Nvidia Reveals More Pascal Architecture Details – 16GB HBM2 Memory at 1TB/S bandwidth
Nvidia recently released some more information about its highly anticipated next generation Pascal GPU architecture during their GPU Technology Conference in Japan and from the looks of things, the wait for this new architecture and its subsequent graphic cards will totally be worth it. Especially, if someone is upgrading from the Kepler series of GPUs, though that doesn’t mean it won’t be a worthy upgrade for Maxwell users.
The Pascal GPUs will be based on the latest 16nm FinFET processing node that should allow significant power savings. As expected, the architecture will also be using the upcoming HBM2 memory type and support up to 32GB of memory although the initial consumers GPUs are expected to launch with around 16GB of HBM2 memory.
They will also offer combined memory bandwidth for chips to a total of 1TB/s. However, the internal bandwidth can go as high as 2TB/s, at least the 16GB ones would, while the enterprise variants such as Quadro and Tesla models should have more bandwidth.
Nvidia also revealed the way they are now enhancing NVLink for dual-GPU cards to offer up to 80GB/s bi-directional data path, letting each GPU access memory controller by the other GPU. This should greatly improve memory management in games that take advantage of newer APIs such as DirectX 12 and Vulkan; and prime the graphics card for higher display resolutions.
There are no details about specific GPU models or release dates yet but Pascal based products are expected to launch in the first half of 2016.