Radeon rumourmongering

Subject: General Tech | January 17, 2019 - 12:32 PM |
Tagged: amd, rumour, Vega VII

There are several rumours bouncing around the internet about the new GPU from AMD, from overall shortages, to selling short through to the non-existence of custom cards from AIB partners.  As usual AMD is fairly tight lipped about unreleased product but did refute the post from Tweaktown stating there will be a mere 5000 cards available at launch, stating that they expected to be able to meet initial demand; a nice change from the issues that plagued the entire industry in 2018.

There are also allegations that the cost of the 16GB of HBM2 will mean AMD will take a loss on every single card sold at the $700 MSRP, which is honestly ludicrous on the face of it.  However there is one rumour that The Inquirer noticed AMD would not comment on, we still do not have confirmation of third party cards.  That would be an odd move on AMD's part, but as this year we will see Intel auction off a high end chip instead of selling them on the open market perhaps we are in for a strange year.

View Full Size

"AMD, however, has dismissed that these supply issues exist, and said in a statement that it has enough supply of the 7nm GPU to meet demand."

Here is some more Tech News from around the web:

Tech Talk





Source: The Inquirer

Video News

January 17, 2019 | 04:34 PM - Posted by MorePowerToAMDAndProfitsSellingDirectly (not verified)

TSMC has got plenty of excess 7nm fab capacity what with smartphone sales going into the crapper. So maybe There is some TSMC 7nm wafer lines that have been brought online to give to Vega 20 die production some boost and even at some discount for AMD.

HBM2 is not as expensive in the 4GB capacity stacks compared to the 8GB larger. And Both Samsumng and SK Hynix have mature HBM2 production ongoing for long enough to stabilize prices. More processor makers than just AMD and Nvidia make use of HBM2 so the economy of scale should be at a point where shortage pricing has given way to a more stable and decreasing price structure for HBM2 over the long run.

AMD selling the Radeon VII directly allows AMD to net retail pricing rather than only wholesale pricing which is lower.

If AMD is using binned Vega 20 dies then those Vega 20 dies are probably not performant enough for Professional usage. So is AMD really taking a loss at $700 MSRP or is that just a smart way to make use of all the Vega 20 wafer output for even those Vega 20 dies that can not be used for Radeon Initinct/Radeon Pro WX SKUs.

Maybe 3rd party cards are not being offered because the AIB's can not afford the markup AMD would have to charge in order to break even or make a smaller profit but who knows. Maybe more Tech Reporters should be working on Keeping track of actual HBM2 pricing from Samsung and SK Hynix. And who is AMD's partner for those first Refrence Radeon VII's that will be sold directly from AMD's website. AND has to cotract with someone to build all theose PCBs and their components.

Of all the GPU makers AMD probably buys the most HBM2 so that's got to equate to some volume discounts from the HBM2 suppliers. 8GB HBM2 stacks are not the highest capacity HBM2 capacitty any longer what with that amended HBM2 standard that JEDEC just adopted as that's now 12GB/Stack using the current die dimensions and will be 24GB/Stack using the newer standard's larger per die dimensions(16 Gb per layer).

So most of the HPC market users will be switching from 8GB HBM2 stacks up to the 12GB and then 24GB HBM2 capacities leaving very little demand for any 4GB HBM2 other than for AMD's more consumer oriented GPU needs. And AMD/Apple uses 4GB HBM2 stacks on that Apple exclusive Vega Discrete mobile variant that used for the latest high end MacBooks, ditto for Intel's Kaby lake G semi-custom offering.

I would not be supprised to see some Vega 20 Die variants make their way into Apple's iMac products and the Mac Pro is supposed to be getting updated also in 2019.

January 17, 2019 | 06:29 PM - Posted by Odin (not verified)

Doesn't hold interest for me at all. When is Navi coming? We want the real next gen architecture, not die shrink Vega.

January 17, 2019 | 08:22 PM - Posted by WaitForAfterNaviOnceNaviIsHereAMDsGotEpyc (not verified)

AMD could just stick the Anyname name on its generations of GCN and gamers would be none the wiser. Nvidia has done that since Maxwell with only Volta(Tensor Cores) and Turing(RT cores) adding anything new.

The GPU micro-archs are mostly the same for both AMD's and Nvidia's GPUs of the last 5 generations prior to Nvidia's Volta/Turing with that being what's mostly been updated by both Nvidia and AMD with minor tweaks anyways for the Raster Gaming focused components on Nvidia's GPUs since Maxwell to Pascal that have only beat AMD's Vega on gaming workloads because of the ROP counts/greater pixel fill rates! That Flagship crown was also won with Nvidia taking its professional GP102 oriented DIE/Tapeout with all those available ROPs(96) and binning out an 88 ROP GTX 1080Ti and enough pixel fill rates to win the FPS competition.

Nvidia's new RT and Tensor Cores features are not adding anything to the Raster oriented gaming/FPS contest above what Pascal already offered.

Gamers can keep blaming GCN but it's not GCN is AMD's lack of a Vega based Die tapeout with a comperable to the GTX 1080Ti's pixel fill rates. With first generation Vega if AMD had included a similar number of ROPs as the GTX 1080TI possesed then AMD's Vega-10 could have easily competed better with the GTX 1080TI in that FPS contest that gamers value above all else.

Navi's coming to replace the Polaris line in the mainstream market slot and Vega-2/Radeon-VII is the closest to matching the RTX 2080 and the RTX 2080 beats the GTX 1080TI. So AMD must have tweaked more on Vega 20 than what could be provided by a process node shrink if Vega 20 is near to the RTX 2080 with only 60 of Vega 20's 64 nCUs enabled. Vega 20 still has 64 ROPs max and it's up against the RTX 2080, 64 ROPs also. But the RTX 2080TI still has a lead in ROPs and more than likely pixel fill rates and higher FPS in raster oriented gaming titles.

It's hard to compete with Nvidia and all those available Tapeouts that Nvidia puts out each generation at the cost of well over a billion dollars while AMD still can only afford one Desktop/High end tapeout Vega 20 that's doing double duty again, like Vega 10, for professional uses and Consumer flagship usage. Just remember this, GP102 and TU102 are professional market tapeouts also with more Quadro Variants and only single consumer variants the GTX 1080ti and the RTX 2080Ti respectively.

So all you flagship gamers only really are gitting the GPUs made from the reject DIE bins of some professional GPU Die/Tapeout. AMD still lacks the funds for 5+ different GPU Base Die tapeouts for each "generation" of latest GPU micro-arch released like Nvidia can afford that cover all aspects of the GPU market top end to bottom, each market with its own targeted Nvidia Die tapeout.

I do not know why Gamers expect anything to change once Navi arrives. And Unless Navi comes with some great difference in ROPs to shaders that can be scaled up to egual more than 64 ROPs and pleny more in the Pixel fill rate metrics, AMD will still struggle against Nvidia with Raster gaming titles. And now AMD has to begin to design Hardware based Tensor Cores and Hardware Based Ray Tracing IP before the gamng software ecosystem begins to adopt Nvidia's RTX technlogy more fully. And Nvidia has the money to pay the games developers to help that adoption rate grow.

But really AMD as a business can still have the last laugh because in a very small number of Business Quarters AMD will begin to earn more revenues on its Epyc/Naples and Epyc/Rome Server/HPC CPUs than Nvidia can produce on GPU sales. So AMD really does not have to be worried about spending any extra beating Nvidia for the flagship GPU crown. More GPU revenues come from the mainstream GPU market than the flagship GPU market anyways and Nvidia has lost over 50% of its market value on illogical fears after the coin mining market crash. AMD, once its Epyc server revenues take primary notice among the investor community as AMD's real strength, will no longer have to worry about any GPU market fluctuations tanking its market cap.

Nvidia only has GPUs mostly, Nvidia lives or dies by GPUs while AMD lives on with its CPUs with GPUs just being AMD's second business. AMD could if it gets enough Vega 20 bins create another Dual DIE/Single PCIe card SKUs and probably take the flagship crown for a short while and why not give another gaming/prosumer Dual GPU/Single PCIe card SKU a try if the defecrive Vega 20 DIE/Bin is filling up the result of Vega 20 die production that always has some DIEs that do not make the grade with enough working nCUs.

I'd also like to know AMD's current integrated graphics market share as that's mostly Vega based now and AMD has to be increasing that with every new Desktop/Mobile Zen/Vega APU sold, consoles not included.

January 17, 2019 | 09:36 PM - Posted by isit3dxpointinhere,orareujusthappytoseeme (not verified)

While i tend to agree, from high end enthusiasts perspective, as I also am eagerly awaiting RTG to compete at the top, I do see this Radeon 7 release as a huge positive however. Moreover, I see it as possible confirmation that RTG will indeed compete at the top at some point in the future, most likely with the Navi.

An msrp of $700 tells me that they (RTG) have something else up their sleeve (Navi?), for which they feel they can charge MORE than $700. Imo, the announcements of Zen 2, Sunny Cove, along with ssd advancements (optane baby!), this year will be very interesting for hardware enthusiasts, and the Radeon 7 does nothing if not stoke an already well fueled fire.

January 18, 2019 | 09:59 AM - Posted by NoMagicBulletButBillionsOfDollarsCanHelpAMDToCompeteInGPUs (not verified)

The MSRP on Radeon VII says that Nvidia's MSRP's and ASPs are high enough that AMD can comfortably price its first 7nm GPU closer to its Nvidia competition. Vega 20 at 7nm and some tweaks to the Vega GPU micro-arch in addition to 4 stacks of HBM2 with a 1TB/s bandwidth helps greatly and 16GB of total VRAM is not really going to be taxed by current gaming titles.

Navi is up RTG's sleeve but it's currently not for any $700+ market! Those Navi DIEs are going to be smaller and replacing the Polaris line of Mainstream Market GPUs. As if really AMD has any time to waste with Flagship GPUs and that small market size compared to the mainstream GPU market's larger size.

AMD better damn well have some discrete mobile Navi GPUs with at least 4GB of HBM2 offerings also on that TSMC 7nm process node. The laptop market needs more discrete mobile GPU competition. And currently the only Vega discrete mobile SKUs that offer 4GB of HBM2 are custom/OEM designs made by AMD for Apple's MacBook high end laptop SKUs. AMD needs to double down on getting its latest GPU micro-arch in a full line of discrete mobile GPU/HBM2 offerings what with Vega/Newer supporting that HBCC/HBC IP that can turn HBM2 into a lsat level GPU VRAM Cache to a larger pool of Virtual VRAM paged to/from regular system DRAM.

Vega's HBCC/HBC IP is ready made for laptop content creators by giving those users as much VRAM as there is physical VRAM + System DRAM available for non gaming graphics workloads that make use of super high resolution images/textures that are well in excess of 4GB in size. So to date Apple is the only maker getting access to that Vega HBCC/HBC-HBM2 IP on its laptop devices and that's a big attraction for content creators even at Apple's over the top retail markups. Navi should have all the IP/features that Vega offers and then some but we are still in wait for Navi to arrive mode to actually find out.

Post new comment

The content of this field is kept private and will not be shown publicly.
  • Lines and paragraphs break automatically.
  • Allowed HTML tags: <a> <em> <strong> <cite> <code> <ul> <ol> <li> <dl> <dt> <dd> <blockquote><p><br>
  • Web page addresses and e-mail addresses turn into links automatically.

More information about formatting options

This question is for testing whether you are a human visitor and to prevent automated spam submissions.