Navi vs Turing architechture comparison.

Discussion in 'Video Cards' started by Stoly, Aug 8, 2019.

  1. sabrewolf732

    sabrewolf732 2[H]4U

    Messages:
    3,984
    Joined:
    Dec 6, 2004
    Believe I covered that in my post

    You're acting as if AMD hasn't released competitive cards, they have. The 290x was a beast when it launched, was competitive vs the 780ti and has aged far better. AMD can release competitive architectures. They just haven't since Pascal (fury WAS competitive with 980/to) and still have nothing abover upper mid-range today.
     
    PontiacGTX likes this.
  2. sabrewolf732

    sabrewolf732 2[H]4U

    Messages:
    3,984
    Joined:
    Dec 6, 2004
    The 2070 super 2080 2080 super and 2080ti are all faster than currently available rdna offerings, what are you talking about?
     
  3. Digital Viper-X-

    Digital Viper-X- [H]ardForum Junkie

    Messages:
    13,728
    Joined:
    Dec 9, 2000
    hmmm, not sure where your optimism is coming from, AMD is barely able to keep up @mid-range , while using a superior manufacturing process. If they were smacking nvidia around I would understand your statement, but 5700xt is just a hair shy of the 2070 super, there is still the 2080 super, 2080ti, and titan rtx
     
    IdiotInCharge and sabrewolf732 like this.
  4. vegeta535

    vegeta535 2[H]4U

    Messages:
    3,138
    Joined:
    Jul 19, 2013
    Wut? Who the cheer leader here? Sure 5700 cards are solid but what does Nvidia have to defend against? The 5700xt barely beats a 1080ti from 3 years ago. AMD has nothing beyond that. You really think Nvidia doesn't have anything I'm the works? Hell they probably have 7nm parts sitting in there shelf ready to go now. They are just being Nvidia and milking people. We not know what Nvidia have does not equate them having nothing at all. A company like Nvidia has a plan probably for the next 10 years. AMD also probably does.
     
    Last edited: Aug 26, 2019
  5. Gamer X

    Gamer X Limp Gawd

    Messages:
    347
    Joined:
    Jul 5, 2019

    I own an RTX2080... and I have a very big disposable income. Many (as in nearly everyone) do not see the 2080ti @ $1400 as a legit gaming card. It isn't, it is a hand-me-down chip from Nvidia's business sector. Turing is not meant for Gamers and as such, it is not as efficient as it can/should be. And because of it, Turing is not priced for gamers... because (AGAIN) it is a hand-me-down from a world, where those same GPU dies sell for double.

    Turing's price is no where near reasonable for Gamers. The hand-me-down hoax is over.... even AMD sent to the Mi50 packing, because it wasn't fair to Gamers to sell them hand-me-downs (ie: Radeon Vii) as legit Gamer cards.

    You have to understand this too... right..?


    Also, the 2080, 2080S, 2080ti,etc are all the same chip...
    A non-sequitur because those cards compete with each other... and if you take my remarks at their word, you'll understand you are feigning ignorance of what is coming from AMD, vrs what is coming from Nvidia... and when. Nvidia is a year away and AMD is month away, so why play coy with this.

    AMD is going to knock Nvidia and the Gamers over the head with a real low price for their 5800 Series. I suggest $499 ~ $699.




    And for Turing's architecture..? No matter what variant, or even the full TU-102.... still won't matter when 5800 Series lands, because TU-102 is not reasonably priced.... $2k GPU being cut-down to serve is a laughing stock. Navi10 is only 251mm^2, a lot smaller than Vega and yet outperformance Vega in games. RDNA's architecture is indeed impressive and efficient.

    If you make a RDNA chip the size of Radeon 7 @ $599 with a cutdown at $499..? It will be above Nvidia's top gaming card, for $300 less..!



    Finally, yes Navi10's RDNA is more powerful than Turing, which also means you can put the $399 5700xt up against the $599 1080ti... because Navi is more powerful than Pascal too...
     
  6. Digital Viper-X-

    Digital Viper-X- [H]ardForum Junkie

    Messages:
    13,728
    Joined:
    Dec 9, 2000
    How is it more powerful than turing? per wat? no... per $? maybe, but that's a result of the state of the market, not technology. Sooooo.... how is it more powerful? :)



    What you suggest and reality are not the same.
     
    IdiotInCharge and sabrewolf732 like this.
  7. andrewaggb

    andrewaggb Limp Gawd

    Messages:
    437
    Joined:
    Oct 6, 2004
    I'd still buy nvidia today. I can get a dual or triple fan cooler, CUDA support, and RTX. I dislike blower coolers, use CUDA, think RTX would be fun to play with even if it's a gimmick at the moment. Saving $100 isn't worth putting up with a blower cooler and losing CUDA to me.

    I appreciate most people don't care about CUDA and that we are very close to having third party RDNA cards with coolers available in numbers you can actually purchase - but that's where I'm at.

    I'm sure a big Navi would compete very well against a 2080ti, I'm also sure nvidia can move to 7nm and get big performance gains just like AMD did. We'll see when it happens.
     
  8. cyklondx

    cyklondx Limp Gawd

    Messages:
    254
    Joined:
    Mar 19, 2018
    Green is 5700 die 251mm^2
    Orange is TU102 NV 752mm&2


    upload_2019-8-26_11-20-7.png
    Navi has 10.3 billion transistors
    TU102 NV has 18.6 billion transistors


    Ayymd wins.
    5700 fits Tu102 8.9x, if size and transistor specs are correct on techspot's article.
    (if amd made a gpu, with the size of 2x 5700 die - it would already have 20 billion transistors more than radeon vii has. (radeon vii has 13bil trans)

    (4x would have over 40 billion transistors, while still smaller than nv at 502mm^2)
    upload_2019-8-26_11-39-51.png

    Obviously this isn't what we would see in reality. Don't get hyped or anything... If big navi comes around i think it would be only around ~350mm^2
     
    Last edited: Aug 26, 2019
  9. sabrewolf732

    sabrewolf732 2[H]4U

    Messages:
    3,984
    Joined:
    Dec 6, 2004
    Looks like you're espousing the benefits of tsmc 7nm process, not Navi.
     
    IdiotInCharge and Araxie like this.
  10. cyklondx

    cyklondx Limp Gawd

    Messages:
    254
    Joined:
    Mar 19, 2018
    yes, but its only summary of current performance vs if taken up the scale as there's no point to scale 1:1 for 89bill transistors. Thus how effective is their arch? Well it is slower and NV at certain things, and faster at other things. I think it hasn't eliminated issue with waiting for resources at all... we are constantly starved, and waiting.
     
  11. Dayaks

    Dayaks [H]ardness Supreme

    Messages:
    7,114
    Joined:
    Feb 22, 2012
    That’s not how math works. It’s mm^2 so 752 is about 3x 251, not 8.9x. So three 5700s fit into one TU102 by area.

    Also the 2080ti doesn’t utilize all the transistors, it’s a cut down chip (some are “off”). The Titan RTX does.

    And comparing die sizes of different processes doesn’t really make sense. You should compare cost per transistor.
     
  12. cyklondx

    cyklondx Limp Gawd

    Messages:
    254
    Joined:
    Mar 19, 2018
    // sarcastic,
    open paint
    Draw a rectangle 752x752, then start drawing inside that rectangle, smaller rectangles with size of 251x251.
    I'll wait.

    (and anyway if you use your maths, you need to remember you are dealing with 2d... so 3x3 = 9)

    The TU102 discussed is the full chip. (didn't discussed cut down Ti's)
    https://www.techpowerup.com/gpu-specs/nvidia-tu102.g813
     
    Last edited: Aug 26, 2019
  13. Stoly

    Stoly [H]ardness Supreme

    Messages:
    6,327
    Joined:
    Jul 26, 2005
    Your right.

    No, wait... isn't there something coming next year? what's it called.... Ampere? on 7nm process with an improved architechture? and 2nd gen RTX?
     
    IdiotInCharge likes this.
  14. sabrewolf732

    sabrewolf732 2[H]4U

    Messages:
    3,984
    Joined:
    Dec 6, 2004
    Edit: Misread your post
     
  15. cyklondx

    cyklondx Limp Gawd

    Messages:
    254
    Joined:
    Mar 19, 2018

    They are supposed to use even "better" Samsung euv lithography.
    Seeing it is believing.
     
  16. NKD

    NKD [H]ardness Supreme

    Messages:
    7,678
    Joined:
    Aug 26, 2007
    Way to assume the worst. RDNA if anything is proof AMD Is on right track. This is called biased opinion. What makes you think AMD made RDNA almost as fast as Radeon VII with 2560 shaders will go down hill? lol. Lisa isnt going to let that happen.

    I guess you probably said the same thing about zen? "Zen is here to stay , AMD will take Zen, and run it into mediocracy" but hey guess what? They didn't.

    AMD is in better position not worst.
     
    PontiacGTX, sabrewolf732 and cyklondx like this.
  17. cyklondx

    cyklondx Limp Gawd

    Messages:
    254
    Joined:
    Mar 19, 2018
    I think every arch is going to become mediocre as time passes, and people forget why it was great. Just like Fermi, we cannot grill anymore - it was the best grill i ever owned.
     
  18. Digital Viper-X-

    Digital Viper-X- [H]ardForum Junkie

    Messages:
    13,728
    Joined:
    Dec 9, 2000
    I think he's talking about future iterations, refinements etc.

    Eg the 7970 was a beast, GCN1.0, and slowly AMD lost their edge with it over and never gained it back.
     
  19. cyklondx

    cyklondx Limp Gawd

    Messages:
    254
    Joined:
    Mar 19, 2018
    290x was even greater beast. Just plagued by couple bad non-ref models.

    Asus made their directcu2 line sht; they cheapen it out.\
    msi gaming omg what a garbo... lack of paste... not to mention design garbo.
    gigabyte windforce wtf... really and it didn't cover half the mem modules... which resulted in them dying quickly.

    only good models were powercolor, sapphire, and msi lightning... while most ppl were stuck on those asus msi and gigbyte cooling setups.
    The 290x was so good it competed with 980.

    Fury is a lost one...
     
  20. Snowdog

    Snowdog [H]ardForum Junkie

    Messages:
    9,217
    Joined:
    Apr 22, 2006
    So you are doubling down, on being wrong...

    Do you understand even grade school math? They are already reporting the Area (hence the use of squared units).

    The die is something 26 mm x 29 mm = 754 mm^2. Not 752 x 752.
     
  21. cyklondx

    cyklondx Limp Gawd

    Messages:
    254
    Joined:
    Mar 19, 2018
    crap you are right. i'm a dumb ass.
     
  22. Snowdog

    Snowdog [H]ardForum Junkie

    Messages:
    9,217
    Joined:
    Apr 22, 2006
    The best bet for these comparisons right now is to look at transistor counts. It will get you closer, but that can be an issue since most direct comparisons end up using dies with disabled parts.

    The best to establish some sort of equilibrium is 5700XT vs RTX 2070. 10.3 Billion vs 10.8 Billion. Very similar performance, very similar transistor count.

    If you do that comparison for Vega 64 vs GTX 1080, you will see that Pascal vs Vega, AMD was way behind, now with Navi vs Turing they have caught up (ignoring RT HW for the moment), but did not leapfrog NVidia.

    This is great because we have a much more competitive landscape today.

    But neither AMD, nor NVidia are really stomping the other anymore.
     
    MangoSeed and andrewaggb like this.
  23. Dayaks

    Dayaks [H]ardness Supreme

    Messages:
    7,114
    Joined:
    Feb 22, 2012
    It was more of a rebuttal to Gamer X’s proproganda if you look at what if I was quoting. Not impossible given history though, which made it even more fun.
     
    IdiotInCharge and NKD like this.
  24. NKD

    NKD [H]ardness Supreme

    Messages:
    7,678
    Joined:
    Aug 26, 2007
    I doubt lisa is that dumb she is going to repeat the same mistake with RDNA. GCN was long overdue, that is given. Its also well known how they had to spin resources to the CPU side a while back or they were as good as dead in water since that was their bread and butter. Now they are kicking that in gear and executing, and first iteration of RDNA out. I see no reason for them to slack on GPU side when they are in a better shape overall due to CPU side.
     
    otg and Maddness like this.
  25. Ready4Dis

    Ready4Dis Limp Gawd

    Messages:
    501
    Joined:
    Nov 4, 2015
    I'm still rocking my fury... OCed and undervolted... Still not much reason to upgrade. :).
     
    sabrewolf732 likes this.
  26. Gamer X

    Gamer X Limp Gawd

    Messages:
    347
    Joined:
    Jul 5, 2019
    I've already explained how "your math" and reality works. I've already explained all of this, and provided illustrations and even suggested that people learn the difference in scaling surface area by a magnitude -vs- multiplying numbers. (As someone just recently re-iterated graphically.)

    I've also mentioned how AMD is using their exclusive multi-chip contract to leverage EOS by making all AMD's GPU & GPU @ 7nm, thus reducing the costs of transitioning over to 7nm. And how Nvidia is staying at 12nm, because they can't afford (to eat) the 7nm costs, because they do not make Gaming specific chips and are have decided to wait an entire year, to move their entire Business to 7nm (EUV). (They don't offer as many products as AMD.)


    Ironically, weren't you the same person who didn't know what economy of scale was.... and mocked me for pages? Only to find out, that you (& many others) were actually clueless on how chips are priced, or what the end-users & gamer costs are.



    I am not bias I am not spreading propganda, but I can see cheerleader's pom-poms all over the hallways of these forums.

    I support RDNA because it is smaller than Vega architecture and does more. Aren't you people confused how little ole navi10 @ 251mm^2 beats out Vega20 @ 335mm^2 in many games..? What else is so weird/tell-tale since being here, is that nobody here wants to see a 350mm^2 Navi chip....

    Why?
     
  27. Dayaks

    Dayaks [H]ardness Supreme

    Messages:
    7,114
    Joined:
    Feb 22, 2012
    I didn’t mock you. I flat out told you you were wrong which is still true. Yield on wafers =/= economy of scale which was your focus previously.

    And you’re still ignoring Vega20 is a partially disabled die.
     
    Last edited: Sep 10, 2019
  28. Ready4Dis

    Ready4Dis Limp Gawd

    Messages:
    501
    Joined:
    Nov 4, 2015
    Lol, you guys still responding... Just stop feeding ;).
     
    MangoSeed, Araxie and Dayaks like this.
  29. Aireoth

    Aireoth 2[H]4U

    Messages:
    2,752
    Joined:
    Oct 12, 2005
    AMD catches up with nVidia’s mid range and last gens top end, like usual, AMD fans celebrating like it’s the second coming. Business as usual.

    Now the good thing i’m seeing is it looks like architecture wise AMD has made some great strides. Maybe next time they can compete in all bracket?

    Or the plot twist is right and Intel pulls a Godzilla and stomps both of em?
     
    Araxie likes this.
  30. Rvenger

    Rvenger [H]ard|Gawd

    Messages:
    1,741
    Joined:
    Sep 12, 2012

    I don't get how people get their panties up in a bunch when AMD has a decent showing especially for Navi. At least it gives hope that competition may be back on the highend.


    I said it twice and I will say it again. Navi and future AMD GPU architectures will favor the consoles, its not designed around beating Nvidia. Go watch the David Canter Navi deep dive.
     
    Aireoth likes this.
  31. Aireoth

    Aireoth 2[H]4U

    Messages:
    2,752
    Joined:
    Oct 12, 2005
    It’s a result of how discourse on the internet works, subtle positions are difficult and people get reduced to the extremes. It’s been particularly damaging to political discourse and issues like climate change.

    GPU fandom is a prime example of that.
     
    Dayaks likes this.
  32. Rvenger

    Rvenger [H]ard|Gawd

    Messages:
    1,741
    Joined:
    Sep 12, 2012

    Read my edit. It will never beat out Nvidia on every metric as long as AMD is designing the GPU for consoles.
     
  33. PontiacGTX

    PontiacGTX Gawd

    Messages:
    696
    Joined:
    Aug 9, 2013
    I mean they can their problem it is they keep releasing the cards 1-2year(s) later than Nvidia also they keep releasing midrange first, and lagging a long time to release the high end after that midrange
     
  34. NKD

    NKD [H]ardness Supreme

    Messages:
    7,678
    Joined:
    Aug 26, 2007
    Hey if one can predict the future based off past all power to them. One could say the same thing about fermi, but that didn't happen. I trust Lisa to not make the same mistakes as old AMD. She has earned that lol.
     
  35. Dayaks

    Dayaks [H]ardness Supreme

    Messages:
    7,114
    Joined:
    Feb 22, 2012
    Well, we know at least the next year will be Pascal-ish performance which is a big leap up for them. A 2080ti performance card isn’t out of the realm of possibility if they can manage thermals / power.
     
    Maddness likes this.