Podcast #514 – NVIDIA GeForce RTX 2080 and 2080 Ti Deep Dive

6 Comments

1. 

   7LivesLikeXP on September 20, 2018 at 6:23 pm

   With Microsoft offering paid
   With Microsoft offering paid extended security support(1) for Windows 7 until January 2023. Does anyone at PCPer Know if AMD will begin to offer some sort of Windows 7 support for its Zen/Epyc Customers and Radeon Pro SKU customers also, ditto for any 8.1 support at all from AMD for any Zen micro-arch based SKUs?

   Apparently Intel is also beginning to also offer some sorts of continued windows 7 support in the enterprise markets considering the new offer from MS for Paid Windows 7 security patches until January 2023.

   My guess for MS offering the paid extended support for Windows 7 until January 2023 is that both Windows 7 and Windows 8.1 are relatively the same under the hood so it’s not going to be too hard for MS to offer Windows 7 Pro/enterprise users the paid extended support on 7! Windows 7 and 8.1 are mostly the same under the hood anyways and security patches for one with also work mostly for the other with very little effort involved for MS.

   (1)

   “Microsoft to offer paid Windows 7 Extended Security Updates

   Microsoft is stepping up its Windows 7 migration campaign with a paid security update option, Desktop App Assure service and more.”

   https://www.zdnet.com/article/microsoft-to-offer-paid-windows-7-extended-security-updates/

   Reply
   • 

     Jeremy Hellstrom on September 20, 2018 at 6:48 pm

     Not sure if this helps, but

     Not sure if this helps, but it applies to volume which is where you are going to see the extended support at a price.

     https://docs.microsoft.com/en-us/windows-hardware/design/minimum/windows-processor-requirements

     I can't imaging a scenario in which AMD or Intel would directly offer support for any operating system … microcode updates maybe but not direct support.

     Reply
     • 

       YesYouCanImagineIfYouActuallyWorkInEnterprise on September 20, 2018 at 9:39 pm

       Just the biggest one as far
       Just the biggest one as far as chipset drivers are concerned, AMD’s an Intel’s chipset drivers, CPUs need drivers also! And that would be because it’s the enterprise customers getting the Extended windows 7 PAID support for any the Xeon and Epyc based operations and also even for the Ryzen Pro/newer Intel branded Business grade PC/Laptop parts that will be what is going into the business grade laptops and PCs.

       So the enterprise customers billions speak pretty loudly to Both Intel, AMD, and MS so there will be plenty of Windows 7 customers until 2023 with that extended Windows 7 PAID support being offered, really demanded by the current windows 7 shops from Microsoft from those enterprises that have loads of money inversted in their current windows 7 certified mission critical software(servers, Business PCs, Business Laptops).

       So remember all those Windows XP enterprise operations that took their sweet time moving to windows 7, again because that mission critical software cost more than anyone’s OS. Well they want to get their money’s worth out of windows 7 before they have to spend the millions necessary for certifying their mission critical software for windows 10, or more than likely Microsoft’s WIN 10 based Long term service branch(The One with all the windows 10 Cruft removed) with the longer term stable release version support that enterprises insist on.

       Chipset Drivers/Firmware have to be certified by the Chipset makers for any OS and the CPU makers also are the Chipset makers even if their Chipsets are on the SOC and not separate on some SKUs now.

       So AMD has been slacking on Chipset Driver support targeting windows 7, and even 8.1(GPUs/Graphics drivers also) for the consumer space! But for AMD, Intel also, they will do what is necessary for Zen/Epyc or Xeon/Business servers, laptop, and PC SKUs as well as the server SKUs because that’s what the big enterpries customers demand.

       So if AMD wants to sell any Zen/Epyc to any big server business clients that may still be windows 7 only shops, most likely on the emterprise’s Business grade laptops/PC that will be Ryzen Pro’s Targeted market also! Well then AMD, or Intel, will be having to certify chipset drivers for windows 7/8.1 a bit more than they had planned.

       Microsoft sure was made to see the light by its enterprise customers as far as Windows 10 was/is concerned hense that LTSB spinup of a version of windows 10 without the nonsense attatched and without any new and shiny added every release cycle. But even with Windows 10 LTSB there is still the cost to some enterprises of that certifing of the mission critical software for any new OS, and so windows 7 becomes the New XP and some shops will be using windows 7 for the next few years after 2020 or MS will lose clients with fat wallets.

       AMD has a line of Ryzen Pro CPUs and Raven Ridge Pro APUs for business clients PCs/Laptops and any enterprises still using windows 7 for their fleets of Business Laptops/PC are most deninitely getting that extended windows 7 Paid support so that verly likely means that Zen/Zen+ and even Zen2 based CPUs/Chipset ecosystem drivers had better have that certified for windows 7/8.1 support for the Enterprise customers that want it.

       Even Intel’s chipset production being moved from 14nm back to 22nm production to make more 14nm capacity available for 14nm CPU production are getting more windows 7 support!

       Windows 7 is the new XP so say the Enterprise customers with windows 7 shops/operations and Microsoft will dance that dance on the Enterprise customers Demand!

       Dance You Little MS Monkeys dance dance Dance!

       Reply
2. 

   BreakingUpIsHardToDoInPhaseChangeLand on September 21, 2018 at 11:34 pm

   Interesting Rumor about
   Interesting Rumor about Intel/Micron XPoint split-up from
   Semiconductor Engineering(1).

   “Chipmakers

   In July, Micron and Intel announced that they were going their separate ways in the development of 3D XPoint. The companies have agreed to complete joint development for the second-generation of 3D XPoint technology, which is expected to occur in the first half of 2019. Technology development beyond the second generation of 3D XPoint will be pursued independently by the two companies.

   3D XPoint is made at Intel-Micron Flash Technologies (IMFT), a joint venture fab in Lehi, Utah. Right now, Intel and Micron are partners in the fab.

   According to equipment sources, Micron will take ownership of IMFT when the companies separate next year. Intel is moving to start a 3D XPoint line, probably at F68 in China, according to sources.

   In response, a spokesman for Intel said: “We don’t comment on rumor or speculation.” “(1)

   (1)

   “Week In Review: Manufacturing, Test

   Tariffs; Intel-Micron saga; mask/litho survey.
   September 21st, 2018 – By: Mark LaPedus ”

   https://semiengineering.com/week-in-review-manufacturing-test-12/

   Reply
3. 

   ReallyLikestoRemainAnonymous on September 24, 2018 at 11:27 pm

   So a point on the RTX series
   So a point on the RTX series that no one seems to be saying much about is that to run DLSS NVIDIA now will require you to use GFE? At least that seems to be the way it works. So now it appears they are locking important features behind their “pay us with your information” wall? That certainly seemed to be what you (and other sources) are saying in the reviews.

   Reply
   • 

     GigaRaysDividedBy1000TimesTheFrameTimeIsNotMuchRays on September 25, 2018 at 1:12 am

     They can have their logon to
     They can have their logon to play nonsense and all AMD would have to do is respin a new Vega Micro-Arch based GPU die tapeout with enough ROPs to match the GTX 1080Ti’s compliment of 88 ROPs and AMD could sell GPUs like mad! And it’s all because most of the gaming titles do not support RTX currently and that will take some time to occur.

     DLSS is a good technology that’s AI accelerated but in Nvidia’s hands that only means lockin, and Nvidia did not invent the Tensor Core or AI. Tensor Cores are just more efficient matrix math units that are used to run the AI Trained denoising/DLSS tasks on Nvidia’s RTX SKUs. Nvidia traines the AI on its supercomputing clusters that run on hundreds of Volta/V100 based Tensor Cores on whatever server CPUs that Nvidia likes. Nvidia trains its DLSS on the gaming title’s actual generated super high resolution Textures/Images/other game image assets that are provided by the game’s maker. So millions of Volta GPU/Tensor Cores hours of AI training to create the custom trained(For each Gaming Title) DLSS AI that is loaded and runs on RTX Turing’s limited amount of Tensor Cores.

     AMD could very well do the same and Vega 20 will have some Matrix Math related extentions to the Vega GPU micro-arch. I’m sure that Microsoft, Sony, are already contracting with AMD’s semi-custom division for console APU based Tensor Cores for that very same DLSS style AI based anti aliasing IP as that represents that best and lowest processing power related upscaling of 1080P content to a higher resolution via using a trained AI running on Tensor Cores! And Nvidia does not have a monopoly on the Tensor Core market or the AI market as a whole, where Microsoft and Sony are also doing said DLSS style research along with everybody and their dog.

     Consoles are really in line to benifit the most from any Trained AI based upscaling/AA where the AI is trained on supercomputing clusters using super high resoultion gaming assets that the games makers generate to be used to train the AI to more efficiently upscale any 1080P game texture/images assets to 1440P and 4K with very little processing horsepower needed on the Tensor Cores that are running that trained AI.

     Trained Tensor Core based AI denoising is the only reason that Nvidia’s crappy Ray Tracing output can be made usable for gaming as that trained AI Denoising is actually the secret sauce to enable Nvidia’s Hybrid “Real Time” Ray Tracing that’s mixed with Raster Output.

     Look at any of Nvidia’s stated 10 or 8 GIGA-RAY/second mertics for its RTX consumer variants and that has to be divided by 1000(1000ms in one second) and the multiplied by 16.67ms for 60 FPS game play or 33.33ms for 30 FPS game play. And you will see on average how many Rays Per Alloted Frame time are actually available for gaming at between 30 FPS and 60 FPS. And that’s not a whole lot of time per frame for generating enough rays per pixel to produce anything but a grainy Image that has to be AI denoised using some Tensor Cores that run the Trained Denoising AI, and some of the Tensor Cores are going to be needed for DLSS in addition to Denoising and any other AI tasks.

     How many million Pixels on the screen at 1080P, or 1440p, up to 4K and that’s maybe one maybe two rays per Pixel, if not all of the scene is ray traced and then there are shadows via Traced Rays to consider and ambient occlusion and environment lighting via Traced Rays and those are all different passes that need Rays Generated. That’s a whole lot of grainy output to be AI denoised in between 33.33ms and 16.67ms what with all that other processing that needs to get done between 33.33ms and 16.67ms or less than 16.67ms for any frame times above 60 FPS.

     Reply

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