Very Improved Tensors are good news. Nothing about RT cores, but there may be some improvements too. Structural Sparsity is mixed bag depending on use. I guess nobody here wants image based on missing information because GPU just threw away something it considered to be too small contributor to accurate visual result. But if it is applicable to something that enables 50% more pixels rendered in same time frame at 90~95% accuracy, resulting downsampled image would look better. (If it is applicable to rendering, which may not be the case.) Yes, I am thinking on how this may benefit gaming GPUs. What I wonder about is 4 Tensors per SM. No mention of RT even in other news post with details. And number of Transistors per SM. But sadly, SM contains quite different blocks than Gaming Turing has. So this is no indication if nVidia made some parts of SM smaller or bigger while improving them. I wonder what are yields for such GPU.

[youtube=nhzEDhm8wSo]

Fox2232:

Very Improved Tensors are good news. Nothing about RT cores, but there may be some improvements too.

Note: Because the A100 Tensor Core GPU is designed to be installed in high-performance servers and data center racks to power AI and HPC compute workloads, it does not include display connectors, NVIDIA RT Cores for ray tracing acceleration, or an NVENC encoder. https://devblogs.nvidia.com/nvidia-ampere-architecture-in-depth/

This might be my first ever titan purchase, wish they would hurry up