OpenGL 2.1 may double performance under Leopard

Posted:
in macOS edited January 2014
A new implementation of the OpenGL industry-standard interface for developing portable, interactive 2D and 3D applications could offer as much as a twofold increase in performance when running under Apple Computer's forthcoming Leopard operating system.



OpenGL improvements



According to a post on Apple's developer website, Mac OS X 10.5 Leopard will support the latest OpenGL 2.1 specification which adds pixel buffer objects, color managed texture images in the sRGB color space, and improvements in the shader programming API.



"Leopard also provides a dramatic increase in OpenGL performance by offloading CPU-based processing onto another thread which can then run on a separate CPU core feeding the GPU," the company told developers. "This can increase, or in some cases, even double the performance of OpenGL-based applications."



QuickTime enhancements



Meanwhile, Apple said improvements to QuickTime in Leopard will offer developers an easy way to build video capture into their applications.



By using the QTKit Capture API (application programming interface), developers will be able to instruct their applications to capture video from an iSight, USB (VDC) and FireWire (IIDC) devices, and DV cameras.



"QuickTime's plumbing is receiving significant upgrades in Leopard," the company said. "There have been significant enhancements in handling the H.264 encoding. Also, transparent alpha layers, an optional part of the H.264 specification, are now supported in H.264-based QuickTime movies."



QuickTime under Leopard will also support 64-bit, Apple added. However, the functionality for 64-bit QuickTime will only be provided via the QTKit framework. The existing C-based QuickTime API will continued to be supported only in 32-bit mode.



Xray for developers



Also of interest to Leopard developers is a new timeline-based performance visualization tool Apple is rolling out called Xray, which gives programmers the ability to see how their application works in a completely new and intuitive way.



"It let's you watch CPU, disk I/O, memory usage, garbage collection, events, and more in the form of graphs tied to time," the company explained. "For example, you can track UI events in real-time and see how they affect I/O activity and CPU load at the same time. And, once the data has been collected, you can scan back and forth through your application's run and see exactly what occurred, and when."
«1

Comments

  • Reply 1 of 31
    boogabooga Posts: 1,082member
    I don't know the details of Apple's OpenGL pipeline, but it sounds like it will provide the most benefit for machines with strong CPUs but weak GPUs. In OpenGL, the stack is required to implement all standard functionality and any available extensions, even if the GPU doesn't. That means that on some machines, more processing is done on the CPU instead, to a greater or lesser degree. If that CPU work is done in a separate thread, it could certainly speed things up, but it seems like you'd get the most speedup on the worst-configured machines, and a top-of-the-line system might not see as much.
  • Reply 2 of 31
    "It let's you watch CPU, disk I/O, memory usage, garbage collection, events, and more in the form of graphs tied to time,"



    This reminds me of something I wish Apple would make available.

    Core Image, Core Video, Core Audio, Core Data, Core Animation....Core Graphing!



    Apple's Keynote and most of it's other apps make very nice looking graphs.

    However, I've seen some pretty hideous graphs in many third party apps.

    Apple should make available an API for making slick looking graphs easily.
  • Reply 3 of 31
    melgrossmelgross Posts: 33,510member
    Quote:
    Originally Posted by Booga


    I don't know the details of Apple's OpenGL pipeline, but it sounds like it will provide the most benefit for machines with strong CPUs but weak GPUs. In OpenGL, the stack is required to implement all standard functionality and any available extensions, even if the GPU doesn't. That means that on some machines, more processing is done on the CPU instead, to a greater or lesser degree. If that CPU work is done in a separate thread, it could certainly speed things up, but it seems like you'd get the most speedup on the worst-configured machines, and a top-of-the-line system might not see as much.



    It looks as though it will be spun off as one thread. I'd rather see it having the capability of being multi-threaded, if possible.



    With 4 and eight cores, using one core will be wasting considerable resources. I see no reason why OpenGl couldn't use two, or even four cores, if required.
  • Reply 4 of 31
    Sweet. Can someone tell me what this means for the PowerPC crowd?

    Will 10.5 be faster on my G4?
  • Reply 5 of 31
    melgrossmelgross Posts: 33,510member
    Quote:
    Originally Posted by nascarnate326


    Sweet. Can someone tell me what this means for the PowerPC crowd?

    Will 10.5 be faster on my G4?



    There will be some speedups, yes. Mostly if you have two cpu's though. That's been true for a while.
  • Reply 6 of 31
    If they're coming out with OGL 2.1, that means that the G80 (8800GTX/GTX) will make its way into the Mac Pro. Oh boy! I love x86!
  • Reply 7 of 31
    dfilerdfiler Posts: 3,420member
    It is likely that there will be some optimizations that will be cross-platform. Better threading would be one of these if you're on a DP G4.



    However, I'd be willing to bet that almost all optimizations will be geared toward intel hardware. One indication of this is the significant rework of h.264 encoding. It is currently fastest on a quad G5, outstripping even the top of the line Mac Pro.



    Nearly all Macs being sold are now intel based. Apple has probably finished off basic support and moved and is now free to optimize for the new architecture.
  • Reply 8 of 31
    dacloodacloo Posts: 890member
    damn. Alpha layers in video. Wow. Finally, no more rendering thousands of alpha images from Maya, and merging them in a painfully slow process inside a video editor.
  • Reply 9 of 31
    walshbjwalshbj Posts: 864member
    Will my DP G5 encode 264 faster with these QT enhancements? Will it take advantage of the G5 64 bit potential?



    I've never understood what the capabilities of my G5 really are in terms of being 64 bit.
  • Reply 10 of 31
    Cwazy AI...it's not OpenGL 2.1 that may double performance...it's the multi-threadedness of the whole thing.



    In fact, this theoretical doubling of the performance exists under 10.4.7 for Mac Pros and 10.4.8 for all Macs with more than one CPU.
  • Reply 11 of 31
    hirohiro Posts: 2,663member
    Quote:
    Originally Posted by melgross


    It looks as though it will be spun off as one thread. I'd rather see it having the capability of being multi-threaded, if possible.



    With 4 and eight cores, using one core will be wasting considerable resources. I see no reason why OpenGl couldn't use two, or even four cores, if required.



    It would be a major bitch to multi-thread the OpenGL stack itself, especially since OpenGL is a single state machine. This way OpenGL does it's thing on one processor and the GPU while all the other stuff can happen on the other CPU(s). It should have been this way for a long time but pre-Tiger kernel issues got in the way.



    I think it is pretty unlikely that a CPU will be a bottleneck for quite awhile now, we will be back to GPU and CPU<-->GPU bandwidth being the bottlenecks.



    And the extra time on the core OpenGL is running on isn't wasted, it becomes available to other threads or processe. OpenGL calls just get core affinity and a high priority, that doesn't limit doing anything else on that core when OpenGL doesn't need it.
  • Reply 12 of 31
    I guess you will see a speed up like Quartz 2D extreme brough to Tiger users.
  • Reply 13 of 31
    eckingecking Posts: 1,588member
    Will this make the current generation of macbooks and mac minis perform better? In what way?
  • Reply 14 of 31
    lundylundy Posts: 4,466member
    Leopard has a new Cocoa API (NSOperation) that lets developers set priority and dependence for different operations. The system then translates that into threads which have the proper priority relative to each other and the proper code to respect the dependencies that their data structures have.
  • Reply 15 of 31
    melgrossmelgross Posts: 33,510member
    Quote:
    Originally Posted by Hiro


    It would be a major bitch to multi-thread the OpenGL stack itself, especially since OpenGL is a single state machine. This way OpenGL does it's thing on one processor and the GPU while all the other stuff can happen on the other CPU(s). It should have been this way for a long time but pre-Tiger kernel issues got in the way.



    I think it is pretty unlikely that a CPU will be a bottleneck for quite awhile now, we will be back to GPU and CPU<-->GPU bandwidth being the bottlenecks.



    And the extra time on the core OpenGL is running on isn't wasted, it becomes available to other threads or processe. OpenGL calls just get core affinity and a high priority, that doesn't limit doing anything else on that core when OpenGL doesn't need it.



    While that's all true, and why I said "if possible", one core will quickly lag the performance of the rest of the program as the number of cores continues to increase.



    In an 8 core machine, not very far away now, Open Gl will still hold everything else back.
  • Reply 16 of 31
    hirohiro Posts: 2,663member
    Quote:
    Originally Posted by melgross


    While that's all true, and why I said "if possible", one core will quickly lag the performance of the rest of the program as the number of cores continues to increase.



    In an 8 core machine, not very far away now, Open Gl will still hold everything else back.



    Not at all. Now we can do all the other stuff that everyone has resisted doing because there isn't enough CPU to go around. Gaming AI is still on about a 10% budget. Newer more realistic physics budgets have crowded an increase in AI cycles out. More processor means more network packet compression making it possible to raise simultaneous networked players by an order of magnitude. More processor unleashes all these with essentially zero negative effect on an OpenGL stack running on one CPU.



    Also remember, not all graphics related processing is openGL stack related. There is a lot of parallelism that can be wrung out in deformable geometry computation, fractally generated level-of-detail for both polys and textures, pre-stack culling and frustrum cropping, all in old-fashioned C++ code. Then the results can be forwarded to the stack for rendering. All techniques used for the last 10 years on big iron SGI Reality Monster rendering mainframes.



    Soon an 8-core desktop with a paired-SLI card will outclass (OK hyperbole), equal a circa 2000 $5 million hardware install.



    ouch.



    Anyone have any questions why SGI faded away?



    Now just relearn the old big $$ simulation Performer tricks for the next couple generations of game engine design.
  • Reply 17 of 31
    almalm Posts: 111member
    I wonder if new GL will have antialiasing already. And means to turn it on/off globally, app-independently, like in windows drivers. Same card in Windows (run on MacPro) show nice smooth AA, and in OS X it's all jagged like hell. There is a way to force AA using some tool from XCode, but it look crappy and performance drop to unusable level.
  • Reply 18 of 31
    Is Open GL 2 far behind Direct X 10?



    Is the new threading stuff exciting? It seems so to be. That will maybe even up some of the frame rate disparity we've seen between the same cards performing on Windows/Mac?



    Lemon Bon Bon
  • Reply 19 of 31
    http://www.barefeats.com/wowmt.html



    Boom!



    If that's not impressive...nothing is.
  • Reply 20 of 31
    MarvinMarvin Posts: 15,312moderator
    Quote:
    Originally Posted by kim kap sol


    http://www.barefeats.com/wowmt.html



    Boom!



    If that's not impressive...nothing is.



    Finally, framerates over 100fps. I don't like that they saw no gain on the MBP though. Maybe they weren't using 10.4.8.
Sign In or Register to comment.