1. This site uses cookies. By continuing to use this site, you are agreeing to our use of cookies. Learn More.

So just how are AMD going to shoot themselves in the foot with the NAVI launch?

Discussion in 'Graphics Cards' started by bru, May 1, 2019.

  1. Calin Banc

    Hitman

    Joined: Aug 14, 2009

    Posts: 788

    Just the usual: expected speeds were reached with the first samples, but for final ones, they need to push the power up and basically AMD engineers want to get this out ASAP so they can finally move to something else.
    It may turn out to be a Vega 2.0, but hopefully more competitive in price.
     
  2. Phixsator

    Mobster

    Joined: Oct 10, 2012

    Posts: 3,017

    i'm not so sure i believe that bold part that much, something about it seems odd to me, timing and all.
     
  3. ilc72

    Gangster

    Joined: Mar 19, 2014

    Posts: 446

    I loved this video, so if the GTX1080ti was it’s own brand it would of taken 35% market share!

    Interestingly whatever will happen with Navi, allegedly NVIDIA announce price cuts to their RTX cards at Computex. NVIDIA is also furiously working to get 7nm to market for their consumer cards since despite 16/12nm being a mature process, yields on a 754mm2 die still aren’t great! Seeing a 1.8x improvement in transit or density at the same power/performance would make the bill of materials cheaper.

    Also the quote that NVIDIA feels their process is better than 7nm is that they were tied to Samsung rather than TSMC, the 7nm process from TSMC is markedly better than Samsung and yields are better than could be expected at this stage of the node cycle.

    Navi as a first release may not tear up the charts, but “Arcturus” was originally thought to be the successor to Navi is fairly nailed on to be the custom Navi variant used in the next-gen consoles rather than a new generation. IMHO the initial Navi release will probably not have the Ray-Tracing, but we will see a higher performance Navi 20 in 2020.

    Finally the Radeon VII is not necessarily a total disaster, why else would NVIDIA actually cave and release a Creators driver for the RTX cards, something they’ve never done to prevent cannibalising Quadro sales.

    At the end of the day the ultimate performance is for the 1% that frequent this forum, rather than the mainstream market, if Navi can deliver a 10-15% performance bump on Vega 64 at a sub £300 price point it will be a success. If it’s nearer £250 then even more so.
     
  4. Grim5

    Hitman

    Joined: Feb 6, 2019

    Posts: 934

    35% of GN's viewers yes - it's steam share is much, much, much lower. Though who cares about the steam survey anyway - we should have a OC UK survey
     
  5. Th0nt

    Soldato

    Joined: Jul 21, 2005

    Posts: 6,134

    Location: N.Ireland

    Its a valid observation that the market will be made up heavily of low end cards and integrated graphics when you spread out into real world data. The consoles hardware would be in this segment too tbh. The reality of what % is made up of high end dedicated gpu's would be rather small.
     
  6. Calin Banc

    Hitman

    Joined: Aug 14, 2009

    Posts: 788

    GCN, in the current trend of PC Games, can't compete with what nVIDIA has at the speeds GCN will actually be ok, power wise. You really need to put the work, including low lvl APIs to get the best out of it and AMD doesn't have the will (an perhaps budget + influence) to help developers with the PC variants. At least not as much as nVIDIA. When 7970 and 7950 launched, they overclocked well, because they were actually running at a decent speed/power ratio. nVIDIA didn't had anything at that time out to compete, so AMD could have on the market such cards. Then gtx680 came in, so AMD launched 7970/7950 GHz Ed and it went full power from that point on. For instance, the difference between an underclocked/undervolted r290 and overlocked/overvolted one (not that much, still intended for regular 24/7 use), for me was about 20-25% (in performance), but power draw was almost double (as per GPU Z), between those 2 cases in my limited testing at the time.

    So with that, since you can't really do a lot with GCN in its current form, the only thing you can actually do is finishing the projects you have and move to something new - and hopefully better. Ergo, that statement can be valid/ plausible.
     
  7. Grim5

    Hitman

    Joined: Feb 6, 2019

    Posts: 934


    Comparing GCN to Intels skylake ringbus would make for a great comparison

    Intel designed its cpus for 4 cores, due to market conditions they had to start scaling it up, because they are now operating outside the design of the product efficiency goes out the door and heat comes with it.

    It’s the same with GCN, it’s being pushed way past it’s intended design

    It’s not like companies don’t know the product limits they just failed to move with the market. In Intels case they did not expect AMD would put out such a great cpu line up so quickly that push out cores coming out of the air.

    In AMDs case I think it’s a financial problem. They couldn’t pony up the RnD cash required and so had to stick with GCN far past it’s use by date
     
  8. Phixsator

    Mobster

    Joined: Oct 10, 2012

    Posts: 3,017

    Sure the statement could be valid, absolutely, my problem with it however is the timing of it. I find it weird that this news about the engineers being tired of navi is only now being mentioned, now that we are getting so close to a possible release. It's almost like someone decided to sandbag. It's all guessing and speculating on my part however and it's certainly going to be interesting to see what will be released eventually.
     
  9. Th0nt

    Soldato

    Joined: Jul 21, 2005

    Posts: 6,134

    Location: N.Ireland

    Always wait for the actual results once people can get hold of it. The trash talking or fudding is the bane of modern era garbage before events whichever walk of life you apply it to.
     
  10. bru

    Soldato

    Joined: Oct 21, 2002

    Posts: 6,873

    Location: kent

    Well the latest rumour has the pricing at $499 and $399 respectively for the RX3080 and RX 3070.


    Which converts to £470 with VAT and £378 with VAT respectively.
    when you look at the RTX2070 founders edition at $599/£549 that doesn't seem quite so bad, but then there is the reality that you can buy a 2070 here on this very store for£420.

    My basket at Overclockers UK:
    Total: £419.99 (includes shipping: £0.00)
    [​IMG]
     
  11. ANDREW GREEN

    Gangster

    Joined: Sep 17, 2018

    Posts: 375

    But then maybe Navi has similar discounts, early days
     
  12. D.P.

    Caporegime

    Joined: Oct 18, 2002

    Posts: 29,965

    More liekly due to limited supply and price gouging prices will go up.

    AMD are not going to discount a product they have just released. That means they set the wrong RRP.

    Nvidia can reduce prices because the product has been out in the channel nearly a year, yields are high, pascal stock has cleared out allow lower price points and the initial demand is now waning.
     
  13. ANDREW GREEN

    Gangster

    Joined: Sep 17, 2018

    Posts: 375

    If supply is that limited and they sell them at that price then they're setting the right price for their profitability even if it isn't what we personally want.
     
  14. LePhuronn

    Mobster

    Joined: Sep 26, 2010

    Posts: 3,747

    Location: Stoke-on-Trent

    Was the Sapphire rep talking US dollars or Singaporean dollars? Am I confusing multiple things in my head, or was this Sapphire rep in/from Singapore when he made the claims?

    Because Singaporean dollars changes the game: $499 - 7% sales tax is about £265
     
  15. bru

    Soldato

    Joined: Oct 21, 2002

    Posts: 6,873

    Location: kent

    That's a good point, could be.
    Not long to go now and we will see whatbis what.
     
  16. D.P.

    Caporegime

    Joined: Oct 18, 2002

    Posts: 29,965

    LOL, so wishful some of you.

    At every release in recent years AMD have set prices for the mid and high end cards that equal Nvidia's. FuryX slower than 980ti but same RRp and more expensive on the street. Vega 64 vs 1080. Radeon 7 vs 2080.


    OF course AMD are going to price these at $499 for 2070 performance. The 7nm node is very expensive.
     
  17. LePhuronn

    Mobster

    Joined: Sep 26, 2010

    Posts: 3,747

    Location: Stoke-on-Trent

    But D.P., that's not the argument. Again, the argument is AMD shouldn't because they can't. How do you propose AMD regain marketshare and mindshare if they charge the same money for an inferior product? You yourself and others have said repeatedly that AMD producing cards at many-year old performance levels is a joke, and charging Nvidia money for it doubly so. The cost of 7nm is irrelevant without context. Yes, as a process it is expensive, but economy of scale, yields, size of the dies and percentage of usable dies from the wafers can all mitigate that initial cost.

    Why do you think the latest leak has such a granular approach to SKUs? AMD could almost literally be using every die that comes off a wafer and slap it onto one or two common PCBs and make a SKU out of it. Getting 70% utilisation out of a wafer makes each wafer cheaper in the long run than, say, 40% utilisation.

    Wishful thinking is perfectly fine, and if AMD do indeed charge what we all consider to be stupid prices then Navi will fail, nobody will buy it and AMD are all out of excuses. I sure as hell won't buy a RX3080 if it performs and costs the same as a RTX 2070, because RT and DLSS for all its woes would be a deal-clincher.
     
    Last edited: May 22, 2019
  18. D.P.

    Caporegime

    Joined: Oct 18, 2002

    Posts: 29,965

    You are asking two thing: 1) what should AMD do, 2) Yields and pricing.

    Answer 2) first as it is simpler. Everything you say applies equally to Nvidia. There is a reason Nvidia have SKUs from Titan->2080ti->2080->2070->2060 and why there may well be things like a 2070Ti, and why there were 2 chip binning for some of these SKUs. The problem AMD faces is 7nm is 2-to3x more expensive than 16/12nm, and yields are lower. AMD's problems are compounded because they do have a smaller market share so have less volume discounts, and the 7nm process is much more in demand with mobile SoCs, while TSMC has spare 12nm capacity.


    For the first question, I don't have the answers. If I did I would be earning 500k a year as a C-level in AMD. The problem is AMD catastrophically cut R&D and this has a delayed and compounded effect that is hard to come back from. I mentioned this way back at 290x release that progress will totally stagnate, which is exactly what happened. Nothing but a serious of minor tweaks and node-bumps, with more rebranding than you can count on your fingers (there is news Polaris will; be rebranded again with 600 series cards!).

    What is clear is if their GPU division is not profitable then it will die. Gaining market share but loosing billions is not sustainable. AMD have tried this, but with lackluster products their market share continued to decline and their profits evaporated. Which is why a few years ago they public stated they can no longer be considered a budget brand and would have to compete with nvidia at a high performance and high profit market position.

    The problem is AMD's lack of year-on-year R&D has meant their GPU products are several generations behind. This is forcing AMD to use a more expensive 7nm process, and forced them to wait another year before they could do anything to compete with nvidia, and despite that all they can achieve is performance parity.

    Lastly, just like Nvidia and every company, th only thing AMD cares about is profits. Selling at tiny margins to gain a small amount of market share is not sustainable, that is what got AMD in top so much trouble to begin with.

    IMO, AMD should be looking at selling off their GPU division but with somekind of cross-licensing in place so APU can be made. The GPU division needs a serious injection of cash to come back. Ryzen is doing well but AMD need to service debt and increase CPU R&D.
     
  19. LoadsaMoney

    Caporegime

    Joined: Jul 8, 2003

    Posts: 27,048

    Location: In a house

    Yup.
     
  20. LePhuronn

    Mobster

    Joined: Sep 26, 2010

    Posts: 3,747

    Location: Stoke-on-Trent

    If the next gen arch next year is a flop then I'll be fully in agreement. I get what AMD were trying to do when they bought ATi, but with the CPU division falling on its ass as well it just messed the company up too much to make good on their plans in any reasonable time frame.

    Give Lisa Su her props though, she seems to have worked some magic and gotten momentum back, but "Next Gen" will be entirely on her watch, not leftovers from RTG still trying to be independent. If her magic touch doesn't extend to graphics as well come Q1 2021 then perhaps it might be time to cut it loose again.

    Unless, of course, graphics is still sustainable and doesn't require propping up with CPU profits.