@davillain- said:
Radeon VII is not what (or many other people) I expected, but Jensen Huang should not talk such trash. It's still 2080 performance for 100$ less. Sure, there's no raytracing or DLSS, but those are gimmicks for now that are used in like two games. So TBH the Radeon VII is a good competitor in the med/high-end segment because as of now raytracing and DLSS aren't ready for prime time & for all games.
I don’t like Nvidia and their shady business practices, but with that Radeon VII ridiculous price tag is even less appealing and I'll still use Nvidia GPU only because they have what I really need and I ain't got nowhere to look but Nvidia so can you blame me? Sink or Swim, pick one.
https://devblogs.nvidia.com/nvidia-turing-architecture-in-depth/
Turing Tensor Cores add new INT8 and INT4 precision modes for inferencing workloads that can tolerate quantization and don’t require FP16 precision. Turing Tensor Cores bring new deep learning- based AI capabilities to GeForce gaming PCs and Quadro-based workstations for the first time. A new technique called Deep Learning Super Sampling (DLSS) is powered by Tensor Cores. DLSS leverages a deep neural network to extract multidimensional features of the rendered scene and intelligently combine details from multiple frames to construct a high-quality final image
VII has added support for INT8 and INT4 for inferencing workloads within it's CU. VII has higher memory bandwidth and 128 ROPS(Anandtech claim) for heavy MSAA at 4K which reduce the need for DLSS based pixel reconstruction AA
Vega 64 has support for INT8 for inferencing workloads within it's CU.
----------
RT Cores accelerate Bounding Volume Hierarchy (BVH) traversal and ray/triangle intersection testing (ray casting) functions.
https://32ipi028l5q82yhj72224m8j-wpengine.netdna-ssl.com/wp-content/uploads/2016/08/169798-A_AMD_RadeonRays_Intro_FNL.pdf
Acceleration strategies The default acceleration structure is Bounding Volume Hierarchy (BVH) using spatial median splits. It maintains fast build times and provides decent intersection performance. You can enable SAH builder using the global option and trade off construction time for better intersection performance
AMD is aware of BVH issue.
https://worldoffries.wordpress.com/tag/deferred-shading/
BVH code on R9-280X era.
https://wccftech.com/titan-v-high-fps-battlefield-v-rtx/
Enabling RT function RTX coreless on Titan V. Artificial software block on older GPUs.
Log in to comment