much like what we say from AMD’s “Fiji” package. Given this information the GeForce X80 Ti and GeForce X80 Titan would be two significantly different products.
The “GP100” silicon will feature 6144 CUDA cores, 384 TMUs, and 192 ROPs. On the X80 Ti you will get 5120 CUDA cores, 320 TMUs, 160 ROPs, and a 512-bit wide GDDR5X memory interface holding 8 GB of memory with a bandwidth of 512 GB/s. The X80 Titan on the other hand will feature all of the cores, TMUs, and ROPs present on the chip and will have a 4096-bit wide HBM2 interface, holding 16 GB of memory with a bandwidth of 1 TB/s. Both of these card will double the pixel and texture fill rates of their previous cards.