A couple questions re: PPC 970 chip

Posted:
in Future Apple Hardware edited January 2014
With all this talk of the (possible) use the new PPC 970 chip in Macs, I have a couple of questions to ask:



1) MHz for MHz, is the 970 faster than the current G4? If Apple ships a 2x1.25 Ghz 970 based machine, will it be faster? If so, how much (roughly)?



2) What are the implications of it being a 64-bit chip? Will it be faster due to this? What is the advantage here?



Thanks (and try to keep any answers simple, please)
«13

Comments

  • Reply 1 of 45
    bungebunge Posts: 7,329member
    There will probably be others that could answer this better, but I'll give it a try.



    1) Yes it will be faster clock for clock, but how much is up for debate. I would guess that it's significantly enough that you could feel the difference between the two machines you describe, but we won't know for sure until they're released.



    2) A 64 Bit chip actually slows down a program just a tad, although most likely not enough to feel the difference in the slightest. The benefits are numerous, but nothing urgent for most people. The maximum amount of RAM a machine can have will be higher, the number of colors a machine could handle will be able to increase when graphics cards are there to exploit it, perhaps nothing major but it's the future of chips in any case.



    If you need a machine now I wouldn't wait.
  • Reply 2 of 45
    [quote]Originally posted by bunge:

    <strong>There will probably be others that could answer this better, but I'll give it a try.



    1) Yes it will be faster clock for clock, but how much is up for debate. I would guess that it's significantly enough that you could feel the difference between the two machines you describe, but we won't know for sure until they're released.</strong><hr></blockquote>

    IIRC, the estimated benchmarks would put a single 970 about on par with a 2GHz P4.



    [quote]<strong>2) A 64 Bit chip actually slows down a program just a tad, although most likely not enough to feel the difference in the slightest.</strong><hr></blockquote>

    Actually, the 970 shouldn't suffer any real performance hit since its 64 bitness only affects apps that are written to run at 64 bit. Otherwise it behaves just like any other 32 bit PPC and will run 32 bit code just as fast. This was part of the PPC ISA from the outset, it is actually a 64 bit ISA. It's just designed to be 32 bit compatible as well, and the 970 takes advantage of this by running 32 code as if it were just "half 64 bit" (loosely speaking).
  • Reply 3 of 45
    blarkblark Posts: 11member
    [quote]Originally posted by bunge:

    <strong>2) A 64 Bit chip actually slows down a program just a tad, although most likely not enough to feel the difference in the slightest. The benefits are numerous, but nothing urgent for most people. The maximum amount of RAM a machine can have will be higher, the number of colors a machine could handle will be able to increase when graphics cards are there to exploit it, perhaps nothing major but it's the future of chips in any case.</strong><hr></blockquote>



    To clarify, a 64-bit chip is capable of addressing far more RAM than a 32-bit chip. For most current 32-bit chips, that most RAM they can address is 4 Gb (that is the upper limit of RAM you can install on the machine, if the hardware allows for it). A 64-bit chip will be able to address much more RAM (4 terabytes is the figure often thrown around, but it is actually higher than that IIRC). That applications that can make use of that much RAM (currently) are primarily scientific, large database, and 3D applications. The 970 can run in 32-bit mode, and will probably see no speed penalties when doing so. there is a slight speed hit when running in 64-bit mode, as memory address pointers etc. will be larger.



    The 64-bitness of a chip has nothing to do with graphics and/or display capability. Current programs like Photoshop and After Effects can manipulate and display graphics in 24- or 48-bit mode (8 or 16 bits per channel or red, green, and blue). Obviously, the greater the bit depth, the larger the file size, and the more RAM is necessary, but current architectures work well for virtually all 2D imaging and compositing work.



    Future colour schemes will probably migrate toward High-Dynamic Range Imaging (HDRI), currently used in some 3D contexts. Here, colour values are stored as floating point numbers rather than as integers (as in current 8- and 16- bit per channel colour). This allows for essentially unlimited variation in brightness and darkness, and more closely emulates luminance values in the real world.



    Imagine taking an HDRI image with your digicam. To be extreme, it could store the values of brightness from the deepest shadows to the variation in brightness across the sun's disc. After the fact, you could re-expose the picture however you like in software. This would obviously be far superior to what conventional photography can do easily...



    Blark
  • Reply 4 of 45
    gargar Posts: 1,201member
    [quote]Originally posted by Tomb of the Unknown:

    <strong>

    IIRC, the estimated benchmarks would put a single 970 about on par with a 2GHz P4.

    </strong><hr></blockquote>



    what's the speed of that specific single ppc970?



    1.8Ghz? <img src="graemlins/bugeye.gif" border="0" alt="[Skeptical]" />



    [ 01-20-2003: Message edited by: gar ]</p>
  • Reply 5 of 45
    [quote]Originally posted by gar:

    <strong>



    what's the speed of that specific single ppc970?



    1.8Ghz? <img src="graemlins/bugeye.gif" border="0" alt="[Skeptical]" />



    [ 01-20-2003: Message edited by: gar ]</strong><hr></blockquote>

    Oops. Yes, that's the estimated benchmarks at 1.8Ghz I was referring to.



    You can find more info <a href="http://www.extremetech.com/article2/0,3973,635220,00.asp?kc=ETTH102099TX1K0100486"; target="_blank">in this link.</a>



    [ 01-20-2003: Message edited by: Tomb of the Unknown ]</p>
  • Reply 6 of 45
    Thanks for the clear answers.



    A clarification, if you please:



    1.8GHz 970 = 2GHz Pentium 4?



    Doesn't a current 1.25GHz G4 = 2GHz Pentium 4?



    Therefore, 1.25GHz G4 = 1.8GHz 970?



    I sure hope not! I want it to be much faster.



    [ 01-20-2003: Message edited by: Borborygmi ]</p>
  • Reply 7 of 45
    gargar Posts: 1,201member
    [quote]Originally posted by Tomb of the Unknown:

    <strong>

    Oops. Yes, that's the estimated benchmarks at 1.8Ghz I was referring to.



    You can find more info <a href="http://www.extremetech.com/article2/0,3973,635220,00.asp?kc=ETTH102099TX1K0100486"; target="_blank">in this link.</a>



    [ 01-20-2003: Message edited by: Tomb of the Unknown ]</strong><hr></blockquote>



    that's very bad news (damn, i wish we lived in creese, 200bc)



    the story apple told us back in 1999 was that a G4 runs almost twice as fast per clock cycle as the P4 did. so a 500Mhz G4 equals a 1Ghz P4.

    or something like that.



    i don't get it, somewhere on these boards i read something completely different like a ppc970 @1.8Ghz kicks a P4 @3.6Ghz ass big time.
  • Reply 8 of 45
    Blark- I'm afraid that I'm going to have to correct you there a bit.

    The 970 is only (IIRC) a 42 bit addressing chip, its 64-bitness comes from having 64 bit integer registers.
  • Reply 9 of 45
    wormboywormboy Posts: 220member
    Oh man, there is sooo much disinformation in this thread it is sickening. The most educated estimates that I've seen come in with the single PPC970 running about the same as a PPC G4 of twice the clock... but of course a statement like that is as wrong as any other in this thread. It would depend entirely upon your metric. If it were released right now, the PPC970 would be the champ. Extra FPU, mindboggling bandwide... and fully capable of appearing in efficient MP configurations!
  • Reply 10 of 45
    outsideroutsider Posts: 6,008member
    Another thing to consider is Apps that make best use of the L3 cache like VirtualPC. It's well known that the 970 does not have support for L3 cache (but it does have 512KB of L2) and VirtualPC works best with a big L3. We'll only know when we have a G4 and 970 system side by side if the 970's wide and fast bus can make up for the lack of L3. I think things will be favorable for the 970 system.
  • Reply 11 of 45
    [quote]Originally posted by Pilmour Boy:

    <strong>Blark- I'm afraid that I'm going to have to correct you there a bit.

    The 970 is only (IIRC) a 42 bit addressing chip, its 64-bitness comes from having 64 bit integer registers.</strong><hr></blockquote>



    AFAIK, that is correct. And, BTW, the G4 has 36-bit addressing, not 32.



    And I don't think that a 64 bit processor will be even slightly slower when in 64bit mode. Some people have been saying that it will slow a little due to more data being moved. It is my understanding that with the 64bitness, the data paths will double in width (except, probably, Altivec). Thus one word of data would move at the same speed as before.



    Is that correct? Or will the paths to memory stay as current widths?
  • Reply 12 of 45
    First I'll reiterate wormboy's point: relative performance will depend on your metric.



    IIRC, based on the only real metric we have for the 970 (estimated SPECmarks), a 1.8 GHz will be about equivalent to 2.5 GHz Pentium4. The G4 shows poorly in SPEC so the 970 is about 4 times faster. This is largely due to memory bandwidth and so is a rather bogus comparison. Realistically I'd expect the 970 to be roughly twice as fast as a G4 at the same clock rate, especially for floating point tasks. For AltiVec tasks, most of which are memory bound, the 970 will be much faster but that will depend on the memory subsystem. In normal scalar compiler generated code the 970 should have quite an advantage over the G4 thanks to its better out of order execution capabilities.



    BTW: the G4 may have a better IPC than the Pentium4, but that advantage has been overshadowed by the memory subsystems, clock rate, and improved caches & tables in the P4.
  • Reply 13 of 45
    [quote]Originally posted by gar:

    <strong>the story apple told us back in 1999 was that a G4 runs almost twice as fast per clock cycle as the P4 did. </strong><hr></blockquote>

    No, that would be true of the Pentium III perhaps and mostly only when taking advantage of Altivec. The current line of G4s (G4+) run about 20-40% faster than a P4 at the same clock speed in terms of actual work done. But of course the P4 makes up for this by clocking way higher than the G4+. So you need dual 1.25GHz chips to approximate a 3 Ghz P4 (and even then, for some things the P4 is still way faster).



    Mind you, this is all very fuzzy and hotly contested in both directions since it's really not fair to compare differenct CPU architectures as its almost impossible to exactly and quantitatively measure performance across architectures.
  • Reply 14 of 45
    gargar Posts: 1,201member
    [quote]Originally posted by Tomb of the Unknown:

    <strong>

    No, that would be true of the Pentium III perhaps and mostly only when taking advantage of Altivec. </strong><hr></blockquote>



    ah, the snail campaign

    that where the good times, back then, with beige G3's, bondy blue imac's and sweet wallstreets.
  • Reply 15 of 45
    progmacprogmac Posts: 1,850member
    [quote]Originally posted by Tomb of the Unknown:

    <strong>

    No, that would be true of the Pentium III perhaps and mostly only when taking advantage of Altivec. </strong><hr></blockquote>



    if a 500mhz G4 is equivalant to a 1ghz P3, that is much more significant than it being the same as a 1ghz P4. A P3 at 1ghz performs about the same as a P4 at 1.3-1.4 ghz. Marketing move from intel.
  • Reply 16 of 45
    [quote] Perhaps more importantly, the front-side bus can transfer up to 7.2 Gbytes per second, roughly four times the bandwidth of the current Pentium 4 front-side bus, <hr></blockquote>



    I like the sound of that.



    The 970s looking to blow the doors of the G4 and give the Pentium 4 a black eye. And that's not even starting on dual and quad and...bleep formations...



    Twice the G4 output. A 1 gig 970 is a 2 gig G4. Oooh. A 1.8 gig 970 is a 3.6 gig G4. Ahhh.



    Whether it works out that way in reality, who knows?



    Still, with that extra fpu in there and that bandwidth...it should crush the G4 at anything vaguely Lightwave-ish...drool...der-rool....



    Lemon Bon Bon
  • Reply 17 of 45
    powerdocpowerdoc Posts: 8,123member
    [quote]Originally posted by Programmer:

    <strong>First I'll reiterate wormboy's point: relative performance will depend on your metric.



    IIRC, based on the only real metric we have for the 970 (estimated SPECmarks), a 1.8 GHz will be about equivalent to 2.5 GHz Pentium4. The G4 shows poorly in SPEC so the 970 is about 4 times faster. This is largely due to memory bandwidth and so is a rather bogus comparison. Realistically I'd expect the 970 to be roughly twice as fast as a G4 at the same clock rate, especially for floating point tasks. For AltiVec tasks, most of which are memory bound, the 970 will be much faster but that will depend on the memory subsystem. In normal scalar compiler generated code the 970 should have quite an advantage over the G4 thanks to its better out of order execution capabilities.



    BTW: the G4 may have a better IPC than the Pentium4, but that advantage has been overshadowed by the memory subsystems, clock rate, and improved caches & tables in the P4.</strong><hr></blockquote>

    The P4 is very strong in Spec, compared to an itanium. Intel is very good to optimize Spec. I don't know how good is IBM at this game compared to Motorola.

    The 970 has two FP unit, compared to one FP unit for the G4, is memory bandwitch will be the best in the desktop aera. The L2 cache wil be twice the size of a G4, he will lack L3 cache (but the memory bandwitch will be 4 time than the G4, thus the L3 cache will be less important).



    My guess will be at equal mhz : 50 % increase in integer operation, 20 % in altivec and 100 % in FP unit. Thus a 1,8 ghz 970 would be equal to a 2,7 GHZ G4 for int and a 3,6 ghz G4 for FP and a 2,1 ghz G4 for altivec. Except some altivec stuff it will kick the ass of any dual G4.

    I will take a single 970 at the place of any dual G4.
  • Reply 18 of 45
    krassykrassy Posts: 595member
    [quote]Originally posted by Powerdoc:

    <strong> &lt;...&gt; (but the memory bandwitch will be 4 time than the G4, thus the L3 cache will be less important) &lt;...&gt;

    &lt;...&gt;I will take a single 970 at the place of any dual G4.</strong><hr></blockquote>



    is there any data of how fast the current L3s transfers are? and i think apple could bring a dual 1.5 so that the "PowerMacs" are really worth the 1700$ to pay (at minimum)
  • Reply 19 of 45
    stoostoo Posts: 1,490member
    Current L3 cache is 250MHz DDR RAm, 64 bits wide, giving a bandwidth of 4GB/s. Not that much faster than the 970's up or down FSB .
  • Reply 20 of 45
    [quote]Originally posted by Stoo:

    <strong>Current L3 cache is 250MHz DDR RAm, 64 bits wide, giving a bandwidth of 4GB/s. Not that much faster than the 970's up or down FSB .</strong><hr></blockquote>



    True, but the latency will be far, far higher. Currently, the L3 can deliver any cache line with just a 16 cycle latency, and keep it up with IIRC, one more cache line every four cycles.
Sign In or Register to comment.