AMD’s Scott Herkelman on CES, Radeon VII and NVIDIA

4 Comments

1. 

   HarderQuestionsShoudHaveBeenAsked on January 14, 2019 at 9:48 pm

   Kyle’s “full” Radeon VII
   Kyle’s “full” Radeon VII specifications, at the time that I read the artice, where missing the Shader core and TMU counts as well as all the relevant TMU: GTexel/s and ROP: GPixel/s maximum theoretical numbers.

   Vega 20 has been out for months and there has to be some relevent metrics that can be extrapolated for the Vega 20 die bin that’s used for Radeon VII.

   DP FP has been diesbled to a lower figure for the Consumer Vega 20 variant and that would imply product segementation. I hope that AMD’s MI50/60 sales are good and that there will be some Radeon Pro WX variants of Vega 20 also. Maybe even a Radeon Pro DUO refresh using Vega 20 with more FP64 enabled for the Prosumer market.

   I wish Kyle’s interview asked more xGMI related questions even if AMD’s answer would have been wait for the release date on Radeon VII.

   Has anyone tried to look closer at the CES Radeon VII images for any signs of xGMI hardware links across the top of the Radeon VII’s PCB.

   Sure JHH’s comments are what they are JHH/Nvidia are more extended into the GPU market for the lion’s share of their revenues than AMD. AMD will in not too many more business Quarters start earning more off of is x86 Server CPUs than Nvidia earns from GPUs. So Nvidia has some CPU envy there and the prospect of Intel entering the GPU race in the beyond 2020 time frame just has to be giving JHH the hebejebes at night and ruining sets of those most expensive Emperor of China silk sheets sets.

   Just look at how much of that Nvidia market Cap that went puff like an overloaded VRM, that’s got to be grating on JHH’s nerves also!

   The AI focused Vega-2 Micro-Arch ISA extentions what about that and any L2 cache tweaks on even tweaks to other parts of Vega 20/Radeon VII GPU Tapeout. With Only 60 nCUs enabled on the Radeon VII bin that’s still showing up in AMD’s, and some other leaked Benchmarks, as competative with the RTX 2080 in raster oriented gaming titles that’s good news for Vega 20/Radeon VII.

   Reply
2. 

   whereisthelink? on January 14, 2019 at 9:58 pm

   “Amd’s Scott Herkelman Syncs
   “Amd’s Scott Herkelman Syncs up CES GPU Shade(rs)”

   Reply
3. 

   othertomperson on January 15, 2019 at 9:26 am

   He keeps repeating the fact
   He keeps repeating the fact that The Division 2 supports “all Radeon VII technology features” without elaborating on what that means. One of these was in the context of in-game ray tracing. Is he implying that Radeon VII supports real-time ray tracing in The Division 2?

   Reply
   • 

     RasterGamingIsStillTheKingOfOz on January 15, 2019 at 5:00 pm

     Well is may not support all
     Well is may not support all of Radeon VII’s features because Radeon VII uses the Vega-2 Micro-Arch and that includes the AI related ISA extentions that are only present on the Vega 20/Vega-2 ISA. And what about Vega 20’s support for xGMI(infinity Fabric) is that even included in the Radeon VII gaming SKU that’s binned from the Vega 20 base die tapeout.

     Is Division 2 making use of the new Vega-2 micro-arch’s AI ISA for any gaming usage, who knows. And Pascal/earlier or Vega/earlier(Back to GCN 1.1) GPUs can accalrate Ray tracing calculations on the GPU’s shader cores. But that’s not likely to be as efficient as RTX/Turing’s hardware based Ray Tracing/BVH IP.

     And even Nvidia’s RTX IP has to have help via tensor core based denoising to make use of the limited numbers of RT cores rays that can be generated in the 33.33nm(30FPS) down to 16.67ns(60FPS)/Lower frame times. Even Nvidia’s limited RT cores output results in grainy images that have to be AI/Tensor Core denoised to be of use in gaming workloads.

     Reflections and refractions requres rays, so do Ray Traced shadows and ambient occlusion passes. Ray traced sub subsurface scattering eats more rays. So even at 10 GigaRays/s that all has to be devided by 1000 and multiplied by the frame time in milleseconds to get a rough idea of the actual numbers of rays that can be generated per frame time. That’s not enough rays generated to create any final image and that has to be denoised and mixed down with the usual raster output to be usable for gaming.

     Even Nvidia’s RT hardware output is insufficient and has to undergo fruther denoising vai the Tensor Cores before the output becomes acceptable.

     Most gaming titles are raster oriented still and nvidia and the games makers can not change that overnight.

     I do not trust marketing VPs, and the NDAs are still in effect and HardOCP did not ask any hard questions even in order to see what subject matter related to Radeon VII is sill waiting for the NDA to expire. Sometimes those no comment to some questions asked can reveal what product informstion may still be restricted.

     And really that “all Radeon VII technology features” quote answers the question about only the Radeon VII’s feature sets that does not include Ray Tracing hardware. If Radeon VII supported Ray Tracing hardware AMD would have already announced that. But AMD’s got no worries about having to compete with RTX/Turing’s new features! Not with gamers more opsessed with that FPS metric over any other gaming metric. So Radeon VII’s FPS metrics against the RTX 2080 in raster gaming titles is what matters currently what with the majority of gaming titles not RTX ready just yet.

     Reply

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