Page 3 of 9 FirstFirst 12345 ... LastLast
Results 21 to 30 of 90

Thread: AMD To Put An ARM Core On Future Fusion APUs

  1. #21
    Join Date
    Oct 2007
    Location
    Toronto-ish
    Posts
    7,544

    Default

    Yes, it is possible to implement OpenCL on HD4xxx although the performance is not good in a lot of common cases (for the reasons you mention). No we are not including that in the initial open source implementation that Tom is working on although we certainly wouldn't object to someone adding it (it is open source in a public repository after all).

    Feel free to call that "bullshit" if you must although many people would call it "focusing on the hardware where we can deliver a broadly useful solution".

    Are you sure the other poster you are referencing is running the LLVM shader compiler ? What was his performance like with the default compiler ? I suspect there is some other difference than the shader compiler.

    Maybe we could continue these discussions to more relevent threads ? It would be really convenient if we could move the HON performance discussion back to the HON thread.
    Last edited by bridgman; 06-13-2012 at 07:38 PM.

  2. #22
    Join Date
    Nov 2008
    Location
    Germany
    Posts
    5,411

    Default

    Quote Originally Posted by bridgman View Post
    Q, please take a deep breath.

    Yes, it is possible to implement OpenCL on HD4xxx although the performance is not good in a lot of common cases (for the reasons you mention).
    Bitcoin as a benchmark prove this is wrong.

    my HD4770 do "50 MHash/s" source: https://bitcointalk.org/index.php?topic=18355.0

    my PhenomII X4 only do "12MHash/s" @3,4ghz source: https://de.bitcoin.it/wiki/Mining_hardware#AMD

    this means my hd4770 is ~5 times faster than my CPU !

    and you call this: "performance is not good" LOL... what a "lie"

    sure i shold use my CPU 5 times slower instead because "bridgman" told me so.

    Quote Originally Posted by bridgman View Post
    No we are not including that in the initial open source implementation that Tom is working on although we certainly wouldn't object to someone adding it (it is open source in a public repository after all).
    yes yes i already know that but the reasons are based on a "lie" : "performance is not good"

    Quote Originally Posted by bridgman View Post
    Feel free to call that "bullshit" if you must although many people would call it "focusing on the hardware where we can deliver a broadly useful solution".
    its bullshit because using the CPU in my case "phenomII x4" is 5 times slower!

    "50 MHash/s" instead of "12MHash/s" is a useful solution!

    but feel free to "lie" otherwise.



    Quote Originally Posted by bridgman View Post
    Are you sure the other poster you are referencing is running the LLVM shader compiler ? What was his performance like with the default compiler ? I suspect there is some other difference than the shader compiler.
    i'm sure there is room for improvements. maybe the hd4000 performance grows in the future.
    and you can ask him what is is using.

  3. #23
    Join Date
    Aug 2007
    Posts
    6,646

    Default

    Well about the real topic:

    will that arm core be accessable in a direct way? if i remember it correctly there was a system that had a x86 and arm core and it was possible to use only the arm one to boot a tiny linux system. will that be possible here as well?

  4. #24
    Join Date
    Oct 2007
    Location
    Toronto-ish
    Posts
    7,544

    Default

    Quote Originally Posted by Qaridarium View Post
    Bitcoin as a benchmark prove this is wrong.

    my HD4770 do "50 MHash/s" source: https://bitcointalk.org/index.php?topic=18355.0

    my PhenomII X4 only do "12MHash/s" @3,4ghz source: https://de.bitcoin.it/wiki/Mining_hardware#AMD

    this means my hd4770 is ~5 times faster than my CPU !

    and you call this: "performance is not good" LOL... what a "lie"
    Sigh...

    OK, logic 101. My statement was "performance was not good in a lot of common cases". Your identification of a case where performance *is* good does not disprove my statement and certainly does not make it a lie.

    Quote Originally Posted by Qaridarium View Post
    i'm sure there is room for improvements. maybe the hd4000 performance grows in the future.
    and you can ask him what is is using.
    You are making the claim that the shader compiler is responsible for the performance difference (which contradicts all our experiences so far) then basing conclusions about our evil-ness on that claim, so perhaps you could ask him ? You only base your arguments on "facts", correct ?
    Last edited by bridgman; 06-13-2012 at 07:54 PM.

  5. #25
    Join Date
    Nov 2008
    Location
    Germany
    Posts
    5,411

    Default

    Quote Originally Posted by bridgman View Post
    Sigh...

    OK, logic 101. My statement was "performance was not good in a lot of common cases". Your identification of a case where performance *is* good does not disprove my statement and certainly does not make it a lie.
    no you are wrong because i show up facts REAL benchmarks and you just claim stuff without sources.
    show me your real benchmarks to prove your "Claims"
    bitcoin is not a calculation of special case other applications do the same calculations.
    in my point of view you are disproved until you show me benchmarks with other results.
    but i think you "can't" because my card is so "Powerful"
    the "Radeon HD 4870 X2" for example is so powerfull this card outperform my CPU in factor 20

    but you will stil claim that its not usefull to have a acceleration of factor "20"


    Quote Originally Posted by bridgman View Post
    re: shader compiler -- you are making the claim that the shader compiler is responsible for the performance difference (which contradicts all our experiences so far) then basing conclusions about our evil-ness on that claim, so perhaps you could ask him ? You only base your arguments on "facts", correct ?
    source: http://phoronix.com/forums/showthrea...n-driver/page2

    Pontostroy:"pentium g620
    opensuse + my repo with git mesa, kernel, xf86-drivers, etc.
    Shader caching and llvm shader compiler are enabled."

  6. #26
    Join Date
    Aug 2007
    Posts
    6,646

    Default

    you should not forget that even such cheap intel pentium cpus have got a faster single core speed than highend amd ones... usually games only use 1-2 core(s).

  7. #27
    Join Date
    Oct 2007
    Location
    Toronto-ish
    Posts
    7,544

    Default

    Quote Originally Posted by Qaridarium View Post
    but you will stil claim that its not usefull to have a acceleration of factor "20"
    No, I claimed nothing of the sort. I said two things :

    - performance was poor in a lot of common cases (or something like that)

    - we were not including HD4000 OpenCL support in the initial work we were doing

    Nothing more.

    Quote Originally Posted by Qaridarium View Post
    Thanks. I asked him about whether he saw a performance diff from enabling the LLVM shader compiler.

  8. #28
    Join Date
    Jul 2011
    Posts
    59

    Talking bridgeman + Q

    Greatest couple in the history of open-source graphics drivers discussions ever, seriously!

  9. #29
    Join Date
    Nov 2008
    Location
    Germany
    Posts
    5,411

    Default

    Quote Originally Posted by bridgman View Post
    No, I claimed nothing of the sort. I said two things :

    - performance was poor in a lot of common cases (or something like that)
    prove this!... and i'm sure you "can't"
    you only don't get a acceleration on lowend cards like hd4350 or hd4550.
    you need at minimum a hd4650 to get a speed up compared to the cpu.
    and this is your real secret point you claim linux users only use lowend hardware and on lowend hardware the hd4000 generation can not "accelerate"
    anything above hd4650 is much faster than the "CPU"

    Quote Originally Posted by bridgman View Post
    - we were not including HD4000 OpenCL support in the initial work we were doing
    Nothing more.
    and again i know this long time before you write about this here in the forum.
    i'm not stupid.

    Quote Originally Posted by bridgman View Post
    Thanks. I asked him about whether he saw a performance diff from enabling the LLVM shader compiler.
    thank you... maybe we get useful informations.
    Last edited by Qaridarium; 06-13-2012 at 08:18 PM.

  10. #30
    Join Date
    Nov 2008
    Location
    Germany
    Posts
    5,411

    Default

    Quote Originally Posted by Linuxxx View Post
    Greatest couple in the history of open-source graphics drivers discussions ever, seriously!
    LOL :-) and sometimes useful informations are created out of the battles

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •