Speed boosts without Quartz Extreme - gcc 3.x

nostradamusnostradamus
Posted:
in macOS edited January 2014
Rumors are suggesting an instant 10-20% speed boost with Jaguar recompiled under gcc 3.x, and even more so for PPC745x G4 processors. This along with several hundred kernel tweaks and optimizations should give a significant increase to perceptible speed.



Without Quartz Extreme, Jaguar will do to 10.1 what 10.1 did to 10.0.



So yes, you will see a significant speed boost even if your machine does not support Quartz Extreme.



[ 05-06-2002: Message edited by: Nostradamus ]</p>

Comments

  • Reply 1 of 11
    buonrottobuonrotto Posts: 6,368member
    Does it matter which API you use? My impression from the coverage at MacCentral is that C++ apps will see major benefits of stuff like this. Does that mean for Obj-C and Java apps?
  • Reply 2 of 11
    nostradamusnostradamus Posts: 397member
    Well, obj-c will but I don't know about Java and the JVM. Since the whole OS will be recompiled, we might see some benfits trickle down due to gcc 3.x?



    I'm sure Apple has also made several optimizations to the JVM and the kernel. We should see some boost with Java, if not a massive one.



    [ 05-06-2002: Message edited by: Nostradamus ]</p>
  • Reply 3 of 11
    nostradamusnostradamus Posts: 397member
    *bump*
  • Reply 4 of 11
    leonisleonis Posts: 3,427member
    Okay. So a HUGE list of updaters for my existing apps would be done by summer
  • Reply 5 of 11
    bluegeckobluegecko Posts: 11member
    The speed improvements will be for C, C++ and Objective-C. Objective-C message sending may or may not be improved from its current state; I would assume that it will increase about as much as the other two languages, which is to say that it will still be slightly slower but yet an improvement. (Objective-C message sends are about three times slower than a function call in C and about three times faster than Java. The flipside is that the extra functionality you get from ObjC can often result in fewer messages being sent, which evens things out.) Java support is another matter entirely: its speed depends upon the VM on which it runs, and so whether or not Java applications receive a speed boost depends more on whether Jaguar ships with Java 1.4 than on anything else. (I suppose tha the Swing-Aqua layer could be improved as well, but the effects there are likely to be far less noticeable.)
  • Reply 6 of 11
    airslufairsluf Posts: 1,861member
  • Reply 7 of 11
    ghost_user_nameghost_user_name Posts: 22,667member
    Actually, Andrew Welch has been experimenting with the gcc 3.1 beta that's available and said that you can get around a 5% boost in most areas from a simple recompile, not 10-20%. Though, I'm sure much depends on what areas you're optimizing your code.



    Good little moki. moki want banana?
  • Reply 8 of 11
    dennis the phantom menacedennis the phantom menace Posts: 64member
    Still, 5% is better than 0%.
  • Reply 9 of 11
    nostradamusnostradamus Posts: 397member
    [quote]Originally posted by starfleetX:

    <strong>Actually, Andrew Welch has been experimenting with the gcc 3.1 beta that's available and said that you can get around a 5% boost in most areas from a simple recompile, not 10-20%. Though, I'm sure much depends on what areas you're optimizing your code.



    Good little moki. moki want banana? </strong><hr></blockquote>



    ArsTestbenchX scores are about 5-15% faster on all G3/G4's when recompiled over the previous version of gcc.



    5% is being conservative, after all, it's still gcc 3.1 BETA and Jagwire is still a several months away.



    Eugene can elaborate on the ArsTestBenchX scores better than I can.



    [ 05-06-2002: Message edited by: Nostradamus ]</p>
  • Reply 10 of 11
    eugeneeugene Posts: 8,254member
    I really haven't been following Ars testbench, but Nibs, BadAndy and Ccox have been working on optimizations, so the scores aren't purely gcc 2.95.2 vs gcc 3.1. However, I have done my own comparisons with LAME 3.92, and that app alone has seen a 5% encoding speed increase.



    The CFLAGS I used for 3.1:

    [code]-mcpu=7450 -mtune=7450 -mdynamic-no-pic</pre><hr></blockquote>



    and for 2.95.2:

    nothing...since none of the above flags exist in that version.
  • Reply 11 of 11
    ghost_user_nameghost_user_name Posts: 22,667member
    [quote]Originally posted by Nostradamus:

    <strong>5% is being conservative</strong><hr></blockquote>I'd much rather make conservative estimates than wildly optimistic ones. Otherwise, I might get very let down if Jaguar's apps aren't "x times as fast" as I was expecting. I'd rather be pleasantly surprised than mildly disappointed.



    The problem here is that some of the kiddies reading these boards will see "20% speed increase" and get that mantra stuck in their heads and if Apple doesn't deliver at least that well or better, they'll throw a tantrum. Don't you remember the debacle about "removing debug code" that was supposed to (some said) nearly double OSX's speed?
Sign In or Register to comment.