CONFIRMED: MPC 7457 with up to 1833 Mhz

1235789

Comments

  • Reply 81 of 173
    mokimoki Posts: 551member
    [quote]Originally posted by KidRed:

    <strong>



    Again, we will get the 970 in the fall of next year. In Jan/Feb we will get a new G4 with new mobo possibly the one mentioned in this thread.</strong><hr></blockquote>



    Hey KidRed -- drop me an eMail -- the hotmail address you have registered here doesn't seem to work anymore...
  • Reply 82 of 173
    amorphamorph Posts: 7,112member
    Why? It's better than what we have, it will probably perform well, considering that performance has scaled surprisingly well with clockspeed even given the MPX bus (cache is useful!). The die shrink means faster G4s in PowerBooks, and possibly a G4 in the iBook as well.



    Unless you want Apple to be stuck at 1.25GHz for an entire year, until the 970 is expected to appear? Even if it appears sooner, the PB, iMac, eMac and iBook need somewhere to go...
  • Reply 83 of 173
    [quote]Originally posted by mmicist:

    <strong>



    IBM have a very nice set of compilers (xlc/xlf) which are able to optimise for the POWER4, and I'm sure they have been altered to optimise for the 970 (it wouldn't take that much work), this will be what they used for assessing the SPEC scores.

    The change from gcc 2.95 to 3.2 for SPEC scores is not very great, maybe 10-20% on a good day, and gcc cannot even compile all the SPECfp subprograms, as it cannot do FORTRAN90.

    Forget any possibility of doubling these scores.



    [edit for spelling]



    michael



    [ 10-23-2002: Message edited by: mmicist ]</strong><hr></blockquote>



    Thank you for your concern, but scalar code from 3.2 was 64% faster than vectorized code from 2.95 for the dual 1Ghz system Eugene tested it on, if I remember correctly. If you have ever compiled for PPC in the last couple of months, you'd know that there IS a big difference.



    2) The 970 is not a Power4, it's a PPC. They have different instruction sets.



    3) I guarantee that IBM hasn't a compiler optimized adequately for VMX/Altivec.



    Please do your research.
  • Reply 84 of 173
    [quote]Originally posted by Splinemodel:

    <strong>



    Thank you for your concern, but scalar code from 3.2 was 64% faster than vectorized code from 2.95 for the dual 1Ghz system Eugene tested it on, if I remember correctly. If you have ever compiled for PPC in the last couple of months, you'd know that there IS a big difference.



    2) The 970 is not a Power4, it's a PPC. They have different instruction sets.



    3) I guarantee that IBM hasn't a compiler optimized adequately for VMX/Altivec.



    Please do your research.</strong><hr></blockquote>



    Did you do all of your research?

    I don't think you can necessarily say that IBM does not have a optimzed compiler...they just may.
  • Reply 85 of 173
    mmicistmmicist Posts: 214member
    [quote]Originally posted by Splinemodel:

    <strong>



    Thank you for your concern, but scalar code from 3.2 was 64% faster than vectorized code from 2.95 for the dual 1Ghz system Eugene tested it on, if I remember correctly. If you have ever compiled for PPC in the last couple of months, you'd know that there IS a big difference.



    </strong><hr></blockquote>



    I gave figures for improvements in SPEC scores, not any other type of programme. I compile for PPC every day, including some of the SPEC sub-units.



    [quote]

    <strong>



    2) The 970 is not a Power4, it's a PPC. They have different instruction sets.



    </strong><hr></blockquote>



    The POWER4 is a PPC. The 970 and the POWER4 don't have exactly the same instruction sets, the 970 includes VMX instructions, that's as much as I can be certain about at the moment.



    [quote]

    <strong>



    3) I guarantee that IBM hasn't a compiler optimized adequately for VMX/Altivec.



    </strong><hr></blockquote>

    A) What do you mean by optimised for VMX? Able to produce code for Altivec, or one that auto-vectorises?

    B) What do you mean by adequately? Definitiely a weasel word.

    C) How can you guarantee this?



    [quote]

    <strong>



    Please do your research.

    </strong><hr></blockquote>



    Please don't make assumptions.



    michael
  • Reply 86 of 173
    bigcbigc Posts: 1,224member
    Iseries info <a href="http://www.iseriesnetwork.com/resources/artarchive/index.cfm?fuseaction=viewarticle&CO_ContentID=1523 8&channel=art&PageView=Search" target="_blank">When is PowerPC not PowerPC</a>



    [ 10-24-2002: Message edited by: Bigc ]</p>
  • Reply 87 of 173
    strobestrobe Posts: 369member
    [quote]Originally posted by ruthifren:

    <strong>



    This is correct. A lot can be learned from Apple's internal hardware roadmap. You used to see a lot of 4-way G4 multiprocessor prototypes with 1 main processor responsible for processing, 1 dedicated to sheets, 1 for the genie effect, and the remaining processor doing gaussian blur on text. This changed a bit with the advent of Quartz Extreme when it was realized that current state of the art 3D hardware was sufficient for text blurring. Add to that the prospect of "scale" effect replacing genie and it's feasible to ship a consumer class machine with just the 2 G4 processors which is exactly what they did.</strong><hr></blockquote>



    I don't know what you've been reading, but I doubt Quartz Extreme performs "text blurring" even if by this you mean anti-aliased text. Quartz Extreme uses the GPU to blend surfaces, not to render their contents.



    FYI: anti-aliased text isn't a blur. QuickDraw's "font smoothing" may be, but since 10.1.5 you shouldn't be using crappy font smoothing anyway.



    Sounds like you're talking out of your butt.
  • Reply 88 of 173
    strobestrobe Posts: 369member
    [quote]Originally posted by Outsider:

    <strong>



    It also helps that Linux for PPC is getting more and more optimized for Altivec.</strong><hr></blockquote>



    erm...



    Where did you get that idea?



    Perhaps some video/audio codecs are using VMX for FFT, but I don't see this as anything to do with Linux/PPC.
  • Reply 89 of 173
    outsideroutsider Posts: 6,008member
    [quote]Originally posted by strobe:

    <strong>



    erm...



    Where did you get that idea?



    Perhaps some video/audio codecs are using VMX for FFT, but I don't see this as anything to do with Linux/PPC.</strong><hr></blockquote>



    erm..



    <a href="http://www.redhat.com/about/presscenter/2002/press_motorola.html"; target="_blank">it's old news?</a>
  • Reply 90 of 173
    tjmtjm Posts: 367member
    [quote]Originally posted by strobe:

    <strong>



    I don't know what you've been reading, but I doubt Quartz Extreme performs "text blurring" even if by this you mean anti-aliased text. Quartz Extreme uses the GPU to blend surfaces, not to render their contents.



    FYI: anti-aliased text isn't a blur. QuickDraw's "font smoothing" may be, but since 10.1.5 you shouldn't be using crappy font smoothing anyway.



    Sounds like you're talking out of your butt.</strong><hr></blockquote>



    Umm, you may want to check with the publisher of your browser for an update. Looks like you need an update on the SSR (Sarcasm and Satire Recognition) module.
  • Reply 91 of 173
    [quote]Originally posted by Bigc:

    <strong>Iseries info <a href="http://www.iseriesnetwork.com/resources/artarchive/index.cfm?fuseaction=viewarticle&CO_ContentID=1523 8&channel=art&PageView=Search" target="_blank">When is PowerPC not PowerPC</a>



    [ 10-24-2002: Message edited by: Bigc ]</strong><hr></blockquote>



    Good article. As it says near the end though, while they do tend to refer to all these chips as "PowerPC", many of them actually support multiple architectures. I don't recall seeing anywhere in the 970 presentation where they explicitly stated that it did not support those other architectures, so I don't think we can assume either way.
  • Reply 92 of 173
    ed m.ed m. Posts: 222member
    Well, who better to ask than someone who was on the PPC design team? I asked my friend and to paraphrase what he said that the short answer (even after reading the article) is: yes, the Power4 and 970 (since it's a derivative) is/are a PowerPC(s).



    Simply put, the PowerPC has a specific technical meaning defined in architectural documents that ALL three AIM members agreed upon. Furthermore, IBM has publicly said that the 970 and Power4 are fully compatible with that definition. Incidentally, Power3 was also fully a PowerPC.



    --

    Ed M.
  • Reply 93 of 173
    bigcbigc Posts: 1,224member
    My jist of the article was that any of their chips can be a PowerPC with the switch of a couple bits. Pretty amazing.
  • Reply 94 of 173
    zosozoso Posts: 177member
    [quote]Originally posted by strobe:

    <strong>Sounds like you're talking out of your butt.</strong><hr></blockquote>



    Get a life...



    ZoSo
  • Reply 95 of 173
    Well, ruthifren you've really has made this thread worthwhile; i think yr "informed, well-thought-out comments" & th hilarious responses should be saved & displayed as perfect examples of posting at its peak. Everybody - if u haven't read these on pages 2 & 3 of this thread, please do - & here's a sample:

    ruthifren--

    [quote]

    I'm told that a substantial percentage of the transistors on the new 970 are dedicated to what is being called the WRU (Window Resizing Unit). I saw a technology demonstration behind closed doors where a TextEdit document with over 8 (!) lines of text was sized all the way to the minimum and back again without dropping a frame.



    iCal, iChat, iPhoto, and iSync are extremely bandwidth intensive, power-hungry, and compute bound apps that will continue to require workstation class hardware to keep mouse&lt;-&gt;processor latencies under the 8 second mark.

    <hr></blockquote>

    (....and much more...)

    lemon bon bon--

    [quote]

    I find the last post very enticing. Why? Because it actually seems ......sub solutions which bypass traditional design flaws are being dreamt up by Apple's classic 'make things easier' team.

    I like the notion that Apple is using the advantages of the 970's superior architecture to tackle the speed problems in a sideways fashion.

    I applaud.

    Nice post. There's more than one way to cook an Egg. Intel. Take note. <hr></blockquote>

    strobe--

    [quote]

    quote:

    ------------------------------------------------------------------------

    Originally posted by ruthifren:



    This is correct. A lot can be learned from Apple's internal hardware roadmap. You used to see a lot of 4-way G4 multiprocessor prototypes with 1 main processor responsible for processing, 1 dedicated to sheets, 1 for the genie effect, and the remaining processor doing gaussian blur on text. This changed a bit with the advent of Quartz Extreme when it was realized that current state of the art 3D hardware was sufficient for text blurring. Add to that the prospect of "scale" effect replacing genie and it's feasible to ship a consumer class machine with just the 2 G4 processors which is exactly what they did.

    ------------------------------------------------------------------------



    I don't know what you've been reading, but I doubt Quartz Extreme performs "text blurring" even if by this you mean anti-aliased text. Quartz Extreme uses the GPU to blend surfaces, not to render their contents.



    FYI: anti-aliased text isn't a blur. QuickDraw's "font smoothing" may be, but since 10.1.5 you shouldn't be using crappy font smoothing anyway.



    Sounds like you're talking out of your butt. <hr></blockquote>

    a beautiful bit of bait ruthifren, & u hooked a coupla beauts!! keep it coming... <img src="graemlins/lol.gif" border="0" alt="[Laughing]" />
  • Reply 96 of 173
    krassykrassy Posts: 595member
    [quote]Originally posted by TJM:

    <strong>



    Umm, you may want to check with the publisher of your browser for an update. Looks like you need an update on the SSR (Sarcasm and Satire Recognition) module. </strong><hr></blockquote>





    <img src="graemlins/lol.gif" border="0" alt="[Laughing]" /> <img src="graemlins/lol.gif" border="0" alt="[Laughing]" /> <img src="graemlins/lol.gif" border="0" alt="[Laughing]" />
  • Reply 97 of 173
    Oh, and to answer the original question about whether IBM has a compiler that is optimized for the 970:



    The 970 is, at least, virtually equivalent to the POWER4 running in 32 or 64 bit PowerPC mode and there are several compilers for it. Furthermore, different implementations of a particular architecture can prefer different code organizations and still be compatible. Remember the G4 7400 -&gt; 7450 transition? The 7450 could run all the existing code, but if it was recompiled with a 7450-aware compiler it ran a bit faster and still ran on the 7400. This is because the compiler can often order instructions a little differently to make one chip or another work a little faster. The POWER4 and 970 are very very similar (if not identical) to eachother, but very different than the 7400 or 7450. Just look at the pipeline lengths -- 5, 7, and 14+. Apple and Motorola compilers aren't optimized for the 970/POWER4 (yet) but IBM's are.



    So yes, IBM has a 970 optimized compiler already.
  • Reply 98 of 173
    mokimoki Posts: 551member
    [quote]Originally posted by Splinemodel:

    <strong>



    Thank you for your concern, but scalar code from 3.2 was 64% faster than vectorized code from 2.95 for the dual 1Ghz system Eugene tested it on, if I remember correctly. If you have ever compiled for PPC in the last couple of months, you'd know that there IS a big difference.



    2) The 970 is not a Power4, it's a PPC. They have different instruction sets.



    3) I guarantee that IBM hasn't a compiler optimized adequately for VMX/Altivec.



    Please do your research.</strong><hr></blockquote>



    I don't know of ANY compilers that will "vectorize" code -- not to any extent as to be useful. AltiVec needs to be hand-coded, you don't just throw a compiler switch and get vectorized code.



    It is likely that the binary was indentical in the above comparison, with no AltiVec instructions in it at all (despite "enabling" it via the compiler)
  • Reply 99 of 173
    [quote]Originally posted by moki:

    <strong>

    I don't know of ANY compilers that will "vectorize" code -- not to any extent as to be useful. AltiVec needs to be hand-coded, you don't just throw a compiler switch and get vectorized code.



    It is likely that the binary was indentical in the above comparison, with no AltiVec instructions in it at all (despite "enabling" it via the compiler)</strong><hr></blockquote>



    I don't think you meant "the binary" since the output of the two compilers is clearly different. But you're right, GCC definitely doesn't do any auto-vectorization.



    There is one AltiVec-enabled auto-vectorizing compiler out there, but the name escapes me at the moment. Apparently it does alright, but doesn't approach the benefits of hand written SIMD.
  • Reply 100 of 173
    [quote]Originally posted by Programmer:

    <strong>There is one AltiVec-enabled auto-vectorizing compiler out there, but the name escapes me at the moment.</strong><hr></blockquote>



    <a href="http://www.psrv.com/altivec.html"; target="_blank">VAST</a>



    [quote]<strong>Apparently it does alright, but doesn't approach the benefits of hand written SIMD.</strong><hr></blockquote>



    And C can't match the speed of hand written assembly. Of course, in practice the costs and lack of portability make the speed benefit of using assembly usually quite irrelevant. The same argument may be used for auto-vectorization. Not having to learn Altivec, portable code and a decent speed improvement may be preferable to writing Altivec code. Unfortunately we don't really know how well VAST performs since there are only 'official' benchmarks. Those benchmarks show very good performance relative to hand-coded Altivec. Of course, you can wonder what the quality of the latter is and whether the benchmarks are not biased.



    [ 10-25-2002: Message edited by: wfzelle ]</p>
Sign In or Register to comment.