Page 173 of 357 FirstFirst ... 73123163171172173174175183223273 ... LastLast
Results 1,721 to 1,730 of 3568

Thread: Vega 10, Vega 11. What we know

  1. #1721
    Senior Member
    Join Date
    Feb 2010
    Posts
    3,443
    To me, Raja showed quite mixed emotions during his speech. At a moment or two, he looked nervous and angry.

    Quote Originally Posted by xsimbyx View Post
    Don't get your hopes up. I really doubt we'll find out anything about pricing or performance. Which is what most people actually wanna know.
    They showed some in-game 4K framerates. Around 60-70-80 FPS...
    Can anyone say something in detail about the demos during the event? Any comparisons to the competitor's product stack?

  2. #1722
    Senior Member
    Join Date
    Dec 2012
    Posts
    7,390
    Quote Originally Posted by GeorgiD View Post
    To me, Raja showed quite mixed emotions during his speech. At a moment or two, he looked nervous and angry.
    I only saw the start (work called :/) but, his mannerisms always seem nervous to me. If it's online I certainly plan to watch it.
    -Q

  3. #1723
    Senior Member
    Join Date
    Jan 2010
    Posts
    3,204
    Raja is engineer in trade, he is not supposed to be on the stage, but AMD is appealing to audience with Raja passionate student like speech.
    Regarding comparisons with competition. 1080ti does 68fps average on sniper elite with 7700k clocked sky-high. AMD I assume used their own Ryzen to demo sniper elite.
    I'm not worried about Vegas performance anymore I'm more worried about these few comments stating about gaming Vega delay.

  4. #1724
    Senior Member
    Join Date
    Jan 2010
    Posts
    3,468
    And to think last year there was a hope that Vega would appear in October or November, and now it looks like it's going to be 3/4 of a year later.

  5. #1725
    Senior Member
    Join Date
    Sep 2010
    Posts
    146
    I'm guessing HBM2 supply is really biting them in the a*. Both Hynix and Samsung dropped the ball here.

  6. #1726
    Senior Member
    Join Date
    Sep 2010
    Posts
    146
    On another note, anyone else thinks Navi based GPU will look exactly like EPYC? MCM with 4 smaller GPUs connected with IF.

  7. #1727
    Senior Member
    Join Date
    Dec 2010
    Location
    Switzerland Geneva Cern
    Posts
    2,902
    Amd themselves confirmed that vega has 1600+MHz boost and probably 1400+MHz base clocks. This is quite interesting as it seems they achieved their bold claims about increasing their gpu clocks to almost match nvidia pascal.

    Also the design overall looks more balanced than fury X. We have a 50% performance uplift from fury X as the minimum extra performance provided so things look good for vega. It might do even better than i previously hoped for (which was a tad slower than the first titan pascal).
    Current rig: i5 3570 @3.8GHz, Asus 7750, Dell 24'' 2560X1440, Corsair Vengence 16GB Ram DDR3 @1600MHz, Chieftec 550W PSU, 480GB OCZ SSD + 500GB Seagate + 1TB WD hd, Windows 10 64-bit

  8. #1728
    Senior Member
    Join Date
    Dec 2010
    Location
    Switzerland Geneva Cern
    Posts
    2,902
    Quote Originally Posted by xsimbyx View Post
    On another note, anyone else thinks Navi based GPU will look exactly like EPYC? MCM with 4 smaller GPUs connected with IF.
    I think it will probably be 2 gpus connected and not 4. Nvidia will have a huge problem competing with this product if they have monolithic dies as the only option. Gpus are more parallel designs and they wont suffer as much as cpus when being on an MCM configuration.

    Imagine for example 2 300 - 400mm2 navi gpus working together with average scaling ~70% across all gpu applications (maybe even on some compute applications??). Nvidia will have to use a +500mm2 chip to compete which will certainly be more expensive than the 2 navi gpus combined.
    Current rig: i5 3570 @3.8GHz, Asus 7750, Dell 24'' 2560X1440, Corsair Vengence 16GB Ram DDR3 @1600MHz, Chieftec 550W PSU, 480GB OCZ SSD + 500GB Seagate + 1TB WD hd, Windows 10 64-bit

  9. #1729
    Senior Member
    Join Date
    Jan 2010
    Posts
    3,468
    Quote Originally Posted by xsimbyx View Post
    On another note, anyone else thinks Navi based GPU will look exactly like EPYC? MCM with 4 smaller GPUs connected with IF.
    It'll likely be on an interposer, and I'm guessing three types of silicon:

    One of I/O module (does all the non-scalable GPU stuff, video outputs, video decode/encode, etc) - 14nm
    N of Compute modules (2048 shaders, One HBM2 controller) - 7nm
    N of HBM2 stacks

    SKUs can be made with 1, 2, 3 or maybe even 4 Compute modules. Which would mean (let's assume 256 disabled on cut-down SKUs), 1792, 2048, 3586, 4096, 5376, 6128, 7168, 8196 shader SKUs.
    That'll be the RX 750, RX 760, RX 770, RX 780 and various RX Navi branded SKUs.

    Obviously the big leap here would be sacrificing non-interposer, non-HBM2 options, but cutting down on the 7nm designs required, and the die sizes required. Less up front costs on a process where a design costs $200m+, but adding per-item costs for a small interposer and HBM2 stack. I guess there would still eventually be a monolithic low-end SKU for products under $150.

  10. #1730
    Senior Member
    Join Date
    May 2012
    Posts
    4,652
    Quote Originally Posted by xsimbyx View Post
    I'm guessing HBM2 supply is really biting them in the a*. Both Hynix and Samsung dropped the ball here.
    Not really, afaia both said that HBM2 wouldn't go into volume ramp until Q1 this year. So why people expected a volume consumer part based on HBM2 before Q2 this year is a bit of a mystery ?
    To find the right answers you must ask the right questions.

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •  
WordPress Appliance - Powered by TurnKey Linux