AMD: Navi Speculation, Rumours and Discussion [2019]

Discussion in 'Architecture and Products' started by Kaotik, Jan 2, 2019.

  1. DavidGraham

    Veteran

    Joined:
    Dec 22, 2009
    Messages:
    2,773
    Likes Received:
    2,560
    It's legit. It appears Navi indeed sacrificed compute to gain more pixel pushing power, just like Digital Foundry predicted/anticipated. A Vega 64 is 12.5 TFLOPS, yet an RX 5700 is 8.5 TFLOPS at typical gaming clocks, and it's faster than a Vega 64.
     
    w0lfram, Pixel, Adrian and 6 others like this.
  2. Bondrewd

    Regular Newcomer

    Joined:
    Sep 16, 2017
    Messages:
    520
    Likes Received:
    239
    Less peak FP doesn't mean it's not doing more real FLOPS on average.
     
    naenrda likes this.
  3. Nebuchadnezzar

    Legend

    Joined:
    Feb 10, 2002
    Messages:
    974
    Likes Received:
    141
    Location:
    Luxembourg
    So is this Navi 10 or 12?
     
  4. Bondrewd

    Regular Newcomer

    Joined:
    Sep 16, 2017
    Messages:
    520
    Likes Received:
    239
    10.
     
  5. Pressure

    Veteran Regular

    Joined:
    Mar 30, 2004
    Messages:
    1,337
    Likes Received:
    271
    It should also be easier to keep 40 CUs occupied.
     
  6. anexanhume

    Veteran Regular

    Joined:
    Dec 5, 2011
    Messages:
    1,520
    Likes Received:
    692
    At 40CU in 255mm^2, that’s roughly the same CU density as Polaris - 36 CU in 232 mm^2. Radeon 7 saw a 33% shrink despite doubling memory bus, so this sounds like a big architecture overhaul.
     
    pharma and DavidGraham like this.
  7. Love_In_Rio

    Veteran

    Joined:
    Apr 21, 2004
    Messages:
    1,452
    Likes Received:
    110
    That doesnt fit at all with the 1,25 IPC gain per clock. You need 10 Navi Tflops to reach 12,5 Vega Tflops performance. In fact it should be a 25 % slower than a Vega 64.
     
  8. Bondrewd

    Regular Newcomer

    Joined:
    Sep 16, 2017
    Messages:
    520
    Likes Received:
    239
    Sandbags.
    Cute little sandbags.
    That was tested at like 4k or something.
     
  9. Love_In_Rio

    Veteran

    Joined:
    Apr 21, 2004
    Messages:
    1,452
    Likes Received:
    110
    And the AMD marketing team misses it?. Nah, it will be 1,25 in the best case.
     
  10. DavidGraham

    Veteran

    Joined:
    Dec 22, 2009
    Messages:
    2,773
    Likes Received:
    2,560
    I would advice to pay little attention to what AMD claims as gains, they will maximize them as much as they could under extreme ideal conditions.
     
    xpea, Pixel and Bondrewd like this.
  11. anexanhume

    Veteran Regular

    Joined:
    Dec 5, 2011
    Messages:
    1,520
    Likes Received:
    692
    1.25x was across a suite of 30 games at 4K. If you call an average of 30 games best case, then I guess you're correct.

    https://www.amd.com/en/press-releas...ion-leadership-products-computex-2019-keynote

    Also, 1900MHz/1546MHz *40 CU/64CU * 12.5 TF = 9.6 TF. IPC is not intrinsically tied to teraflops by a scalar multiplier, as it's ultimately game dependent (hence averaging 30 games gets you 1.25x). Some games it will be higher.

    Given RDNA does seem to be a big architecture change, I would guess the 1.25X will only grow as drivers get optimized.
     
    #691 anexanhume, Jun 9, 2019
    Last edited: Jun 9, 2019
    mahtel, Lightman, del42sa and 3 others like this.
  12. Cat Merc

    Newcomer

    Joined:
    May 14, 2017
    Messages:
    124
    Likes Received:
    108
    You claim an 8.5TFLOP Navi > 12.5TFLOP Vega. That would mean 1.25x is anything but ideal.
     
    naenrda likes this.
  13. DavidGraham

    Veteran

    Joined:
    Dec 22, 2009
    Messages:
    2,773
    Likes Received:
    2,560
    This isn't about the FLOPS, it's about performance (IPC) per clock. RX 5700 runs at 1750MHz typical gaming clock, which enables it to achieve performance comparable to the 2070, which is 10% faster than Vega 64.
     
  14. anexanhume

    Veteran Regular

    Joined:
    Dec 5, 2011
    Messages:
    1,520
    Likes Received:
    692
    Are you confirming the Strange Brigade demo was at 1750MHz?
     
  15. Love_In_Rio

    Veteran

    Joined:
    Apr 21, 2004
    Messages:
    1,452
    Likes Received:
    110
    So, why AMD ,that will give us the best case scenario, says the performance per clock is 1,25 higher?. To get what you say they should say 1,50.
     
  16. anexanhume

    Veteran Regular

    Joined:
    Dec 5, 2011
    Messages:
    1,520
    Likes Received:
    692
  17. Laniakea

    Newcomer

    Joined:
    Apr 16, 2019
    Messages:
    65
    Likes Received:
    82
    Thanks for posting. Interesting information. Nvidias GTX 1080 FE with it's 9.1 TFLOPs (usual boost is 1866 MHz so actually ~9.6 TFLOPs) beats Vega 64 in games. That would mean AMDs TFLOPs in games are better than Nvidias now, which would be nice for a change.

    EDIT: I remember that there was a leak from gfxbench (or something) that suggested turned down compute capabilities but much increased game performance for Navi. Does someone remember it? I can't find the bench result but I believe it was with 20 or 24 CUs while performaning around RX 580 levels.
     
    #697 Laniakea, Jun 9, 2019
    Last edited: Jun 9, 2019
  18. Bondrewd

    Regular Newcomer

    Joined:
    Sep 16, 2017
    Messages:
    520
    Likes Received:
    239
    Less that and outright just lower peak FP.
     
    Cat Merc and Laniakea like this.
  19. Love_In_Rio

    Veteran

    Joined:
    Apr 21, 2004
    Messages:
    1,452
    Likes Received:
    110
Loading...

Share This Page

  • About Us

    Beyond3D has been around for over a decade and prides itself on being the best place on the web for in-depth, technically-driven discussion and analysis of 3D graphics hardware. If you love pixels and transistors, you've come to the right place!

    Beyond3D is proudly published by GPU Tools Ltd.
Loading...