New 970 FX details from macrumors...

124»

Comments

  • Reply 61 of 79
    costiquecostique Posts: 1,084member
    Quote:

    Originally posted by snoopy

    One thing it means is that I was surprised to see theory work out so close to published values...



    I guess that's because IBMers use the same math as you to calculate the published numbers.
  • Reply 62 of 79
    amorphamorph Posts: 7,112member
    Quote:

    Originally posted by snoopy

    For voltage it is a squared relationship. Double the voltage and the power goes up four times its value. Cut the voltage in half and the power gets cut to one fourth its value.



    Given that, ponder the fact that the last generation G4s were run at 1.85v.
  • Reply 63 of 79
    smirclesmircle Posts: 1,035member
    Quote:

    Originally posted by snoopy

    Double the voltage and the power goes up four times its value. Cut the voltage in half and the power gets cut to one fourth its value.



    Only you cannot just reduce the voltage at will. Transistors have a breakthrough threshold that needs to be broken for them to work. If you go below this threshold, they simply stop working.



    I am no expert in baking CPUs, but I guess they construct the things in a way that the normal operating voltage is not too much above the threshold.



    So, you might have a chip that works both with 0.85 and 1 V, but it is unlikely you can just underpower the 1 V piece to 0.5 V.
  • Reply 64 of 79
    This story talks about the 65nm already being sampled.









    http://www.theregister.co.uk/content/54/35130.html
  • Reply 65 of 79
    screedscreed Posts: 1,077member
    Last line:

    "The 980 is expected to be fabbed at 90nm, however."



    Screed
  • Reply 66 of 79
    snoopysnoopy Posts: 1,901member
    Quote:

    Originally posted by Smircle

    Only you cannot just reduce the voltage at will. . .

    So, you might have a chip that works both with 0.85 and 1 V, but it is unlikely you can just underpower the 1 V piece to 0.5 V.






    You are right of course, but I was explaining the linear and squared relationships, not proposing a real chip. I think it is easier to understand the principal by considering a doubling or cutting in half. At least it is quicker to explain it this way.
  • Reply 67 of 79
    snoopysnoopy Posts: 1,901member
    Quote:

    Originally posted by sCreeD

    Last line:

    "The 980 is expected to be fabbed at 90nm, however."



    Screed




    That must mean we will get the 980, or whatever IBM calls it, much earlier than waiting for the 65 nm line to be up and running. Maybe a 65 nm 980FX six months later? Like the 970 and 970FX?



    Who knows, the Xbox and PowerMac may be using the same CPU?
  • Reply 68 of 79
    Quote:

    Originally posted by snoopy

    Who knows, the Xbox and PowerMac may be using the same CPU?



    Well, I doubt they'll actually use the same CPU. They'll probably be based on the same technology, and will likely be mostly (completely?) compatible, but I think the XBox2 CPU will be optimized exclusively for gaming. That's what IBM did for Nintendo's GameCube CPU, isn't it?
  • Reply 69 of 79
    snoopysnoopy Posts: 1,901member
    Quote:

    Originally posted by TrevorD

    Well, I doubt they'll actually use the same CPU. They'll probably be based on the same technology, and will likely be mostly (completely?) compatible, but I think the XBox2 CPU will be optimized exclusively for gaming. That's what IBM did for Nintendo's GameCube CPU, isn't it?







    Just hoping I guess. The advantage would be very high production runs and lower cost for the CPU. Regarding Nintendo, I seem to remember they requested some additional functions and/or instructions. My memory is fuzzy on that one.
  • Reply 70 of 79
    IBM will talk about the new chip in february. I suppose that it will be the 2nd feb because everyone waits for an announcement.



    IBM will specially talk about the PowerManager introduced in the 970fx. That will help it too and think that a Powerbook G5 is the next update for the Powerbook.



    I don't know what to say to a friend of mine that wants to buy one.
  • Reply 71 of 79
    Quote:

    Originally posted by snoopy

    Just hoping I guess. The advantage would be very high production runs and lower cost for the CPU. Regarding Nintendo, I seem to remember they requested some additional functions and/or instructions. My memory is fuzzy on that one.



    The Gekko processor is a ppc750 with a subset of the altivec (Ibm calls it vmx I think) instruction set.
  • Reply 72 of 79
    henriokhenriok Posts: 537member
    Quote:

    Originally posted by cuneglasus

    The Gekko processor is a ppc750 with a subset of the altivec (Ibm calls it vmx I think) instruction set.



    No, it's based on PPC 405 and it doesn't use VMX (witch _is_ AltiVec) but some other SIMD-unit.
  • Reply 73 of 79
    outsideroutsider Posts: 6,008member
    Quote:

    Originally posted by Henriok

    No, it's based on PPC 405 and it doesn't use VMX (witch _is_ AltiVec) but some other SIMD-unit.



    Old article. The processor is 485MHz and it IS based on the 750CXe.



    http://cube.ign.com/articles/100/100...html?fromint=1
  • Reply 74 of 79
    thttht Posts: 4,640member
    The Gekko is a PPC 750 based processor with basically one change. The existing 64-bit precision FPU unit was enhanced to do 2 32-bit precison ops at the same time. Not much of change. So there were only about 2 or 3 SIMD instructions added. You really can't say a subset of the AltiVec/VMX instruction set was added. Maybe a sub-subset. The SIMD instructions probably aren't really related at all.



    The other unique things about the Gekko are obviously the specialized graphics unit and lower-latency memory system using 1T-SDRAM (or something along those lines).
  • Reply 75 of 79
    Quote:

    Originally posted by Outsider

    Old article. The processor is 485MHz and it IS based on the 750CXe.



    http://cube.ign.com/articles/100/100...html?fromint=1




    OK, In that article, and part 2 of it, there's no mentioning of what processor Gekko "relly" is but they do mention that it's not just a 750, so it seems to be some kind of 750. IBM should update their website.



    I'd say, based on those two articles that Gekko is based on 750 plain and simple. I see no evidence of 750CXe.



    And.. doing a Goggle-search on the topic i find nothing other than anectodal evidence supporting the claim that Gekko is indeed derived from 750CXe. The only official word is still that Gekko is derived from 405, even if that article is some months older than article from IGN.



    The fact that Gekko and 750CX/CXe happened to arrive at the same time proves nothing. The articles above clearly describes an ongoing deveopment of Gekko where they've might have been forced to lauch it early. 750CX/CXe is a small evolution from the original 750. On die cache, faster bus, newer fab. I see no reason why Gekko should be based on 750CXe. It seems pretty odd if a specialized chip like Gekko would be launched before the general CPU that it was derived from.



    On the other hand.. It seems logical that a processor destined for game console use would be based on a high end general CPU rather than a low end embedded type.
  • Reply 76 of 79
    Okay, enough of this random speculation...



    The Gekko is a 750-derived processor with a 256KB on-die L2 cache, running at ~485 MHz. It has about a dozen instructions which operate on "paired single" floating point numbers, and this has zero, I repeat, zero relationship with VMX. It also has a couple of data packing / unpacking modes and a streaming writeback buffer, both of which help it communicate with the graphics chip in the GameCube. This chip was developed specifically for Nintendo and that was justified because the changes from the 750 are relatively minor and they knew they would selling many millions of these chips.



    It is not derived from the 405. It has no VMX unit (or even a sub-subset of VMX... it operates on the FPU registers). It has an on-die L2 cache, but that was likely specifically added to this chip's layout and does not imply a relationship with later 750-derived designs.
  • Reply 77 of 79
    haraldharald Posts: 2,152member
    No, Programmer you're just wrong.



    Again.



    You really don't know anything do you?



    joke
  • Reply 78 of 79
    programmerprogrammer Posts: 3,423member
    Quote:

    Originally posted by Harald

    No, Programmer you're just wrong.



    Again.



    You really don't know anything do you?



    joke




    Yeah, you're right... I should just keep my mouth shut.
  • Reply 79 of 79
    krassykrassy Posts: 595member
    Quote:

    Originally posted by sCreeD

    Last line:

    "The 980 is expected to be fabbed at 90nm, however."



    Screed




    yeah - that'll mean a PowerBook G5 in fall with 65nm cpus and a PowerMac G6 (980) in fall with 90nm cpus
Sign In or Register to comment.