It’s an Epic post-apocalyptic GPU Exodus

3 Comments

1. 

   Godrilla on February 14, 2019 at 4:02 pm

   Great review from Guru 3d. My

   Great review from Guru 3d. My 9 year old i7 980xe at 4.3 ghz is still capable of running this game and will likely still be gpu bound and it seems anything more than 6 gigs of vram is overkill at 1080p ive seen videos at 4k extreme it goes ip to 9.5 gigs utilization.

   Just preorder $49.99.

   Reply
2. 

   othertomperson on February 14, 2019 at 4:14 pm

   They should adopt error bars,
   They should adopt error bars, similar to Gamers Nexus. Those bars exaggerate the differences. Between GTX 1080, Vega 56, 64 and 2070 you’re talking 4fps at 4K. I wonder how repeatable these close benchmark results really are.

   Reply
3. 

   TensorCoresAreTheNewSlicedBread on February 14, 2019 at 5:25 pm

   AdoredTV has his late to the
   AdoredTV has his late to the party Radeon VII limited video review up(Got Sampled a Review Radeon VII kit late from AMD)! And some game titles are startng to bump up against the 8GB VRAM limitation and that’s resulting in some interesting occurrences on Radeon VII, and Vega 56/64 with their HBCC/HBC IP enabled when things go past the 8GB VRAM limits on some GPUs compared to GPUs that have 8GB hard VRAM limits that either don’t have HBCC/HBC IP or lack Radeon VII’s(16B VRAM) larger than 8GB VRAM.

   There now should be more focus on the Vega HBCC/HBC IP and its ability to turn the HBM2/VRAM into a last level VRAM Cache(HBC) to a larger pool of Virtual VRAM paged to/from system DRAM by that HBCC on Vega. And even for GPUs with only 8GB of HBM2 if the game titles are now bumping up against the 8GB limit to a few GBs above 8GB on some titles in testing.

   AMD definitely has some advantages with its Vega HBCC/HBC IP and HBM2’s higher effective bandwidth but Nvidia’s in hardware dedicated Tensor Cores are the one thing that AMD has to get on any Post Navi Designs because that DLSS/Tensor Cores Nvidia based IP looks to be the most interseting for increasing the game’s average frame rates.

   Dedicated Ray Tracing hadrware can wait a little longer but not the hadrware Tensor Cores IP, AMD needs that for Replacing the very Shader Core/Raster Pipline stresses that AA puts on any GPU’s resources. Offloading/replacing Multisample anti-aliasing(MSAA) with some Trained AI methods like DLSS on dedicated hardware based Tensor Cores in really putting Nvidia ahead.

   More Questions Need to be asked of Microsoft about their DXR API and maybe offering a separate software code path under DXR for Ray Generation on the GPU’s Shader Cores for older Nvidia GPU generations and AMD GPUs that lack dedicated Ray Tracing hadware. Ditto for and DX12/AI related code paths in software for GPUs that Lack Dedicated Tensor Cores. Also the very same Qusetions to the Khronos Group and their cross platform Graphics API Vulkan.

   Nvidia is writing some RTX specific Vulkan extentions for their new RTX hardware but Khronos also needs to be developing a software code path for Ray Tracing and AI performed on Older GPU’s Shader cores for Ray Tracing and AI related image processing accelerated on the GPU’s shader cores on older generation Nvidia GPUs and AMD’s GCN GPUs.

   Radeon VII’s “Vega-2” micro-arch has some AI related ISA extentions that the Vega 10/Vega-1 GPU Micro-Arch lacks. So who will be cornering AMD’s engineering folks at the next big Tradeshow/Gameing Technology Event and asking AMD about those Vega 20/Vega-2 Micro-Arch AI related ISA extentions and any sorts of AI based AA/Upscaling on Radeon VII/Vega 20 based Consumer Gaming GPU variants, even on Pro Graphics Oriented Radeon WX Branded Vega 20 Based GPU SKUs. AMD will most devinitely be releasing some Radeon Pro WX(formally FirePro) Pro GPUs that are Vega 20 based.

   Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.