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

Navi Rumours... should i really take these seriously?

Discussion in 'Graphics Cards' started by alex crossland, Dec 21, 2018.

Tags:
  1. LePhuronn

    Mobster

    Joined: Sep 26, 2010

    Posts: 3,640

    Location: Stoke-on-Trent

    I don't see any reason why not. I think it was David Wang who said that post Vega they were looking at all memory technologies and would apply whatever worked best. It's clear that compute cards benefit from HBM, but pure gaming cards not so much. Now that Koduri is gone, Lisa Su has a clear long-term plan and the balls to carry it through and AMD aren't quite as skint as they were, I don't see any reason why Navi wasn't designed with a GDDR6 controller from the outset. Hell, AMD might have even designed Navi to work with both GDDR6 and HBM if there's a compute-focussed Navi in the future (although to be honest I doubt that, more likely that's what this "Vega 2" is all about - Navi is a short-term gaming architecture).
     
  2. henson0115

    Hitman

    Joined: Dec 6, 2013

    Posts: 974

    Location: Nottingham

    Actually the amd keynote specifily mentioned 7nm GPU for gaming.
     
  3. Beren

    Mobster

    Joined: Jul 30, 2004

    Posts: 2,816

    Location: Auckland

    I bought a 590 yesterday. It was a good price and the three games made it a great deal.

    I think we are looking at 6 months for a card that is going to cost a bit more than what I just paid.
    BUT I currently game at 1080p I am going to be able to play every game out there on max settings for probably 24 months.
    I have been out of gaming for an age so the three games is also a great buy for me as I know I want 2 of them.

    At that point I will go to a 4k display and then to a card that will run 4k at high.

    So for me this was a purchase I am happy with... For you, who knows.
     
  4. 4K8KW10

    Soldato

    Joined: Sep 2, 2017

    Posts: 5,612

    You mean these leaks:

    [​IMG]

    It is not anything special, even in the slightest.
    Given the weak Vega 10 chip performance, a direct shrink to 7nm will be very close to the above estimates. Without any architectural improvements which should be present there in Navi.
    This is why Vega 64+15% is low.
     
  5. Kaapstad

    Man of Honour

    Joined: May 21, 2012

    Posts: 27,750

    Location: Dalek flagship

    Adored TV

    This is the guy who got the Turing cards completely wrong then tried to defend it.

    I am hoping the next round of AMD cards are going to be good but I won't take notice of anything from sites like the one above.

    I also think the recent RX480, 580, 590 Vega 56 and 64 have all been good cards and think the company can build on this.
     
  6. 4K8KW10

    Soldato

    Joined: Sep 2, 2017

    Posts: 5,612

    Or maybe it isn't that bad. I am looking from the Vega 64 successor view point. But the leak means that Navi 10 replaces Polaris 10/20/30.
    Then, what replaces Vega 10?
     
  7. LePhuronn

    Mobster

    Joined: Sep 26, 2010

    Posts: 3,640

    Location: Stoke-on-Trent

    Well yes, in price point terms these replace Polaris. In performance terms, 3080 and 3070 replace Vega 10, 3060 replaces Polaris, but these would be considered merely sidegrades then.

    As to superseding Vega, AdoredTV made mention of his sources glimpsing a "Big Navi" which was "impressive" in some unquantified manner. I don't think 7nm Vega will see a gaming incarnation, but it is up in the air as to what "Vega II" actually is, if the trademark is an active one and not a concept AMD binned off.

    But then, if these leaked specs prove to be accurate, there's no reason to buy an RTX 2070 or lower and AMD could potentially clean up in the midrange market segment. So would there be any need to have a true Vega 10 successor before a completely new arch with Arcturus? Can AMD actually take on the RTX 2080 given we still don't know they can hit/beat GTX 1080 Ti performance?

    I don't game higher than 1080p so I'm quite stoked at the concept of a fully AMD 3000 system, but it's Arcturus that has my attention if I'm honest.
     
  8. 4K8KW10

    Soldato

    Joined: Sep 2, 2017

    Posts: 5,612

    It can be that Navi 12 is 150W, a 150-200 sq.mm chip with the performance of Vega 64 + 15% for £200-£250, while Navi 10 is the so called "Big Navi" that would cost £700 and is between RTX 2080 and RTX 2080 Ti.

    Polaris 12 is 103 sq.mm and only 50W in RX 550.
    Polaris 11/21 is 123 sq.mm, 75W in RX 560.
    Polaris 10 is 232 sq.mm, 150W in RX 480.
    Polaris 20 the same but 185W in RX 580.
    Polaris 30 the same but 175W in RX 590.
     
  9. LePhuronn

    Mobster

    Joined: Sep 26, 2010

    Posts: 3,640

    Location: Stoke-on-Trent

    Look at your own image post 4K8K :p if those Adored leaks are accurate then Navi 10 is in the RX 3080.

    But it is possible that the 3080 uses a cut down Navi 10, with a full fat version powering something bigger. In fact I'd be inclined to think that is the case given there's not a great performance difference between a full Navi 12 (RX 3070 as leaked) and that Navi 10.

    I'm not sure it would sit between the 2080 and 2080 Ti though. The 2080/1080 Ti is already Vega 64 + 30% and the 2080 Ti is +30% of the 2080. That is a lot of performance to make up. But if that is a cut down Navi 10 in the RX 3080, then a full Navi 10 probably could match the 2080/1080 Ti.

    But why $700? That's the same price roughly as a RTX 2080, so why bother? Also $700 doesn't fit the pattern of those 3 leaked Navi prices either. If those leaked prices are relatively accurate, I would expect a full Navi 10 (RX 3090?) to be about $450-500. Maybe throw another $75 on top for 16GB VRAM :p
     
    Last edited: Dec 30, 2018
  10. bru

    Soldato

    Joined: Oct 21, 2002

    Posts: 6,834

    Location: kent

    It does make you wonder, if the 7nm node is really going to be this good for GPU's and the rumours for AMD's next gen certainly seem very favourable, it makes me wonder what NVidia could do with it?
     
  11. 4K8KW10

    Soldato

    Joined: Sep 2, 2017

    Posts: 5,612

    I am arguing that his leaks are not accurate. Navi 10 should be the biggest chip. Navi 20 should be its second version. Arcturus is a generation beyond, probably on 7nm+.
     
  12. TaKeN

    Capodecina

    Joined: Aug 18, 2005

    Posts: 13,043

    Location: Shropshire

    Add a few Extra Zeros to the end of the price?
     
  13. LePhuronn

    Mobster

    Joined: Sep 26, 2010

    Posts: 3,640

    Location: Stoke-on-Trent

    OK, that makes no sense. Look at the 3 cards again. Navi 10 IS the biggest chip; the most powerful of the 3 cards has Navi 10 on it. If there will only be 3 Navi cards then Navi 10 is the biggest chip. If there is another card to come then it's likely that it'll be a full fat Navi 10, because there is no chance in hell that Navi 10 dies are coming out perfectly, and that perfection is only 20% more performant than a full Navi 12.

    Also, why should Navi 10 be the biggest chip? Who says? What, just because Vega 20 and Polaris 30 didn't appear until a node shrink? If Navi 12 and Navi 10 exist simultaneously, why can't there be a Navi 20 too? Because Arcturus is coming 2020 on 7nm+, so there isn't time for a 2nd generation Navi.

    Yes, there's been rumblings of a "Big Navi" and "Navi 20" but the only mention of it I can recall is the same AdoredTV source as the Ryzen/Navi stuff who said "yeah, I've seen a bigger Navi running and it's immense". That's even more tenuous than the Ryzen and Navi leaks.
     
  14. 4K8KW10

    Soldato

    Joined: Sep 2, 2017

    Posts: 5,612

    But with very poor performance for a 7nm chip. Remember we expect at least 70-80% performance uplift from the node change alone.
    I think that his leaks are not accurate - he just shifted the performance to the left. Claiming Navi 12's performance on the Navi 10.
     
  15. Davedree

    Mobster

    Joined: Mar 27, 2010

    Posts: 2,952

    All you have to do is look at the consumer market cancelled Polaris 40/44 (Scorpio).
    Revisit the thread on here when we discussed reasons why it was cancelled.

    Then I predict it will be redrawn onto the the 7nm with tweaks to the cache and memory controllers for Gddr6, and should be around 200-230mm2 die size without needing a 384 bit bus or 12 memory controllers, to keep the bandwidth up.
     
  16. LePhuronn

    Mobster

    Joined: Sep 26, 2010

    Posts: 3,640

    Location: Stoke-on-Trent

    Now you're just getting ahead of yourself.

    7nm shrink is either half the power or 1.25x performance, not both. This has been said countless times. These Navi leaks are saying AMD chose the former in this instance. And if AMD can get some kind of sanity back to the mainstream market by offering GTX 1080/RTX 2070 levels of performance for about $300 on a single 8-pin PCI-E connector then I say fair enough.

    I'm excited about what AMD are going to do in 2019, and in a perfect world I know what I want and I would love to see, but it seems to be you're almost foaming at the mouth here, twisting what you want to happen into some kind of exaggerated certainty. "the leaks are wrong because it WILL be 1000% more than that, because that's what it should be!"
     
  17. 4K8KW10

    Soldato

    Joined: Sep 2, 2017

    Posts: 5,612

    1.25x performance means every single transistor on the dies should be identical. A direct optical shrink.

    When moving to a new process, the goal is to double the transistors count, as seen from the RV730 to RV740 transition in 2009:
    [​IMG][​IMG]
    [​IMG]

    You see around 70% improvement going from 55nm to 40nm.

    If Navi 10 indeed offers Vega 64 + 15%, this means it is Polaris 10 doubled.
    Which means that again AMD leaves the high-end market empty.
     
  18. LePhuronn

    Mobster

    Joined: Sep 26, 2010

    Posts: 3,640

    Location: Stoke-on-Trent

    Cool graph bro, all that does is corroborate the performance increase side of the 7nm shrink, of which nobody ever disputed. But you're ignoring the other option of the shrink and that's to slash power requirement in half. Again discussed in AMD's slides and nobody has ever disputed. And, as I said, the RX 3000 leaks as presented indicate AMD opted for the 50% power approach to the 7nm shrink.

    And this is suddenly news to you how? From the very beginning AMD said Navi was pitched at the mainstream, and that narrative has only been reinforced by the Navi leaks. Now whether that "mainstream" refers to performance (i.e. cementing 2070/1080's raster performance as the new mainstream benchmark) or price (Navi occupies the £250 and less market segment) matters not. Navi is a mainstream arch. There is no super mega halo product.

    Don't get me wrong, with all the talk and rumours of "Navi 10" for so long, I was surprised when the AdoredTV leaks actually said Navi 12 as well. This does indicate to me that the Navi 10 in the RX 3080 is not the full chip, so there is potentially something bigger available (RX 3090 for argument's sake). Now you can want AMD to out of the blue stomp a 2080 Ti into the dirt with a mythical $700 Navi 20, but there is nothing out there to even suggest that is going to happen. You can want AdoredTV to be wrong all you want and Navi 10 is actually some kind of messianic figure, but it's not going to happen. Yet.

    Navi has never been associated with top-end. Ever.
     
  19. 4K8KW10

    Soldato

    Joined: Sep 2, 2017

    Posts: 5,612

    AMD desperately needs to make a 400 sq.mm 7nm GPU that is within 5% or similar of the RTX 2080 Ti.
    And to call this very GPU a new mainstream.
     
  20. 4K8KW10

    Soldato

    Joined: Sep 2, 2017

    Posts: 5,612

    Well, I can guess: they will shrink and double GP104 with its 314 sq.mm, call it the new 3080 and with 10-20% higher performance than RTX 2080 Ti.
    This is what is gonna happen. As usual.