AMD Releases Polaris Based Radeon Pro Graphics Cards

Discussion in 'Mac Pro' started by xsmi123, Aug 8, 2016.

  1. xsmi123 macrumors regular

    xsmi123

    Joined:
    Jun 30, 2016
    Location:
    Sylvania, OH
    #1
    I didn't realize that people missed this announcement. I believe these will be the cards used in the Mac Pro up dates when we get them.





     
  2. keysofanxiety macrumors 604

    keysofanxiety

    Joined:
    Nov 23, 2011
    #2
    Wowowww this looks amazing.

    $10,000 for the development kit alone, though! It'll be released in 2017 at the earliest. So unfortunately it probably won't be coming to a Mac Pro near you – not in the next couple of years, anyway :(
     
  3. xsmi123 thread starter macrumors regular

    xsmi123

    Joined:
    Jun 30, 2016
    Location:
    Sylvania, OH
    #3
    Well, not yet...But the WX series should pretty soon I'm hoping.
     
  4. koyoot macrumors 601

    koyoot

    Joined:
    Jun 5, 2012
    #4
    As for SSG. I think to make it work you need coherent fabric connecting the SSD to GPU and CPU.

    I wonder if it is possible to connect both GPUs in Mac Pro with coherent fabric and SSD in the same way, to create one huge computing unit.
     
  5. AidenShaw macrumors P6

    AidenShaw

    Joined:
    Feb 8, 2003
    Location:
    The Peninsula
    #5
    It would almost certainly be doable.

    The questions are whether it can be done within the space/power/thermal constraints of the cylinder, and whether it would be very expensive with few applications that could exploit it.
     
  6. koyoot macrumors 601

    koyoot

    Joined:
    Jun 5, 2012
    #6
    One thing. Low-Level API called Metal. There you are - fixed your problem.
     
  7. AidenShaw macrumors P6

    AidenShaw

    Joined:
    Feb 8, 2003
    Location:
    The Peninsula
    #7
    The API doesn't matter if the task cannot be parallelized.
     
  8. koyoot macrumors 601

    koyoot

    Joined:
    Jun 5, 2012
    #8
    For example?
     
  9. tuxon86 macrumors 65816

    Joined:
    May 22, 2012
    #9
    Any task that depends on the result of another task can't be run in parallel.
     
  10. AidenShaw, Aug 9, 2016
    Last edited: Aug 9, 2016

    AidenShaw macrumors P6

    AidenShaw

    Joined:
    Feb 8, 2003
    Location:
    The Peninsula
    #10
    It's the law.

    Which is what tuxon86 said, but explained using precise language like:

    amdahl.png
     
  11. xsmi123 thread starter macrumors regular

    xsmi123

    Joined:
    Jun 30, 2016
    Location:
    Sylvania, OH
    #11
    I wouldn't want to see it in a cylinder. I, for one, am hoping that No One was telling the truth about the test mule he saw with four PCI slots and cards the same color as the case.
     
  12. Ph.D. macrumors 6502

    Joined:
    Jul 8, 2014
    #12
    This card and the Mac Pro's cards intersect in one interesting way: One of the Mac Pro's graphics cards already holds up to 1 TB of SSD (and the other makes reservations for it). The Pro's cards have actually been working towards this sort of solution.
     
  13. xsmi123 thread starter macrumors regular

    xsmi123

    Joined:
    Jun 30, 2016
    Location:
    Sylvania, OH
    #13

    How? This is the first I've heard of this.
     
  14. koyoot macrumors 601

    koyoot

    Joined:
    Jun 5, 2012
    #14
    Yep, I forgot about that, thanks for reminding me.

    However, how big is the market in this specific scenario? Because I think that solution we are discussing is for 99% of users that want powerful Mac, and have need for it, and have money for it. Apple is going all in with parallel computing, so I do not think Amdahl's Law is accurately used in this particular case.
     
  15. lowendlinux Contributor

    lowendlinux

    Joined:
    Sep 24, 2014
    Location:
    North Country (way upstate NY)
    #15
    I didn't miss it, I just don't think it's relevant to anything in the Mac line up, but we'll see.
     
  16. Asgorath macrumors 65816

    Joined:
    Mar 30, 2012
    #16
    That's just where the nMP's hard drive is, as it's probably the only place they could find room for it. Not the same as the AMD tech.
     
  17. xsmi123, Aug 10, 2016
    Last edited: Aug 10, 2016

    xsmi123 thread starter macrumors regular

    xsmi123

    Joined:
    Jun 30, 2016
    Location:
    Sylvania, OH
    #17
    So you feel the nMP will continue... I don't. At least, I hope not.

    Or do you feel that the D series is not a true Firepro Card?
     
  18. Ph.D. macrumors 6502

    Joined:
    Jul 8, 2014
    #18
    No, not the same, of course. But it's there. And the other card has vestigial support for a second connector. I'm just pointing out the curious "coincidence."

    Personally, I don't think that it's entirely a mere accident of fate (e.g., couldn't fit otherwise). On the other hand, SSD on the graphics cards is probably a little premature for mainstream use.
     
  19. tuxon86 macrumors 65816

    Joined:
    May 22, 2012
    #19
    Parallel computing isn't a new thing. Thing is, only relative few task can effectively be parallelized. Graphics can and have been using it since they can be splitted in distinct unrelated rendering task. But for most other task that we use computer for the overhead of making sure task don't depend on the result of other task make it not worth the effort.
     
  20. Zarniwoop, Aug 11, 2016
    Last edited: Aug 11, 2016

    Zarniwoop macrumors 6502a

    Joined:
    Aug 12, 2009
    Location:
    West coast, Finland
    #20
    In future, more and more specialized tasks will be given to co-processors. Currently many tasks need a lot of single threaded non-parallel CPU time but same time the process needs access to unified memory to talk with different devices. Today that takes expensive CPU time to transport data from here to there.

    The co-processor future is already happening with Apple's' Ax SoC series; there are M9/DSP/ISP... etc. but they are difficult to benchmark, and comparing Apple's' Arm-chips to Intel variants is not fair. Most likely A9X could run Logic Pro X same speed as Intel desktop CPU, because DSP could take most of the CPU load.

    For Mac's, Intel is the road block preventing Macs to be fully HSA compliant and so co-processors can create a lot of overhear on CPU and limits its use. The main problem is Intel's closed memory architecture; it cannot share / doesn't want to share memory with co-processors. No support for universal heterogeneous Unified Memory Access architecture.

    So, it could be that Apple wants AMD to make a good x86 CPU for them to replace Intel. To fund this, they've used AMD only GPU's for years now. Zen or a custom chip could give Apple the key to be different in personal computing space. They first fully HSA compliant computer. Maybe this is why we havn't seen a lot of updates recently? Because Apple want's to make a jump?

    Second tech blocking Apples way to create HSA architecture (where they'd have keys for the system) was CUDA. So it had to go too... sure, latest CUDA supports HSA, but it is a closed system. OpenCL 2.0 supports shared virtual memory, which is a key component coming with HSA. So, no proper openCL 2.0 support before macOS and Mac hardware are HSA ready.

    Well, this is my theory? Got better?
     

Share This Page