nVidia awesomeness waits for 2014

Discussion in 'MacBook Pro' started by whitedragon101, Jan 3, 2013.

  1. whitedragon101 macrumors 6502a

    Joined:
    Sep 11, 2008
    #1
    Just felt like reposting the pending awesomeness that is Maxwell. Basically if you don't already know nVidia will release Maxwell GPUs next year (it was meant to be this year) which is a massive leap in performance. Not just incremental but an epic increase . See graphic :

    http://www1.pcmag.com/media/images/329235-nvidia-gpu-roadmap.jpg?thumb=y

    15 DP GFLOPs per watt people !!!

    I used to work in supercomputing and may be disproportionally excited about it. But put it this way :

    Arma 3 on a laptop !!!!!
     
  2. AirThis macrumors 6502a

    Joined:
    Mar 6, 2012
    #2
    I wonder if everyone will jump on the CUDA bandwagon... This is what irks me about Safari on the rMBP. You've already got a chip that's massively powerful in the machine and which supports CUDA. By using that technology, you should be able to eradicate any browser lag. The computational power in CUDA can be used for practically anything (biocomputing, encryption, image rendering, etc.). Certain programs used in Biocomputing already perform calculations up to 20 times faster with CUDA enabled. So why not use this same technology to enhance the daily browser experience? The answer seems obvious to me: Apple is pro-OpenCL, and Nvidia is obviously not. So I'm not sure we have the right chips in our machines. I wouldn't be surprised to see Apple ditch Nvidia at some point in the future.
     
  3. Erasmus macrumors 68030

    Erasmus

    Joined:
    Jun 22, 2006
    Location:
    Hiding from Omnius in Australia
    #3
    Irrelevant.

    All modern GPUs support OpenCL, even NVIDIA ones. The fact that Safari does not run using CUDA or OpenCL has nothing to do with NVIDIA. There are no processor intensive operations performed by Safari that are massively parallel, and therefore would not benefit from running on a GPU. In fact, it would run considerably slower.
     
  4. AirThis macrumors 6502a

    Joined:
    Mar 6, 2012
    #4
    How good is the OpenCL support for Nvidia chips on OSX? Do you really think it's optimized?

    As for Safari, there are plenty of images to load on a web page. I have a very hard time believing it would be slower.
     
  5. Squilly macrumors 68020

    Squilly

    Joined:
    Nov 17, 2012
    Location:
    PA
    #5
    What about Haswell? Is this thats predecessor?
     
  6. duervo macrumors 68000

    duervo

    Joined:
    Feb 5, 2011
    #6
    Apple (or any manufacturer, for that matter) isn't going to go all in with CUDA very quickly, if at all. If it does happen, it will be slow going, as it will tie them to nVidia for the time that they do make use of it. They won't make a move like that lightly.

    Being able to switch between AMD/ATi and nVidia between product cycles gives them a lot more leverage than they'd have if they went solely with one over the other, which is the road they'd be going down if they decide to make heavy use of CUDA. I doubt that they'd want to place that much power in the hands of any of their suppliers.
     
  7. cirus macrumors 6502a

    Joined:
    Mar 15, 2011
    #7
    I wouldn't necessarily trust that graph. The dates are wrong (about a year off). Furthermore, so far, kepler gets worse (or very similar) dp gflops per watt than Fermi (absolute dp compute is weaker but power consumption is a bit lower too). Single precision compute has been improved but GCN gets about 4-5 times more dp than kepler (talking about 7970/7950 vs 670/680).
    The only good dp kepler cards are the k20 and k20x and these are not available to the public or most companies.
     
  8. simsaladimbamba

    Joined:
    Nov 28, 2010
    Location:
    located
    #8
    Haswell is Intel's microarchitecture (CPUs and chipsets with integrated graphics processor (IGP)) for 2013 for mobile and desktop computers. It is the successor of Ivy Bridge (2012), which is the successor of Sandy Bridge (2011).

    Or do you mean something else?
     
  9. bill-p macrumors 65816

    Joined:
    Jul 23, 2011
    #9
    Safari lag has already been fixed. Install Webkit nightly.

    As for OpenCL and CUDA, most of what they accelerate involve linear math that benefits a lot from threading (splitting the algorithm into different parts and processing each part in parallel). Typically, that's a lot faster (about 2-10x faster) than regular CPUs. But when it comes to branching (if, else, logical statements, that sort of thing), they have to branch, or warp within the GPU, and that usually results in dropped performance. Markedly so.

    In fact, in some cases, CPU might well be faster than GPU when it comes to branching (logical operations).

    Unfortunately, Safari and other such browsers rely heavily on logical operations. Especially when it comes to interpreting or recompiling Javascript. So they won't benefit at all from CUDA or OpenCL.

    Unless, of course, you're talking about a website that's purely made up of a ton of images... but even then, those are rare, and in most cases, branching operations still affect performance a lot more while scrolling and navigating than shaving a few seconds off of loading a big image.
     
  10. Squilly macrumors 68020

    Squilly

    Joined:
    Nov 17, 2012
    Location:
    PA
    #10
    Then what is Maxwell?
     
  11. Ploki macrumors 68000

    Joined:
    Jan 21, 2008
    #11
    Nvidias architecture.

    CPU in rMBP is Ivy Bridge, GPU is Kepler.

    2013 rMBP will probably be Haswell (CPU) + Kepler GPU,
    and 2014 rMBP will be Broadwell CPU + Maxwell GPU.

    Nvidia and Intel both have their own architecture roadmap.
     
  12. nill1234 macrumors member

    Joined:
    Dec 22, 2012
    #12
  13. daneoni macrumors G4

    daneoni

    Joined:
    Mar 24, 2006
    #13
    Haswell is Intel's 2013 next gen CPUs whilst Broadwell (2014) is it's successor.

    Maxwell is Nvidia's 2014 next gen GPUs
     
  14. Squilly macrumors 68020

    Squilly

    Joined:
    Nov 17, 2012
    Location:
    PA
    #14
    Is rMBP supposed to use Kepler in their gpu or Maxwell (in 2013)?
    Guessing broadwell and maxwell will be much better?

    ----------

    Don't they say that about every new GPU or CPU architectural design?

    ----------

    Look above.
     
  15. simsaladimbamba

    Joined:
    Nov 28, 2010
    Location:
    located
    #15
    Since Maxwell does not arrive before 2014, Kepler (already released) is the most likely candidate.

    As with almost every new architecture in computer business, a new architecture will be faster than the older on.

    What do they say? That they have a roadmap?
     
  16. Squilly macrumors 68020

    Squilly

    Joined:
    Nov 17, 2012
    Location:
    PA
    #16
    That it'll be better/a large difference.
     
  17. simsaladimbamba

    Joined:
    Nov 28, 2010
    Location:
    located
    #17
    It is business to do so, some kind of marketing voodoo, and in many cases, a newer architecture, unless botched, brings speed gains. Many people will of course not take advantage of that, just like with current technology and past technology, but for the few that do, and the even fewer that care about this, those speed gains will be noticeable.
     
  18. whitedragon101 thread starter macrumors 6502a

    Joined:
    Sep 11, 2008
    #18
    When people realise a rMBP can be a gaming machine (if they deliver) people will care.
     
  19. leman macrumors 604

    Joined:
    Oct 14, 2008
    #19
    If something supports CUDA, it will also support OpenCL. Its simply a different API for essentially the same thing.

    And besides, good luck utilising CUDA/OpenCL whatever for website layout. This is a completely different thing from biocomputing or any other embarrassingly parallel problem.

    ----------

    It already is. It plays every modern game with reasonable settings and reasonable performance.
     
  20. jasonvp macrumors 6502a

    jasonvp

    Joined:
    Jun 29, 2007
    Location:
    Northern VA
    #20
    Loading images isn't an arithmetic process that can be parallelized. I think you're sort of missing the point of CUDA and OpenCL. Safari may not be any slower with OpenCL/CUDA support added, but it certainly won't be any faster, either. I'm pretty sure Apple knows that and won't bother with the trouble of adding either of them.

    jas
     
  21. whitedragon101 thread starter macrumors 6502a

    Joined:
    Sep 11, 2008
    #21
    I agree you can play modern games with 'reasonable' settings and get 'reasonable' performance. I am not criticising the current gen at all. But it isn't a go to choice for gamers who want a machine specifically for gaming as for that you need to be able to turn up all the dials and get 60fps (i.e alienware, asus monster laptops etc). With 3x the raw GPU compute performance per watt(again if they deliver) Maxwell could make the rMBP a really decent gaming system.
     

Share This Page