Moved: G4 / Bus Speed?

Posted:
in Future Apple Hardware edited January 2014
OK, this is for the ultra informed of you out there. How likely is it that Moto has tweaked/modified the bus intereface on the 7470 (or whatever the new G4 we will be getting soon is) to run at 333MHz? I know the current chips won't benefit much cause the MPX/133 bus they use can't talk to the memory any faster.



Is this something that Moto could have overcome for the 7470, thus allowing it to talk to memory faster, or are we still looking at a fast chip, that can't fully tap into fast memory bus speed, aka DDR 266 or 333?



[ 08-13-2002: Message edited by: Jamie ]</p>

Comments

  • Reply 1 of 17
    telomartelomar Posts: 1,804member
    I'm not ultra informed but I wouldn't consider it highly likely. A FSB at 166 MHz seems quite probable but otherwise to get full benefit from DDR memory I expect they'd have to take the memory controller on-chip, which I doubt has been done yet.
  • Reply 2 of 17
    bartobarto Posts: 2,246member
    Ok, we know that the G4 now runs on a 166MHz bus.



    A 333MHz MPX bus would probably force a packaging redesign (ceramic -&gt; organic?), and even then I doubt more than 1% of chips could reach 333MHz.



    If you are talking about DDR, then it would mean a whole new bus. You can't just "tweak" something into DDR. Apple uses Discovery northbridges, so until a DDR Discovery is made, you wouldn't even be able to use it.



    If you are talking about 128-bit (for the equivalent of 333MHz at 64-bit), then you would see costs run up with Apple forced to produce more expensive motherboards, to accomodate for more "tracers" running away from the chip. Say goodbuy to sawtooth also, as you would need the socket to double in size.



    The only plausible way out for Motorola is to get rid of MPX, and replace it with more serialized, higher MHz busses like Rapid-I/O or HyperTransport, and/or an on-chip memory controller. Motorola seems to have little time left as an Apple customer, and for embedded uses the G5 HAS Rapid-I/O and DDR333, and is far more suited to tasks the G4 was used in (eg routers).



    I don't see Motorola doing anything more with the G4. It's in its death throws.



    Barto



    [ 08-11-2002: Message edited by: Barto ]</p>
  • Reply 3 of 17
    Discovery chipsets? I was under the impression Apple designed their own. All of the block diagrams show either a "Pangea" single chip and a UniNorth/KeyLargo chipset.



    [ 08-11-2002: Message edited by: Analogue bubblebath ]</p>
  • Reply 4 of 17
    bartobarto Posts: 2,246member
    Uninorth = Apple's name for discovery.



    I think (guess) the original UniNorths were Apple designs, and switched to Discovery with the Gigabit PowerMacs (onboard gigabit ethernet is part of the Discovery's specifications, but the earlier powermacs had intel 10/100 controllers).



    If you pry off the northbridge heatsink off a new Power Mac, a Discovery part number is visible.



    Barto
  • Reply 5 of 17
    henriokhenriok Posts: 537member
    Motrola have said that they don't have any plans for DDR on the MPX, but they might do a little speed bump. We know that this is done considering last weeks deicovery of G4-processors @ 1167 MHz.



    166/333 DDR-RAM is up and running in the 8560 embedded processor so we know that Motorola is capable of implementing it.. but as far as the G4 design goes, faster than regular 166 MHz is highly unlikely. If they do make a CPU with DDR on die, then that processor will surely be dubbed G5.
  • Reply 6 of 17
    costiquecostique Posts: 1,084member
    [quote]Originally posted by Barto:

    [QB]Ok, we know that the G4 now runs on a 166MHz bus.



    A 333MHz MPX bus would probably force a packaging redesign (ceramic -&gt; organic?), and even then I doubt more than 1% of chips could reach 333MHz.

    <hr></blockquote>



    Do you mean no double-pumped 166MHz in new PowerMacs?
  • Reply 7 of 17
    [quote]Originally posted by Barto:

    <strong>Uninorth = Apple's name for discovery.



    I think (guess) the original UniNorths were Apple designs, and switched to Discovery with the Gigabit PowerMacs (onboard gigabit ethernet is part of the Discovery's specifications, but the earlier powermacs had intel 10/100 controllers).



    If you pry off the northbridge heatsink off a new Power Mac, a Discovery part number is visible.



    Barto</strong><hr></blockquote>



    Interesting, I went to marvell's website and none of the Discovery northbridges had any mention of the AGP controller.
  • Reply 8 of 17
    bartobarto Posts: 2,246member
    I would be extremely surprised if Motorola introduced ANY new bus for the G4.



    The 166MHz MPX has been announced, maybe we will see a 200MHz in another 18 months, at the end of the G4's life (assuming it survives another 18 months).



    Barto
  • Reply 9 of 17
    bartobarto Posts: 2,246member
    [quote]Originally posted by Analogue bubblebath:

    <strong>



    Interesting, I went to marvell's website and none of the Discovery northbridges had any mention of the AGP controller.</strong><hr></blockquote>



    I know. I read somewhere that Apple uses PCI -&gt; AGP conversion. Judging by the fact that the Discovery has no AGP bus, yet Apple uses a chip that looks, walks and talks like a Discovery, I must assume this to be true.



    This is possible as the Discovery has DMA PCI, and primarily AGP is PCI + DMA. 33MHz 64-bit is converted to 66MHz 32-bit AGP 4X.



    66MHz 64-bit will (presumably) be converted to 133MHz 32-bit AGP 8X for the next Power Macs.



    Barto



    [ 08-12-2002: Message edited by: Barto ]</p>
  • Reply 10 of 17
    big macbig mac Posts: 480member
    [quote]Originally posted by Barto:

    <strong>I would be extremely surprised if Motorola introduced ANY new bus for the G4.



    The 166MHz MPX has been announced, maybe we will see a 200MHz in another 18 months, at the end of the G4's life (assuming it survives another 18 months).



    Barto</strong><hr></blockquote>



    This is the first time I've heard about 166MHz MPX. If so that's a major indication of what we're getting this month. This is nothing great, but I'm kind of surprised Motorola would give us any bus improvements whatsoever.
  • Reply 12 of 17
    [quote]Originally posted by Barto:

    <strong>



    I know. I read somewhere that Apple uses PCI -&gt; AGP conversion. Judging by the fact that the Discovery has no AGP bus, yet Apple uses a chip that looks, walks and talks like a Discovery, I must assume this to be true.</strong><hr></blockquote>



    <a href="http://developer.apple.com/techpubs/hardware/Developer_Notes/Macintosh_CPUs-G4/PowerMacG4/2Architecture/Accelerated_cs_Port_Bus.html"; target="_blank">From Apple's PowerMac G4 developers section:</a>



    "The accelerated graphics port (AGP) bus is a 128-MHz, 32-bit bus connecting the AGP graphics card to the Uni-N IC. The AGP bus provides faster access to main memory than previous designs using the PCI bus. The bus is an AGP-4x bus with twice the performance of the AGP-2x bus, supporting peak transfers of 512 MB/s.



    The AGP bus is a superset of the PCI bus, with the addition of separate address lines so it does not multiplex address and data when running in AGP mode. Having a separate address bus allows the AGP bus to pipeline addresses, thereby improving performance."



    Doesn't sound like PCI-to-AGP conversion to me. Maybe you're thinking of the Xserve? (No Discovery chipset here either, as the mem. controller is DDR.)



    Also, UniNorth includes the FireWire controller , unlike the Discovery. And the Discovery chipset supports up to 4GB of SDRAM, while UniNorth supports up to 1.5GB. Those seem like pretty major differences, don't you think?



    [ 08-12-2002: Message edited by: Analogue bubblebath ]</p>
  • Reply 13 of 17
    bartobarto Posts: 2,246member
    "The accelerated graphics port (AGP) bus is a 128-MHz, 32-bit bus connecting the AGP graphics card to the Uni-N IC. The AGP bus provides faster access to main memory than previous designs using the PCI bus. The bus is an AGP-4x bus with twice the performance of the AGP-2x bus, supporting peak transfers of 512 MB/s."



    That doesn't say that there is no bridge chip. In I/O Reg Explorer, it shows that my system has 3 PCI channels. A PCI bridge (part of Discovery, going to the 3 PCI slots), a channel with a Gigabit Ethernet controller (part of Discovery), and (gasp!) a channel with 2 devices. One called "[email protected]", and the graphics card. Smells like an AGP converter to me.



    "Also, UniNorth includes the FireWire controller , unlike the Discovery."



    Sort of. It provides a connection for a FireWire controller, and even that may be non-specific.



    "And the Discovery chipset supports up to 4GB of SDRAM"



    4 GBs of SDRAM, using 512MB SDRAM sticks. If a G4 has 4 slots (like mine), 2GB is acheivable. If a G4 has 3 slots (2001 onwards), only 1.5GB is acheivable. This does not rule out the Discovery. Presumably if the G4 had 8 slots, 4GB would be achievable.



    It is very confusing, and I, like you, would like to know for sure what is going on.



    The fact that the chip has a Discovery part number printed on it though... Anyone here have photos of the northbridge with the heatsink pryed off?



    Barto
  • Reply 14 of 17
    [quote]Originally posted by Barto:

    <strong>

    That doesn't say that there is no bridge chip. In I/O Reg Explorer, it shows that my system has 3 PCI channels. A PCI bridge (part of Discovery, going to the 3 PCI slots), a channel with a Gigabit Ethernet controller (part of Discovery), and (gasp!) a channel with 2 devices. One called "[email protected]", and the graphics card. Smells like an AGP converter to me.</strong><hr></blockquote>



    The <a href="http://developer.apple.com/techpubs/hardware/Developer_Notes/Macintosh_CPUs-G4/PowerMacG4/2Architecture/Block_Diagram_and_Buses.html"; target="_blank">block diagram</a> doesn't show any bridge chip . And that applies to all AGP-equipped PowerMacs, 'cause I've seen them all(the block diagrams).



    [quote]<strong>Sort of. It provides a connection for a FireWire controller, and even that may be non-specific.</strong><hr></blockquote>



    <a href="http://developer.apple.com/techpubs/hardware/Developer_Notes/Macintosh_CPUs-G4/PowerMacG4/2Architecture/FireWire_Controller.html"; target="_blank">That's not what it says here.</a>



    [quote]<strong>It is very confusing, and I, like you, would like to know for sure what is going on.



    The fact that the chip has a Discovery part number printed on it though... Anyone here have photos of the northbridge with the heatsink pryed off?</strong><hr></blockquote>



    Maybe this is some sort of custom-designed Discovery northbridge just for Apple? Kinda like the iPod processor? I'm telling you, there's no way the UniNorth is the same as any of the Discovery northbridges listed on Marvell's website.



    [ 08-12-2002: Message edited by: Analogue bubblebath ]</p>
  • Reply 15 of 17
    thttht Posts: 3,115member
    <strong>Originally posted by Barto:

    This is possible as the Discovery has DMA PCI, and primarily AGP is PCI + DMA. 33MHz 64-bit is converted to 66MHz 32-bit AGP 4X.</strong>



    AGP 2x (533 MByte/s) is 32 bit 66 MHz DDR which would be equivalent to 66 MHz 64 bit (533 Mbyte/s) PCI bus.



    AGP 4x is 32 bit 66 MHz QDR (with some sort of low voltage differential on 2 DDR signals I believe, note sure). 1066 MByte/s. This is twice as much as a PCI 2.x bus can deliver. There is pretty much no way a PCI 2.x bus can be converted to AGP 4x without making it feel like it is AGP 2x because PCI 2.x can only deliver 533 MByte/s.



    <strong>66MHz 64-bit will (presumably) be converted to 133MHz 32-bit AGP 8X for the next Power Macs.</strong>



    AGP 8x is 32 bit 66 MHz "octo" data rate. It's supposed to deliver 2066 MByte/s bandwidth while the latest and greatest PCI-X spec 64 bit 133 MHz only delivers 1066 MByte/s. Same sort of problem with AGP 4x and a PCI bus. The bandwidths don't make sense.



    So either Apple is doing some magic to one of those Discovery PCI buses to get it devliver 2x bandwidth than the PCI spec is designed for, Apple is lying about the specs, or Apple designed their own core logic. Perhaps they had Marvell heavily modify their core logic, but it simply isn't a plane jane Discovery core logic chipset.
  • Reply 16 of 17
    razzfazzrazzfazz Posts: 728member
    [quote]Originally posted by Barto:

    <strong>That doesn't say that there is no bridge chip. In I/O Reg Explorer, it shows that my system has 3 PCI channels. A PCI bridge (part of Discovery, going to the 3 PCI slots), a channel with a Gigabit Ethernet controller (part of Discovery), and (gasp!) a channel with 2 devices. One called "[email protected]", and the graphics card. Smells like an AGP converter to me.</strong><hr></blockquote>



    Not at all.

    AGP always appears to the system software as a second PCI bus bridged to the primary one. It's exactly the same for x86 machines, or any other architecture. That's because AGP was designed to be backwards-compatible (software-wise!) with PCI, so developers wouldn't have to start from scratch when writing drivers for their devices.



    Bye,

    RazzFazz



    [ 08-13-2002: Message edited by: RazzFazz ]</p>
  • Reply 17 of 17
    jambojambo Posts: 3,036member
    Moving to Current Hardware...
Sign In or Register to comment.