Anyone concerned about L2 cache on G4s?

Jump to First Reply
Posted:
in Current Mac Hardware edited January 2014
Am I the only one who feels a little concerned about the small L2 cache on the current batch of G4s? I mean, I just read a benchmark test that showed in non-Altivec apps the 900MHz G3 in the current iBook actually outperforms (just barely) PowerBooks, eMacs and iMacs with 700MHz-1Gig G4s.



That's supposedly due to the larger/faster L2 cache in the G3, and the G4s WAY outperformed the G3 in Altivec apps. Still, if I'm going to pay the big bucks for a G4 shouldn't it outperform a G3 no matter what app I'm running?

Comments

  • Reply 1 of 5
    mcqmcq Posts: 1,543member
    Not reallly concerned. No contest with current PowerBooks, as they all have 512K L2 Cache, same as the iBooks.



    As far as the iMac/eMac, the performance difference on non-Altivec shouldn't be significant, so the small tradeoff (if any) is more than worth having Altivec available for the apps that take advantage of it.
     0Likes 0Dislikes 0Informatives
  • Reply 2 of 5
    machemmachem Posts: 319member
    Quote:

    Originally posted by MCQ

    Not reallly concerned. No contest with current PowerBooks, as they all have 512K L2 Cache, same as the iBooks.



    As far as the iMac/eMac, the performance difference on non-Altivec shouldn't be significant, so the small tradeoff (if any) is more than worth having Altivec available for the apps that take advantage of it.




    Yeah, my 1.25GHz PB kicks most Macs all over the map, except for duals >800MHz, and G5s. I don't expect it to replace a (similarly-priced) desktop, but it sure does what it is supposed to do, especially in my real-world use.
     0Likes 0Dislikes 0Informatives
  • Reply 3 of 5
    Is there a difference in die size between a current G3 and G4? If not, then I say it is fair game based on 2 different ways to use up die space. One way will give superior vector op performance (by sacrificing some embedded L2 space for a vector module) while the other will give improved operation for general ops (by going for all-L2 with no special acceleration modules). That said, the question really becomes is it right for Apple to charge a premium for G4 chips even if they use the same die size as a G3 (assuming that condition is true, of course). You could argue "yes" in that Altivec certainly does accelerate things substantially given the right conditions, and that represents an added value...or you could argue "no" since it is still the same size slice of silicon which should cost about the same to produce and process no matter what is printed on it.
     0Likes 0Dislikes 0Informatives
  • Reply 4 of 5
    The G3 is much better designed, as far as I know, than the G4. That's why a G3 with altivec is so stimulating a prospect.



    The 750GX reaches 1.1Ghz on a 4 stage pipeline architecture. That's pretty impressive. It has a typical power dissipation of something like 5.6W at 867Mhz. The G4 is not this tight a package.



    However, as you'll see with altivec, a lot of OS X uses it. Some tests run by xlr8yourmac.com a while ago showed that a G4/600 (a 7410, not even 1:1 L2!!!) beat a G3/800 (750FX, nearly identical to 750GX) on almost all tests. The G4's advantage is a greater memory "bandwidth" (they really mean throughput).



    But damn. I want a powerbook with a 1.6Ghz G3+altivec variant. Should be a better package than a 7457/7447, especially when there's no conceivable justification at this time for multiprocessing in a mobile package.
     0Likes 0Dislikes 0Informatives
  • Reply 5 of 5
    smirclesmircle Posts: 1,035member
    Quote:

    Originally posted by Splinemodel

    The G3 is much better designed, as far as I know, than the G4.



    No, this is an urban legend that seems to spread uncontrollably. The 750 scales even worse than the 74xx, it has no multiproc capabilities, a worse bus interface and no SMD unit.



    Quote:



    The 750GX reaches 1.1Ghz on a 4 stage pipeline architecture. That's pretty impressive.




    It does not reach 1.1Ghz. The GX is not available yet and, as the last iteration of the 750 (slated to reach 1Ghz in spring) has shown, IBM has huge problems getting the clock rate up. 1.1Ghz by end of year means it is almost a year late.



    Quote:



    But damn. I want a powerbook with a 1.6Ghz G3+altivec variant.




    At the rate the 750x is scaling, that will be a year and a half away. Besides, the SMD unit will up peak power consumption quite noticebly. And, to feed a SMD unit, you need to redo the bus interface, else you are going to starve the chip even worse than the current G4 is bandwidth-starved. Faster bus interface means faster ASICS means more power consumed.
     0Likes 0Dislikes 0Informatives
Sign In or Register to comment.