The Speed of the 970

Discussion in 'Wasteland' started by XnavxeMiyyep, Apr 30, 2003.

  1. macrumors 65816


    I heard somewhere(I think Lowend Mac), that the 970 would go 4 times faster per mhz than the G4. Let's say a Dual 1.42 Ghz G4 goes about the same speed as a 2 Ghz Pentium 4. (just an estimate, the G4 is probably much faster). Wouldn't that mean a 970 of the same speed would go at the speed of an 8 Ghz Pentium 4? Anyway, what do you think about the speed difference?
  2. macrumors 604


    Well, there are many elements that contribute to the overall
    speed. The 970 won't offer many benifits when running 32 bit
    apps. So I wouldn't expect 4 ghz speeds from a 1 ghz 970.
  3. macrumors G5

    Sun Baked

    The single biggest factor that will lead to the performance boost will be -- a faster bus (talking actual data moved per second, not MHz).

    The second will be an extra FP unit...

    And the third boost will be larger available memory space (fewer page swaps).


    Though if the rumors of Motorola killing off the 7457-RM are true -- we'll never find out how much the MPX bus was a bottleneck (G4s 1 GB/s MPX bus is definitely a lot less than the 970s 6 GB/s).
  4. macrumors 6502

    Speed of the 970

    I wouldn't worry about whether the PPC 970 allows bragging rights over Intel's Pentium4 line. It is plenty fast, it is 64 bit, and it has a pedigree roadmap that would make SUN drool.

    The more important question is, if they do adopt the 970 (cross fingers), what new, and interesting things could the PowerMac do?

  5. macrumors 6502a


    Yes.. no benefits.. that amazing bus which will allow much better memory throughput will have no impact whatsoever if the app is 32-bit.
  6. macrumors 603


    same to you, buddy. you're getting reported.
  7. macrumors 604


    what are you on?
    Higher bus speed will create performance in both 64 AND 32 Bit apps... There is currently a bottleneck with the G4 having a high clock speed but relatively small bus speed.

    look at others 400, 533, we got what? 166?
  8. Moderator emeritus


    come nice ;)
  9. macrumors 6502

    Re: The Speed of the 970

    No, it is not "4 times faster." If you look at the SPECint2000 and SPECfp2000 scores, the 970 is roughly twice as fast as the 7455 at equal clock speeds (a little less than twice as fast for integers, and much more than twice as faster for floating point). So, all other things being equal, this would suggest that a 1.8 Ghz PPC 970 would be equivalent to a 3.6 Ghz 7455 (which would certainly be at least as fast as a 3.6 Ghz P4).

    However, not all other things are equal! First of all, you might question whether the SPEC2000 suite is an accurate representation of the types of programs you are planning to run. Certainly a valid question. BUT, even if you are running the types of tasks in the SPEC2000 suite, there is still the problem that SPEC is HIGHLY compiler dependent. So, for example, the P4 (and to a lesser extent the Athlon and PIII as well) score very well in the SPEC suite because they use the Intel Reference Compiler. It's a compiler that provides very good performance, but it's not what people actually use (they use Microsoft Visual C++, which is a compiler that provides....not so good performance...perhaps no surprise given the company that made it). So if you looked at SPEC scores, you would actually think that the P4 is faster than the G4 PER CLOCK CYCLE! Yes, according to SPECint2000 and SPECfp2000, a 700 mhz P4 (or possibly even 500 mhz, for terms of floating point), should be faster than a G4. But while the P4 is a very impressive chip with a very high clock speed and all, it does NOT actually keep up with the G4 on a PER CLOCK CYCLE basis in most real world applications, let alone beat it.

    The point of all of this is that the G4 SPEC scores that I used to compare the 7455 and the 970 are generated using GCC (Gnu C Compiler). We don't know what compiler IBM used to estimate the PPC 970 SPEC scores. If they used GCC (I doubt it), then comparing the 970 scores and the 7455 scores may be a valid exercise. If they used their own compiler, and it is better optimized than GCC (at least in terms of the SPEC suite), then all bets are off - maybe the 970 will be twice as fast as the 7455 at a given clock speed, or maybe it won't be. You just can't compare the results, especially if most of the apps running on the 970 will be compiled using GCC rather than the IBM compiler.

    I suspect the story may be something similar to what happened when the 7450 (G4e) was released or when the Pentium 4 was released. Initially both had pretty disappointing performance, but both improved as software was better optimized and recompiled for them. Now, unlike in those cases (where the new chips could barely beat the old chips they replaced), I am sure the 970 will destroy the 7455 when it is released, but it may not be by quite as large a margin as you would expect from the SPEC scores. But that margin will grow over time as apps are recompiled and so forth.

Share This Page