AMD: R7xx Speculation

Discussion in 'Architecture and Products' started by Unknown Soldier, May 18, 2007.

Thread Status:
Not open for further replies.
  1. A.L.M.

    Newcomer

    Joined:
    Jun 2, 2008
    Messages:
    144
    Likes Received:
    0
    Location:
    Looking for a place to call home
    Ok. So the two gpu share framebuffer AFR related content through CF connection. With an internal port like the CF-X sideport, each gpu gains direct access to the other gpu's MC hub, maybe having the chance to directly write into the other gpu framebuffer or maybe even to share some resources of the other gpu core...
     
  2. MistaPi

    Regular

    Joined:
    Jun 12, 2002
    Messages:
    353
    Likes Received:
    4
    Location:
    Norway
    For 1-2 years down the line 1GB vram might be quite usefull?
    Any word on a 4870 with 1GB vram?
     
  3. A.L.M.

    Newcomer

    Joined:
    Jun 2, 2008
    Messages:
    144
    Likes Received:
    0
    Location:
    Looking for a place to call home
    It can be, especially if you use 1080p res and up. I think that by the end of July the production of GDDR5 could have ramped up enough to permit a decent availability of 1GB products (given the fact that R700 should be launched at the end of August).
     
  4. Silent_Buddha

    Legend

    Joined:
    Mar 13, 2007
    Messages:
    15,932
    Likes Received:
    4,889
    Sort of like how G71 was praised for its efficiency in comparison to R580 even though R580 was generally quite a bit faster, although G71 could catch up in a few titles and surpass it in OGL titles? Sorta like Rv770 compared to GT200 (4870 vs GTX 280). Although to be fair the GTX 280 is more often faster by a fair bit than the 4870. The flip side though is it's also significantly more inefficient than the 4870. When viewed in light of a R580-G71 comparison (for perf/mm^2).

    Let's see, what did Nvidia do about that? Hmmm, oh yeah the GX2. ;) And what's ATI going to do about the current situation? Oh yeah 4870x2 (or whatever R700 end up being called).

    The more things change...the more things stay the same.

    Either way the Rv770 does deserve a significant amount of praise for what it has accomplished. Not only in comparison to where it came from (R600) but also in light of the competition (GT200).

    Regards,
    Croaker
     
  5. wishiknew

    Regular

    Joined:
    May 19, 2004
    Messages:
    332
    Likes Received:
    6
    Wow. No brain drain on Ati's part.

    So are the new Ruby and scorpion demo out yet?
     
  6. Lukfi

    Regular

    Joined:
    Apr 27, 2008
    Messages:
    423
    Likes Received:
    0
    Location:
    Prague, Czech Republic
  7. sireric

    Regular

    Joined:
    Jul 26, 2002
    Messages:
    348
    Likes Received:
    22
    Location:
    Santa Clara, CA
    No, the L1/VC and L2 have their own network of busses (~384 GB/s) to communicate, which allows the L2 to broadcast their respective datas to the L1's. The L1s are tied to SIMD cores, while the L2's are tied to memory channel.

    The L2s and other heavy hitters (raster backends mainly) are tied to pairs of memory channels directly. A hub now also exists, but is only used by clients that have low bw. It's a mixture of 600/500 memory subsystems.
     
  8. Lux_

    Newcomer

    Joined:
    Sep 22, 2005
    Messages:
    206
    Likes Received:
    1
  9. 3dilettante

    Legend Alpha

    Joined:
    Sep 15, 2003
    Messages:
    8,118
    Likes Received:
    2,860
    Location:
    Well within 3d
    Is the high-level diagram used by techreport directly from AMD?
    That's the image that has 16 sections per SIMD, but if the little yellow dots are ALUs, there are 160 per SIMD.
    A minor blooper, I suppose.

    I did notice that, compared to other diagrams of R600, the tesselator plugs into more things, and shader export plugs into the vertex assembler.
     
  10. AlexV

    AlexV Heteroscedasticitate
    Moderator Veteran

    Joined:
    Mar 15, 2005
    Messages:
    2,528
    Likes Received:
    107
    Yeah, that's from AMD, and yes it's blooped, since the one with the SIMD array has the proper count...unless they have, gasp, 1600 ALUs, with 800 for redundancy:D.
     
  11. fellix

    fellix Hey, You!
    Veteran

    Joined:
    Dec 4, 2004
    Messages:
    3,484
    Likes Received:
    396
    Location:
    Varna, Bulgaria
    Hm, peering at the RV770 die-shot now, why I have the weird feeling that this CF sideport is just a second PCIe switch alongside the "host" one...!? :???:
     
  12. tuteja1986

    Newcomer

    Joined:
    Feb 23, 2007
    Messages:
    34
    Likes Received:
    0
    Location:
    Australia , Sydney
    Doesn't look its working :!

    Check the Beyond3d pics , it should work something like that.
    http://doomstar.awardspace.com/Edge_detect_AA_mode.html

    Anyways , Edge days are over.

    If you have a 38XX and want the best quality at least performance hit then use :
    8x Edge Detect + Quality Adaptive AA

    [​IMG]

    If you have a 48XX and want the best IQ at least performance hit then use:
    8x box + Quality Adaptive AA

    http://anandtech.com/video/showdoc.aspx?i=3341&p=10
    "The RV770 actually shows a bigger performance hit from enabling their tent filters than RV670. This is likely because the filters are run on shader hardware in both cases while RV770 has faster hardware resolve that can be used for normal AA resolve. If RV670 resolves "box" filtered AA on the shader as well this would explain the flatter performance in that case. Even more so than the image quality question, the fact that they perform lower really should be the nail in the coffin for AMD's tent filter garbage. "

    http://www.rage3d.com/reviews/video/atirv770/architecture/index.php?p=6

    "the RV770 handles 4AA samples per clock and is thus capable of outputting 16 pixels per clock even with 4X AA, whereas the R600 could only manage 2 samples per clock, which resulted in it outputting only 8 pixels per clock with AA enabled."

    "Many of these improvements are tied to the redesign of the Color Blender block (CB), which took away unnecessary functionality and allowed the aforementioned doubling of AA rates."

    So tent is out , MSAA is in.
     
    #4792 tuteja1986, Jun 25, 2008
    Last edited by a moderator: Jun 25, 2008
  13. mczak

    Veteran

    Joined:
    Oct 24, 2002
    Messages:
    3,012
    Likes Received:
    112
    If you look at the techreport figures, you can see the same:
    http://techreport.com/articles.x/14990/7

    There's not much reason to use the tent filters (at least those using less than 8 samples) since quality will be worse and performance lower than a base 8x mode. Edge detect 12x seems to be useless too on a quality/performance metric compared to 8xMSAA. Only if you want the best quality possible use 24x ED.
    However, it should be noted that MSAA scaling can be FAR different depending on the game (and possibly even within the same game depending on the area I guess).
    http://www.computerbase.de/artikel/...0_cf_hd_4870/5/#abschnitt_aa_und_afskalierung
     
  14. Geo

    Geo Mostly Harmless
    Legend

    Joined:
    Apr 22, 2002
    Messages:
    9,116
    Likes Received:
    213
    Location:
    Uffda-land
    Folks, the day has finally come when this thread has outlived its usefullness. Please take general conversations about 48xx to the reviews threads in 3D Hardware. If specific instances of theoretical discussion of some element of RV770 or the cards based on it still seem warranted, feel free to open new thread(s) specific to those elements.
     
Loading...
Thread Status:
Not open for further replies.

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...