Will the new Radeons on 7nm compete with Nvidia?

Discussion in 'Video Cards' started by Agamen, Dec 19, 2018.

  1. Shadowed

    Shadowed Limp Gawd

    Messages:
    461
    Joined:
    Mar 21, 2018
    The upper mid range is where the sales are anyways. If they can get a card close to 2070/1080 performance at a significantly lower price, then they are gonna get some of that desperately needed market share.
     
    N4CR likes this.
  2. Boil

    Boil [H]ard|Gawd

    Messages:
    1,224
    Joined:
    Sep 19, 2015
    You mean like the rumored Radeon RX 3080...?!?

    amd-rx-3080-1480x833-png.png
     
    Shadowed and Maddness like this.
  3. N4CR

    N4CR 2[H]4U

    Messages:
    3,427
    Joined:
    Oct 17, 2011
    That's exactly where I expect the top small navi. V64 +15% is faster than a 1080, it contradicts the 2070/1080 mark estimate on the bottom line, I don't expect it to go that fast out of the box, more like V56-64/2070/1080 area of speed. Many tweaked vegas will do +15% anyway.

    It's still GCN...
     
  4. Pieter3dnow

    Pieter3dnow [H]ardness Supreme

    Messages:
    6,362
    Joined:
    Jul 29, 2009
    Yeah a lot of people are stuck on GCN ;) Does not matter much.
    The thing is that Vega is harder to produce and larger in die size that is why we got Navi and better power consumption to boot...
    There are still many things different from Vega.
     
    Maddness likes this.
  5. Hagrid

    Hagrid [H]ardness Supreme

    Messages:
    7,993
    Joined:
    Nov 23, 2006
    I don't think they will surpass a 1080ti, but maybe at least they will work and not play any space invaders?
    I wish they would bring a high end card to the table, but I doubt it. Sucks.
     
  6. tajoh111

    tajoh111 Limp Gawd

    Messages:
    203
    Joined:
    Jan 26, 2012
    I can see GTX 1080/Vega 64 for their 250 dollar product but not at 150 watts. And this is still on the optimistic side of expectations given what 7nm has brought to the table for Vega 20. 20% more performance for the same power on 7nm.

    In some ways, Navi has even more going against it for improvement than vega 20. This is because cost have to be controlled and this put some restrictions in the design of the chip.

    Because this is a midrange part, it will only have 32rops on a 256 bit bus. This severely limits how much shaders they can put on the chip before running into bottlenecks. How this hurts the performance expectations is two ways. They have to use either a wider, larger more expensive die which will be bottlenecked by the ROPs or use a smaller die, clocked higher which will eat into the performance per watt.

    Considering this is still GCN and this is the 5th or 6th iteration of it, it is a a very mature architecture and as a result, there is less room for improvements.

    IF the rumors don't come to fruition, which is likely, considering all the rumors paint AMD basically throwing money in the garbage by not increasing their margins(because the above price/performed would lead to sell outs), it leads me to confirm 1 more things along with what Kyle has said.

    Kyle indicated his rumors mostly aligned with Adored. Afterwards AMD contacted Adoredtv asking how people were reacting to the rumors, not who the leak was. To me that means more than anything that AMD planted these rumors as part of their guerrilla marketing tactics and was asking for feedback on how to shape and make these rumors more effective in the future.

    It's like a thief sending a friend to the scene of the crime and not asking the most important question(who did it) but rather asking what clues or mistakes were left(so that the thief could better improve their theft next time).

    As we can see from how popular these rumors are, they are an effective marketing tool for AMD when they are not releasing new products. How many people are saying wait for Navi now which is a familiar mantra which we have heart anytime Nvidia releases something with nothing from AMD in the competing market.

    The sad thing is, the same occurred with wait for Fiji, wait for Polaris, wait for Vega. Considering the performance of the actual product has performed at the bottom rung in terms of expectations and well below that of rumors(check the previous rumors/expectations thread), you would have expected people to smarten up by now. But nope. AMD is just making the rumors more juicy this time so people can't resist.

    The irony of situation is Adored asked the consumers to believe his rumors because if they don't, we will make the higher prices come true. Considering AMD has let us down compared to the hype 4 times as far as performance if we include Tonga, it's reality that should be making us question why these price/performance are too good too be true along with power and performance metrics.
     
  7. Pieter3dnow

    Pieter3dnow [H]ardness Supreme

    Messages:
    6,362
    Joined:
    Jul 29, 2009
    If you crunch some numbers on Navi with 12nm vs 7nm process you would get something near 300 Watt and the performance slightly below Vega56 . Those numbers do not seem that terrible at all but nothing special either. And that is why the 150 Watt number should not be that far fetched and the rest of the line up (Navi 12) does reflect the same power usage nothing that screams impossible...
    AMD wants to sell cards it has a few things going for them but one of them not so much , marketing is AMD their weak spot.
    No one ever said wait for anything it is the people that are driving forums with their own opinion. I would even suspect that some of the Nvidia guerilla marketing does this to ridicule AMD fans by rubbing it in badly after people waited for so long and felt let down.

    Why as an adult would people be led by rumours this whole argument reeks. It does not matter how juicy because no one that has bought a RTX card is looking for something as Vega +15% performance. The only thing that is worthwhile is the price performance ratio which is incredibly good (shocker for people who buy AMD?).

    Rumours or not why would someone as AdoredTV let himself be used for purposely revealing false information and why would AMD purposely lead AdoredTV on that is a recipe for burned bridges. It is the biggest lose lose scenario no one wins this makes no sense whatsoever .....
     
    euskalzabe and Maddness like this.
  8. Factum

    Factum [H]ard|Gawd

    Messages:
    1,485
    Joined:
    Dec 24, 2014
    AMD’s problems are far bigger than “Poor Volta”.
    They are lacking an entire ecosystema behind their hardware compared to NVIDIA.

    So to sum up:
    -Bad marketing
    -Hardware consuming more power for less/same performance.
    -The lack of a software eco-system (CUDA is thaugth at schools eg.)

    So the marketing, the hardware and software is a problem for AMD...and AdorkTV is just AMD PR to me...
     
  9. Maddness

    Maddness [H]ard|Gawd

    Messages:
    1,159
    Joined:
    Oct 24, 2014
    I think almost everything PC parts related on YouTube is PR related. I very occasionally will watch something but always take it with a grain of salt. None of them stick there neck out. It's always I've heard from my guy inside. Most of what they say has already been rumors on some other site, so it's really second hand anyway. That's my main reasons for coming to sites like this. I'm personally so glad that Kyle doesn't consider YouTube to get his message across.
     
    Auer likes this.
  10. polonyc2

    polonyc2 [H]ardForum Junkie

    Messages:
    15,950
    Joined:
    Oct 25, 2004
    AMD is back...7nm will be the new 9700 Pro
     
    dvsman, Krenum and Maddness like this.
  11. bobzdar

    bobzdar [H]ard|Gawd

    Messages:
    1,576
    Joined:
    Jun 6, 2003
    What's interesting is that at lower clocks, vega becomes competitive. For example, laptop vega56 at 120w competes with 110w gtx 1080 maxq performance wise. I think they just pushed the clocks too high in the desktop parts in order to match desktop Pascal, well past where they're efficient. Clock for clock vega is actually faster, it just hits a wall around 1500mhz and sucks a ton of power beyond that. 7nm may fix a lot of that.
     
    euskalzabe and Auer like this.
  12. tajoh111

    tajoh111 Limp Gawd

    Messages:
    203
    Joined:
    Jan 26, 2012
    What numbers can we crunch?

    We have no Navi power consumption figures on 12nm, so how can we crunch numbers for 7nm's? The only figures we have to go on, is GCN current iterations with Vega and Polaris and the numbers we have for Vega 20 which is already 7nm's. And if we take these numbers, it does not paint a favorable picture towards those expectations.

    Take Polaris 30 or the RX 590 and you have a card that already consumes 210 -225watts at 1565mhz with performance 60% of a GTX 2070 and we have Vega 20 on 7nm which would have about 100% of the performance of an RTX 2070 going off Hardocps numbers at 300watts.

    Considering the process is only giving a 20-25% uplift in performance at the same wattage, applying 7nm above for polaris 30 atleast means currently Polaris 30 performance at 110watts(if the 50 percent power reduction at same performance) or slightly slower than a GTX 1070 at 210-225 watts(20-25% more performance at the same wattage).

    Similarly Polaris 20 at 1800mhz which has performance comparable to RTX 2070 if we assume performance scales with clocks mostly is already consuming 300watts, twice the 150 watts stated in the rumors. But it does this with 7nm's and HBM2 helping it. To get to Vega 64 performance + 15%, AMD has to double their performance per watt vs Vega 20 through pure architecture since Vega 20 is already 7nm's. That's almost impossible particularly since most of this performance deficit will have to be achieved with clocks because a wide architecture(more Cu,s more rops, bigger bus) makes the chip larger and more expensive and not midrange anymore, certainly not $250. HIgh clocks eats into that power budget makes that 150watt figure impossible.

    Every iteration of Polaris and Vega past the RX 480 are getting increasingly worse for performance per watt. The only way a chip the size of the RX 480 gets down to 150watts in a mass production setting(not chip to chip underclocking/undervolting), is by giving it conservative clocks and lesser performance.

    I think something along the lines of Vega 56 at 150 watts is possible. This would be a 50% uplift in performance per watt vs RX 480(Vega, RX 580, RX 590 all have worse performance per watt than the RX 480) and this is not a given.

    Looking at the performance per watt jump between the 270x and RX 480, the jump in performance per watt is 35%.

    perfwatt_1920_1080.png

    So a 50% jump in performance per watt is not even a given considering that the improvement between 28nm to 14/16nm finfet is greater than the transition between between 12/14/16 nm to 7nm.

    https://www.anandtech.com/show/11337/samsung-and-tsmc-roadmaps-12-nm-8-nm-and-6-nm-added

    https://www.anandtech.com/show/12677/tsmc-kicks-off-volume-production-of-7nm-chips

    AMD only managed a 35% jump between 28nm and 14nm on the same chip size which is why a jump to performance along the lines of 50% over an RX 480 or Vega 56 performance is something reasonable and a tad optimistic for 150 watts. Getting another 30% more performance on top of Vega 56 when GCN power consumption climbs very quickly is not very likely at 150watts.

    GCN is very mature at this points and the area for improvements are becoming less and less much like Intel's architecture for the last decade or even the turing series from Nvidia.

    And as far as noone looking at a RTX card is looking for Vega + 15% more performance is BS. That's the performance of the RTX 2070 which is likely the best selling card in the whole turing lineup.

    The reason why Adored would let himself get used is people have short memory and he likes the hits and views. He made similarly optimistic predictions about Vega(initially), polaris and Fiji. People forget how wrong he was about Polaris and Fiji. And AMD guerrilla marketing tactics are much smarter nowadays because the leaks come from people not tied to AMD directly anymore. That means they don't get their hands dirty and do it from a proxy on behalf of AMD. This is the best way for dirty work to be done. Doing it directly can have negative consequences which AMD learned the hard way.

    If anyone remembers, there used to be someone on these forums and really all tech forums called JFAMD or John Fruehe who hyped up AMD products, particularly bulldozer.

    https://forums.anandtech.com/threads/john-fruehe-jf-amd-no-longer-with-amd.2263923/

    As it turned out, lying to people and being directly tied to AMD in forums is a bad thing. People wanted to lynch him after and he also got fired. So how you can you use this form of marketing without the fallout? Basically do what AMD I suspect has been doing every since.

    Seed fake leaks and even if the products don't line up with expectations, there is no fallout because none of this can be linked to AMD. However, what you will notice is there are people that come on this board and other boards with new account that support these rumors and disappear shortly after.
     
  13. Pieter3dnow

    Pieter3dnow [H]ardness Supreme

    Messages:
    6,362
    Joined:
    Jul 29, 2009
    The reason why you have no numbers is because you fail to apply the process which can process numbers. 7nm characteristics turned back 12nm performance we know what the process improvement is and that would work for assuming performance.
    Like I said before Navi is not that special nor high end.
    That is the worst case of apples and oranges ever Navi is way different from Polaris and Vega. Maybe you should check AMD Horizon video where it is explained what the 7nm process is explained on power/performance.
    https://www.tsmc.com/english/dedicatedFoundry/technology/7nm.htm

    Here is where AdoredTV explains in detail how the performance and power consumption is for AMD on Zen2 process from AMD slides themselves.


    Density *2 Power 0.5 Performance 1.25.

    This is how you crunch numbers and find out if Navi is something special or not. Instead of what you did where there is nothing of value.

    Everyone but AdoredTV goes out of his way to correct mistakes or apologizes when his rants on stuff gone to far. He even did one on Polaris (explaining numbers). And nowhere does he ever claim that what he says is the absolute truth

    How would you know where the leaks come from in the first place and how would you know about the people contacting AdoredTV ?

    John Fruehe was a member of this forum as well. Where he was under the impression of where something would increase beyond what was logical and certain places he was made fun of rather then portrait as being a liar. That he lied about something he did not understand was not good but not in the way you portrait it. You pretend that everyone of AMD just lie and that is wrong. When you try and deal in absolutes then it becomes where I have to tell you fail to convince everyone of your point.

    In the end lying about things will catchup and will do the opposite of what you are trying to achieve an example of not lying but not being effective was AMD campaign where they took certain Vega features and try and sell the new Vega on points that mattered little or had to be implemented by programmers instead of being supported in the driver by default.

    I'll explain it slowly to you again:
    If AMD leaks numbers that are false then uses AdoredTV for leaking those false numbers and AdoredTV notices this what is his incentive to participate?
    If there is no incentive for people posting numbers that no one will take it serious any more.

    If there are no products that come close to what the leaks say then everyone loses AMD lost AdoredTV lost and people will not care about leaks from AMD that AdoredTV covers (the boy who cried wolf).
     
    Last edited: Dec 26, 2018
  14. Spoonie_G

    Spoonie_G Limp Gawd

    Messages:
    287
    Joined:
    Jan 10, 2007
    McDonalds makes the most money from selling food than other restaurants. That should only matter to folks who are shareholders.
     
    ZeqOBpf6 likes this.
  15. Pieter3dnow

    Pieter3dnow [H]ardness Supreme

    Messages:
    6,362
    Joined:
    Jul 29, 2009
    But it does not have any influence on the price (which was the comment about). Nvidia can ask any price and people will pay for it.
     
  16. Factum

    Factum [H]ard|Gawd

    Messages:
    1,485
    Joined:
    Dec 24, 2014
    So NVIDIA is a the only company in the world that is not afected by supply and demand?
    Take a step back and look at your hyperbole...
     
  17. NKD

    NKD [H]ardness Supreme

    Messages:
    7,415
    Joined:
    Aug 26, 2007
    My crystal ball is saying wait till CES!
     
    euskalzabe, N4CR, atp1916 and 3 others like this.
  18. NKD

    NKD [H]ardness Supreme

    Messages:
    7,415
    Joined:
    Aug 26, 2007
    if they launch mid range navi then you are looking at vega 64+ 10-15% performance. This is the reason I am leaning towards them launching navi. Rumors were Navis is shaping out to be better than they expected. So if its already vega +15% then there is not much incentive to launch vega 2 unless vega 2 has significantly improved in the IPC in gaming and competing in between 2080-2080ti. I am leaning more towards mid range navi announcement and launch earlier then expected since rumors were its shaping out to be better.
     
    Maddness likes this.
  19. Maddness

    Maddness [H]ard|Gawd

    Messages:
    1,159
    Joined:
    Oct 24, 2014
    This is also what i believe. They may also announce the new machine learning Vega, but i doubt out will be a consumer version.
     
  20. KazeoHin

    KazeoHin [H]ardness Supreme

    Messages:
    7,794
    Joined:
    Sep 7, 2011
    All I hope is that AMD does NOT follow the "30X0" naming system.

    Its an obvious and cheap trick.
     
    Maddness likes this.
  21. Pieter3dnow

    Pieter3dnow [H]ardness Supreme

    Messages:
    6,362
    Joined:
    Jul 29, 2009
    You afraid that millions of Nvidia users might buy this by accident? Hey it says 3080 on the box okay it must be Nvidia ;)
    If it sticks with the Ryzen 3000 series then I can't see that AMD has no valid opportunity for doing this on their graphics card.
    I had a very funny response to this one but I settled for the cold hard truth.
    https://www.cnbc.com/2018/11/16/nvi...er-as-demand-for-mining-chips-evaporates.html

    So when you made let us say 40 million extra chips that were intended for use in those mining GTX 1080/1070 and now start selling them as GTX 1060. And still no price drops on those new 1070/1080 cards?

    Or did I miss the parts that the 1070 and 1080 are selling way below msrp now ?
     
  22. XoR_

    XoR_ Gawd

    Messages:
    596
    Joined:
    Jan 18, 2016
    Navi with 2560 shaders and performance of RTX 2070 is a pipe dream.
    Polaris have 2304 shaders and 53% the performance.
    With merely 11% shader count increase they would need:
    - massively raise clocks - which anyone with any knowledge of how power consumption raise with clock (due to voltage requirements) stand completely againsts claims of this card having also low TDP.
    - massively raise shader performance - since AMD made no improvement between GCN in Fury X and Vega 64: https://www.hardocp.com/article/2017/09/12/radeon_rx_vega_64_vs_r9_fury_x_clock_for it seems unlikely...

    Technically 2560 shaders and 7nm node (and clocks it enables) is more than enough to beat 2304 shader 12nm RTX 2070 at the same TDP but not without serious architecture improvements.
    And then there is 250$ price tag which is just unrealistic....

    My most optimistic guess is 30% performance improvement to RX 580 with similar price tag and slightly lower TDP but still higher than RTX 2070.

    With 4096 shaders they could easily match and outperform 2070 but still 150W TDP doesn't seem at all realistic as is 250$ price tag. More like 400$ while being optimistic...
     
  23. Hisshadow

    Hisshadow n00b

    Messages:
    56
    Joined:
    Mar 11, 2017
    wish they would make a tremendous leap forward as amd did. For one, in their entire history just be #1 outrun a geforce and keep going...

    its possible, 3dfx had it happen to them, by nvidia..

    intel just got a taste of it lol

    I dont know.. can amd make a video card that fast? maybe they should let whoever does their cpu's... draw his own design for a gpu
     
  24. Factum

    Factum [H]ard|Gawd

    Messages:
    1,485
    Joined:
    Dec 24, 2014
    Your are just confirming my point...NIVIDIA is affected by supple and demand...just like all other companies...aka drop the hyperbole.
     
  25. Pieter3dnow

    Pieter3dnow [H]ardness Supreme

    Messages:
    6,362
    Joined:
    Jul 29, 2009
    That is not what I said ...
     
  26. Trmntr

    Trmntr [H]Lite

    Messages:
    103
    Joined:
    Mar 6, 2012
    Vega64 got a performance like 1080, I expect to see 2070 performances easly
     
    _chesebert_ and Sulphademus like this.
  27. Pieter3dnow

    Pieter3dnow [H]ardness Supreme

    Messages:
    6,362
    Joined:
    Jul 29, 2009
    And where is the logic connected to that statement ?

    I have seen people getting along fine , I here by demand world peace, not likely to happen is it ?
     
  28. Trmntr

    Trmntr [H]Lite

    Messages:
    103
    Joined:
    Mar 6, 2012
    I do not know if those new AMD cards named "3070" will be the new AMD high end, or just replace the old 580s/590s, but if there are releasing something along the lines of Mid-High/HighEnd they should be aiming for a perf like the 2070 to remaing competitive don't you think so?
     
  29. Pieter3dnow

    Pieter3dnow [H]ardness Supreme

    Messages:
    6,362
    Joined:
    Jul 29, 2009
    There in a different price bracket ;)

    Competitive does not always follow naming rules (Nvidia themselves break rules with naming for the same product that has physical different components).

    AMD might choose to stick with the 3000 series for the sake of sounding like it is part of the Ryzen series 3000 more or less branding rubbing of on the Navi series..
    Which is well within their right to do so .... And it makes sense as well..
     
  30. NKD

    NKD [H]ardness Supreme

    Messages:
    7,415
    Joined:
    Aug 26, 2007
    I hope they do to go along with ryzen 3000 naming. Unless nvidia owns those names there is nothing wrong with it. AMD can name them whatever they like as long as they can.
     
    Pieter3dnow likes this.
  31. /dev/null

    /dev/null [H]ardForum Junkie

    Messages:
    13,754
    Joined:
    Mar 31, 2001
    Yep, definately. Gotta spend at least as much on videocards as your desk.
     
  32. funkydmunky

    funkydmunky 2[H]4U

    Messages:
    2,198
    Joined:
    Aug 28, 2008
    Navi initially appears to be the Polaris replacement. It is rumored to be far ahead of schedule so we should see it announced with a spring release. That still leaves Vega2 7nm. Scrap it, release it for home market? If it can do 1080ti level I think AMD will launch it and hold on until high-end Navi ships later on.
     
  33. NKD

    NKD [H]ardness Supreme

    Messages:
    7,415
    Joined:
    Aug 26, 2007
    I think if vega 2 is performing in between 2080 and 2080ti then they will likely release it as high end. That avoids them having to leave a hole at the high end while they release navi. I think That could easily last until their next-gen card for 2020 launch. It will make total sense to have Navi at mid-range and vega competing with 2080 at the higher end.
     
    Last edited: Jan 1, 2019
    Boil likes this.
  34. funkydmunky

    funkydmunky 2[H]4U

    Messages:
    2,198
    Joined:
    Aug 28, 2008
    This makes sense to me. Even with the cost of HBM (when is this not going to be an issue?) I hope AMD can make Vega2 more affordable then what Nvidia is pulling.
     
  35. NKD

    NKD [H]ardness Supreme

    Messages:
    7,415
    Joined:
    Aug 26, 2007
    Adoretv has been right more time then you as far as I can tell. Pick on the guy all you want but he doesn't just bullshit. Did you miss the fact that he also leaked the RTX and the naming for NVidia turing cards? Common! Hate on the guy all you want but he is more right then not! Not his fault people don't pay attention when he is speaking.
     
  36. NKD

    NKD [H]ardness Supreme

    Messages:
    7,415
    Joined:
    Aug 26, 2007
    HBM2 shouldn't be too much of an issue right now. It was with original vega due to shortage initially. But don't see them having issues with HBM2 at all. Been out for too long now.
     
  37. viivo

    viivo Gawd

    Messages:
    946
    Joined:
    Sep 7, 2005
    Otherwise known as a scumbag price gougers wet dream. Actual consumers may be able to buy one at MSRP sometime around 2023. Of course that's if it's even true, which it isn't.
     
  38. bobzdar

    bobzdar [H]ard|Gawd

    Messages:
    1,576
    Joined:
    Jun 6, 2003
    If vega II performs between a 2080 and 2080ti and they price it at $599, undercutting the 2080 by $100-$200, I don't see how they wouldn't make money. Hbm2 isn't that expensive or they'd have lost money on every v56 and v64 sold - and we know those cards were financially successful for them in the mining boom at msrp - it was resellers that marked them way up. The v64 was a $499 card, so unless each 7nm chip costs them $100 more than the 14nm equivalent (not remotely likely) they'll be in the money at $599 and likely more profitable than vega64 to boot.

    I'm sorry, but from what I've seen so far, dlss and ray tracing aren't worth $50 more, let alone $100-$200.
     
    Algrim likes this.
  39. Factum

    Factum [H]ard|Gawd

    Messages:
    1,485
    Joined:
    Dec 24, 2014
    Everything you see in games are hacks to simulate ray traycing.
    Of you don’t want ray traycing...you have to give up on new games...the devs want RT...Microsoft has provided the DX12 DXR API...time to sell your gaming PC.
     
    IdiotInCharge likes this.
  40. NKD

    NKD [H]ardness Supreme

    Messages:
    7,415
    Joined:
    Aug 26, 2007
    Ray tracing is nice but not at the performance cost. Even that can be ignored at times but not at the price Nvidia is charging. Problem is Nvidia knows they can charge what they want and fanboys will pay for bragging rights. Its just the way it is. Plus its a huge die and they are charging a pretty penny for it. It will just need time. I think AMD will probably do something similar but only when costs are less.