http://www.anandtech.com/show/9124/amd-dives-deep-on-asynchronous-shading
It's been discussed many times how the table is misleading AND the written info is wrong, why it's wrong, we can speculate (NV PR?) but it doesn't change the fact this article is being used to educate readers on hardware topics relating to a major DX12 feature.
https://community.amd.com/community...blade-stealth-plus-razer-core-gaming-solution
It was obvious back then that Kepler & Maxwell cannot have parallel graphics & compute execution within its engines.
It's even more obvious when you read the official NV programming guide for GameWorks VR where they specifically say their pipeline can stall due to not being able to prioritize timewarp compute. Basically if the pipeline is rendering graphics, a priority context switch for async compute cannot proceed until the graphics task is finished. This is the reason for their high motion to photon latency for VR, which they DO admit on their official NV blog. Whereas AMD can get down to 10ms, NV's best efforts have been ~25ms.
Perhaps AT find it worthwhile to clarify the topic and maybe even ask NV PR why they saw fit to create a situation that made it seems like their hardware can support Async Compute/Shading when it cannot?
It's been discussed many times how the table is misleading AND the written info is wrong, why it's wrong, we can speculate (NV PR?) but it doesn't change the fact this article is being used to educate readers on hardware topics relating to a major DX12 feature.
https://community.amd.com/community...blade-stealth-plus-razer-core-gaming-solution
It was obvious back then that Kepler & Maxwell cannot have parallel graphics & compute execution within its engines.
It's even more obvious when you read the official NV programming guide for GameWorks VR where they specifically say their pipeline can stall due to not being able to prioritize timewarp compute. Basically if the pipeline is rendering graphics, a priority context switch for async compute cannot proceed until the graphics task is finished. This is the reason for their high motion to photon latency for VR, which they DO admit on their official NV blog. Whereas AMD can get down to 10ms, NV's best efforts have been ~25ms.
Perhaps AT find it worthwhile to clarify the topic and maybe even ask NV PR why they saw fit to create a situation that made it seems like their hardware can support Async Compute/Shading when it cannot?
Last edited:
Facebook
Twitter