NVIDIA unveiled new details about its new Pascal architecture at the Graphics Technology Conference in Japan. This architecture is close to offering twice the performance per watt consumed over the current Maxwell architecture by applying the latest technologies such as stacked HBM2 (high-bandwidth memory 2).

NVIDIA Pascal will have up to 4-gigabyte HBM2 stacks, giving a total of 16GB (32GB for the Quadro / Tesla). The consumer GeForce variant is expected to serve up 16 GB. The combined bandwidth of the chip will be 1TB, while its top is fixed to the 2 TB / s.

The NVIDIA Pascal GPU will boast 16GB HBM2 with 1TB/s bandwidth at Launch


On the other hand, it is more than confirmed that NVIDIA build its Pascal chips under the manufacturing process of 16nm FinFET (while AMD could make the direct leap to 14nm FinFET, although not ruled 16nm FinFET), the most important point is innovating a new interconnect called NVLink.

Currently, dual GPU cards are essentially two graphics cards in a common PCB, this downside is due to the bandwidth of the PCI-Express slot is shared by a bridge-chip; while the SLI bridge connects two GPUs, each in a slot but the bridge itself generates a bottleneck. With NVLink, the two GPUs will be interconnected via a  bidirectional data of 80 GB / s, allowing each GPU directly addresses the memory controlled by the other. This should greatly improve memory management in games that take advantage of the latest API like DirectX 12 and Vulkan and performance thereof under high-resolution monitors.

Previously at the GTC March, 2015, we heard a lot from NVIDIA about Pascal based architecture, but a few changes and improvements are noticed as compared with previous one.

  • 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.

NVIDIA is expected to launch its first products based on Pascal architecture during the first half of 2016.


Via: TechPowerUp