[Rumor] Today we have some details on next-Gen Pascal-based GPUs for which NVIDIA will use both HBM2 and GDDR5X memory depending on the model. Currently, silicon GM200 powering the very high end graphics cards like the GeForce GTX TITAN X or the GeForce GTX 980 Ti, while the GM204 features more affordable graphics cards such as the GeForce GTX 970 or 980.
NVIDIA would use the GDDR5X memory on Pascal based-GPUs, HBM2 for Business cards
You Might Like: NVIDIA shows off its next-gen HBM-based Pascal GPU at GTC 2015
With the arrival of Pascal situation will be the same, except that the silicon GP104 would use successor of GDDR5 memory called GDDR5X and to be released with the next generation GeForce graphics products. Reportedly, the GDDR5X memory remains the same memory interface of 256 bits that can be found on the GTX 980, while the bandwidth of memory grows to 448 GB / s (twice), almost the same as Fury Radeon X (512 GB / s) with HBM memory. The logic of all this is that the GDDR5X memory would be even more cheaper than HBM. HBM2 is still obviously on the cards, and will include a 4096-bit memory bus at 1GHz with memory bandwidth hovering at 1TB/sec.
Furthermore, the GPU based on silicon GP100 could be a future Titan X, based on Pascal architecture GeForce GTX 1080Ti, hypothetical Quadro or other successor would use the HBM memory (HBM2) with a memory interface of 4096 bits running at a frequency of 1.00 GHz will be able to offer a bandwidth of 1TB.
According to the report, the new GP100 chip will feature the following;
- Pascal micro architecture.
- DirectX 12 feature level 12_1 or higher.
- Successor to the GM200 GPU found in the GTX Titan X and GTX 980 Ti.
- Built on the 16FF+ manufacturing process from TSMC.
- Allegedly has a total of 17 billion transistors, more than twice that of GM200.
- Taped out in June 2015.
- Will feature four 4-Hi HBM2 stacks, for a total of 16GB of VRAM for the consumer variant and 32GB for the professional variant.
- Features a 4096bit memory interface.
- Features NVLink and support for Mixed Precision FP16 compute tasks at twice the rate of FP32 and full FP64 support. 2016 release.