Yup. I was being a little too quick with my mouth, and obviously I have written about these very things.

Excellent take thank you Count_de_Money,

I am especially afraid that AMD is going to drop the ball and fluff workstations: but I would rather have better support for Epyc than plump up Threadripper, because the unmet demand that’s for grabs in graphics has been satisfied most unsatisfactorily so far by adapting 4C platforms and as you say, RAM appetite is unlimited for moving picture graphics and professional users are already juggling multiple maxed out Xeon workstations. This market wants v4 lanes and single root topology and maximum L3 and SP Epyc is a bargain and we simply need a push for memory compatibility and ISV certification to be the dominant part in 2021.

Adobe has a feature for using an AI(Tensor Cores) based trained algorithm to pull out a part of an Image from its background and doing that on a image/video and no need for any chroma screen setup to key over parts of an image/images over another background and having to use a complicated compositor setup. So greatly improved workflows in some video and single image tasks that make use of AI trained algorithms to get the work done.

So AMD’s WX branded Professional Graphics Workstation cards better be CDNA based with more of the RDNA graphics capability included for the WX branded Pro Graphics Workstation branded GPU parts. Nvidia’s RTX Quadros doing 2D and 3D Animation rendering workloads and making use of Tensor Cores and Ray Tracing cores to speed up the rendering process via dedicated GPU hardware blocks will have AMD fruther behind unless AMD starts to focus more on Professional Graphics oriented GPU hardware features that require RDNA and CDNA branded GPU feature sets.

But I’d imagine that even CRNA based GPUs will still have those tessellators, TMUs(Texture Minipulation Units), and ROPs(Raster Operation Pipelines) in addition to the compute additions that the Professional Graphics Workstation end user will make use of and AMD can not pass up the better markups in the Professional Graphics Workstation market where currently Nvidia is king, GPU accelerator market as well.

Nvidia has its NVLink interfacing its GPUs with OpenPower Power9/10 CPUs and that level of CPU to GPU coherency and AMD will have that fully implemented by the Time of Epyc/Genoa using that full Infinity Architecture(Rebranding/Suprabranding of the Infinity Fabric branding) to include both Epyc CPUs and Radeon Instinct/Other GPUs interfaced more coherently via the infinity Fabric that will be AMD’s fully in-house answer to Nvidia’s/OpenPower’s CPU to GPU NVLink interfacing.

AMD’s CEO and CFO where big on stressing gross margins and gross margin growth so AMD needs to give more deference to the Professional Graphics Workstation market segment where like the HPC/AI market the GPU markups are well above any consumer gaming GPU margins/markups. So AMD’s still not fully focused across all the potential TAM without mentioning that Professional Graphics Workstation part that’s the Pro Graphics Workstation GPU market segment.

There also needs to be Some Epyc Branded CPU workstation capable parts that are above an beyond Threadripper’s limited ECC memory support(Does Not support RDIMMs) and limited to 256GB max memory size. And animation rendering needs can vastly outstrip 256GB and need 1TB larger memory sizes. So that’s requiring Epyc and 8 full memory channels instead of only 4 for any Threadripper branded parts. AMD really needs to brig that Infinity Architecture to the Professional Graphics Workstation market with Epyc/Genoa and some Workstation Genoa parts that’s directly interfaced via the Infinity Fabric/Infinity Architecture and bring a more coherent HSA to fruition where Nvidia will have no ability to directly compete unless Nvidia take up an OpenPower license of its own and brings up some CPU competition as well as GPU.