Please Help Put the 970 in Context

Posted:
in Future Apple Hardware edited January 2014
Dual 2.3 GHz processors (that's 4.6 GHz of PowerPC to kick Intel's clock-frequency-spewing ass) would be amazing, but it would also be interesting to see what sort of overall design Apple puts these chips in. Will it be conservative, or will the system architecture and componentry blow us away too? There are some substantial posts on technical issues at the Ars OpenForum, though they're geared to people with a more technical knowledge of hardware. I would appreciate it if Programmer or anyone else could try to distill some of those types of discussions for us, such as the merits of different bus topologies, what kind of RAM ought to be used, etc.
«13

Comments

  • Reply 1 of 56
    rhumgodrhumgod Posts: 1,289member
    No offense but isn't there enough 970-laced topics already that this could go in?
  • Reply 2 of 56
    Why didnt you use one of the other 20.000 970 threads?



    In before lock!



    [Edit:] DAMN! I was to slow!
  • Reply 3 of 56
    Forgot about "970 Motherboards to enter production soon (MacBidouille)"... still, I posit that it is worth having a thread around dedicated purely to exploring the reasons behind design decisions, rather than just the current 970 mishmash threads of fantasy machine specs, unverifiable rumours and occasional rationality.
  • Reply 4 of 56
    hmurchisonhmurchison Posts: 12,423member
    Ok maybe we should outline what functions the 970 should have compared to a G4.





    Then we can discuss the differences, merits or whatever for each section.



    Someone good at creating outlines..get busy.



    This will be the definitive "known info" thread. Light on speculation Heavy on verfiable facts.



    Discuss.
  • Reply 5 of 56
    bodhibodhi Posts: 1,424member
    This is soooo close to being locked but it looks as if it may go in a good direction.



    But in the future please use one of the current 970 threads instead of creating ANOTHER thread about the 970.







  • Reply 6 of 56
    programmerprogrammer Posts: 3,458member
    We really don't know enough yet to make predictions about bus topology or what kind of system level hardware Apple will throw at these machines. We have some preliminary estimated benchmarks from IBM, given before they had silicon to test with and at the announced 1.8 GHz level. There are very few facts that we can state with certainty.



    The performance of these machines is going to be really good compared to the current G4 machines, there is no doubt about that. In general code that has not been optimized specifically for the Mac (i.e. just ported from the PC) will perform substantially better than on the G4. G4 optimized code should run noticably better on a clock-for-clock basis as well, although there may be exceptions. The increase in memory bandwidth should see a substantial jump in AltiVec performance since the majority of AltiVec code out there is bandwidth bound. Scalar floating point code should run roughly twice as fast as on the G4 at the same clock rate (this is both single and double precision).



    Between the bandwidth, clockrate, and internal processor improvements the machines should be cleanly twice as fast... but this will vary widely depending on what you are using to measure performance. In some cases it might be much more than twice as fast, and I'm sure there will be pathalogical cases here and there where people will manage to find something that actually runs slower (it always happens when a new CPU comes along). As software gets recompiled with the optimizer 970 turned on the speed difference should increase.





    Aside from these vague statements there really isn't much that we can say with authority right now.
  • Reply 7 of 56
    fran441fran441 Posts: 3,715member
    Bodhi, if I hadn't seen you had already posted, I would have closed this thread as well. There are definitely more than enough 970 threads to go around.
  • Reply 8 of 56
    bigcbigc Posts: 1,224member
    How do you put something in "context" that doesn't exist.
  • Reply 9 of 56
    Even if we can't speculate on precisely what Apple is going to do with regard to topologies, I would certainly like to know what the benefits and drawbacks of these systems are. What I gather from the Ars thread is that a star topology would allow the throughput of each processor's bus to be used simultaneously, whereas a token ring approach would introduce latency (because the token gets passed around the ring, everybody needs to wait their turn) and also limit the overall throughput of the system to that of one processor's bus, with the advantage of not having to create companion chips with different pin counts. First, this is obviously a basic description, but is it at all accurate? If so, and if we assume that Apple is only interested in single and dual configurations, what is the benefit of a ring topology?
  • Reply 10 of 56
    Quote:

    Originally posted by Bodhi

    This is soooo close to being locked but it looks as if it may go in a good direction.



    But in the future please use one of the current 970 threads instead of creating ANOTHER thread about the 970.











    Told you so, told you so! (<--Gez, I want that old smiley back..)









    In before lock, TWICE! And this time, WITH warning!
  • Reply 11 of 56
    Quote:

    Originally posted by Bigc

    How do you put something in "context" that doesn't exist.



    You look at the few facts there is, shuffle it together with the rumors, add some water, 20 minutes in a hot oven, and WOAW!



    "Proof" of a new chip!
  • Reply 12 of 56
    thttht Posts: 5,433member
    Well, Apple will use the technology corresponding to the prices and margin they want to sell it.



    Which means at the current price points, they will use PC2700 and PC3200 DDR SDRAM (DDR333 and DDR400). They will support up to 4 GB of memory. Everything else will stay the same but for typical evolutionary improvements: possibly 1 64 bit 66 MHz PCI slot, 3 64 bit 33 MHz PCI slots, AGP 4x/8x, Firewire 800, SATA, 128 MB standard graphics card, etc. A HyperTransport southbridge may be there - Apple isn't in the HT consortium for nothing. The PPC 970 looks to have BGA packaging, so an Apple machine will likely have a new socket and daughtercard. I believe a new case design is in the works, and will have 2 5.25" external bays and 4 3.5" internal bays like there is now. It would be slick if it had xServe/xRaid design language, but I doubt it.



    What really confounds me is how the system ASIC will bridge the point-to-point source-synchronous 970 bus with main memory, AGP bus, et al. It has a DDR processor bus operating at half the clock of the CPU. If this info doesn't change, Apple is going to have a tough time maximizing clock rate grades. [They should want to sell boxes from 1.2, 1.4, 1.6, 1.8, 2.0, up to 2.2 GHz or whatever, and not just jump up in CPU grades.] If there is a memory bus to processor bus multiplier, for a hypothetical 167 MHz memory bus, you'll see 1.33, 1.67 and 2.0 GHz machines. This is a huge waste in business terms.



    The system ASIC will have to multiplex and demultiplex 2x32 bit data and addresses from the 970 bus to a 2x64 bit memory bus with separate lines for data and addresses. That's Intel 850 complexity. This isn't going to be a simple ASIC to do. Which means, no PCI Express, or any other up and coming tech.



    As far as what they ought to use, well, I'm a Rambus fan. Yellowstone, quad-channel RDRAM, Redwood, lots of bandwidth there. My 4+ channel Rambus backside cache dream is still alive.
  • Reply 13 of 56
    shawkshawk Posts: 116member
    Wasn't there a description and a picture from IBM Germany for a 970 mobo/blade that sounded/looked a lot like the motherboard that Mac Whispers described going out for bid?



    Wasn't that IBM information hurridly pulled? Does anybody have a pointer to the cached information?



    Not that I'm suggesting the obvious but...
  • Reply 14 of 56
    That was an odd hash of components. It had fast Ethernet, PC-133 SDRAM, some Xilinx FPGA which some thought to be serving as the 'companion chip', and a nice big heat sink.
  • Reply 15 of 56
    programmerprogrammer Posts: 3,458member
    If the FSB is a ring then the output of the chipset goes to the proc1's input, the output of proc1 goes to proc2's input, and the output of proc2 goes to the chipset's input. All data goes all the way around the ring and is thus transmitted (in turn) on 3 output buses and 3 input buses. This means the total bandwidth available is 3.2 GB/sec for all processors. Star topologies means each processor gets 3.2 GB/sec in each direction.
  • Reply 16 of 56
    Quote:

    Originally posted by Programmer

    We really don't know enough yet to make predictions about bus topology or what kind of system level hardware Apple will throw at these machines. ...snip... There are very few facts that we can state with certainty.



    Very true, but when has that ever stopped anyone?



    One thing we do know, if not for a fact, at least for a near certainty, is that this chip has been in development for at least two years. And given the altivec unit, Apple has known about it for two years.



    This means that Apple has had two years to think about what technologies they want to build into the systems that will be powered by this chip. And given Apple's propensity for innovation, I suspect this means the overall architecture will be appropriately robust. Add that to the rumors that the motherboard's design was only finalized this month, then throw in Apple's recent forays into the enterprise market and you might be forgiven for thinking that the rest of the system could be as radical and new as the PPC970 itself.
  • Reply 17 of 56
    rhumgodrhumgod Posts: 1,289member
    Quote:

    Originally posted by shawk

    Does anybody have a pointer to the cached information?



    You mean this one? It is scaled down a bit from the 2MB that it was on the IBM site. I have the original and can add it to my website if anyone cares to see it in it's burlap-sack-bed virginity.
  • Reply 18 of 56
    netromacnetromac Posts: 863member
    Quote:

    Originally posted by T'hain Esh Kelch

    You look at the few facts there is, shuffle it together with the rumors, add some water, 20 minutes in a hot oven, and WOAW!



    "Proof" of a new chip!




    Which temp??? My "proof" always seems a little unfinished
  • Reply 19 of 56
    kroehlkroehl Posts: 164member
    Quote:

    Originally posted by NETROMac

    Which temp??? My "proof" always seems a little unfinished



    Put it on grill for a few minutes and you get a nice crispy crust on your proof. Don't overdo it or your proof will be hard to swallow!



    The proof is in the pudding anyway.



    Kroehl
  • Reply 20 of 56
    placeboplacebo Posts: 5,767member
    The Powermac 970 is kinda like a black hole: you don't know that it exists, but you see its effects. Read: (MoBo, Panther, etc.) We don't know what it looks like but...it's there. As for the additional PM 970 components, I thinks that Apple will not hold back on this release.
Sign In or Register to comment.