Apple modular Mac Pro launch coming in 2019, new engineering group formed to guarantee fut...

18910111214»

Comments

  • Reply 261 of 269
    cgWerkscgWerks Posts: 1,605member
    Morphroth said:
    Are we to believe that the iMac Pro was designed and built over a couple of weekends, using high-end parts thrown together, randomly and without even a hint of optimisation… and built, and sold?
    No, I think the iMac Pro was intended to be the end of the line Mac to appease video editors and developer types for the next few years... in other words, they didn't intend to make another Mac Pro.

    ZestyMordant said:
    What's stopping a future iteration of Thunderbolt from using PCIe 4 or 5?
    True, but the question is if it will always lag behind, just due to technology and physics limitations (or budget). I'm not sure why we can't have the optical version of TB right now that Intel originally sold the concept on. Is there some physical thing stopping that? Would it just cost too much? Or, are they just rolling it out slowly, version by version, to maximize the money they are making on it?

    Or, is it a problem like desktop/mobile performance, where the mobile keeps getting faster, but will never catch the desktop because it does too, and doesn't have the power-limitations.

    nht said:
    No, each port is typically on a TB switch so if you grab two ports on the same switch you're still bandwidth limited.  Bonding two TB switches isn't doable today.
    But TB4 or beyond could or it may get replaced by some other interconnect standard that's just an external 16x PCIe 5 connection with expensive and short cables...
    I meant that in regards to the mentioned "128GBps external pci expansion chassis"... that isn't possible for the Mac, unless you could bond multiple TB ports. You're just meaning it's being done in other industries, right?
  • Reply 262 of 269
    bkkcanuckbkkcanuck Posts: 811member
    cgWerks said:

    I meant that in regards to the mentioned "128GBps external pci expansion chassis"... that isn't possible for the Mac, unless you could bond multiple TB ports. You're just meaning it's being done in other industries, right?

    PCI Express 3.0's 8 GT/s bit rate effectively delivers 1 GB/s per lane.

    PCI Express 4.0's 16 GT/s bit rate effectively delivers 2 GB/s per lane.

    PCI Express 5.0's 32 GT/s bit rate effectively delivers 4 GB/s per lane.

    The maximum that your average consumer grade CPU's expansion capabilities are 24 PCIe lanes (I think 4 goes to the chipset; so 20 net).  Your average Xeon workstation chip gives you 44 lanes PCIe lanes - for dual CPU Xeon it effectively doubles as it is per CPU.  The benefits of Xeon over consumer CPUs is more bandwidth throughput and ECC memory (among others).  Thunderbolt only expansion effectively eliminates the first benefit.

    Thunderbolt uses either 2 or 4 lanes depending on the implementation.  

    So the entire bandwidth for a consumer grade CPU can handle currently is theoretically 20GB/s.  A Xeon workstation chip would be 40GB/s.

    The maximum that 4 lanes to the CPU would have would be 4GB/s or 40Gbps.  

    ---

    PCI Express 4.0 will bring with it OCuLink version 2 (an alternative to Thunderbolt) will have up to 16 GT/s (8 GB/s total for × 4 lanes).

    Graphics card (if you have one) typically slots into an x16 slot -- if you put them in x4 slot you would be bottlenecking the graphics card (assuming it ran).

    ---

    For high bandwidth expansion -- PCIe for the foreseeable future provides the best bandwidth and cheaper cost of expansion.

    cgWerks
  • Reply 263 of 269
    ascii said:
    Yep, I suspect a lot of people will just skip 4.0. I also wonder why we were stuck on 3.0 so long.
    Thanks to the Great GPU Shortage of 2017-18, I wonder if anything will even be made to use PCIe 4. As slowly as ATI and nVidia come out with products these days, who says they won’t just skip it, too?

    And speaking of 3, why are we still on SATA 3? Did they look at M.2 and say, “Well, we better not even try”? SATA 3.2 has 16 Gb/s, but… Oh, I see that M.2 is a derivative of the SATA standard. Okay; guess they’re just moving to that entirely. Now to wait another decade for products that use it to cost less than a kidney.
    M.2 SSDs are pretty much the same price as the SATA 3 ones; NVMe SSDs are a bit more expensive though.
  • Reply 264 of 269
    cgWerks said:

    True, but the question is if it will always lag behind, just due to technology and physics limitations (or budget). I'm not sure why we can't have the optical version of TB right now that Intel originally sold the concept on. Is there some physical thing stopping that? Would it just cost too much? Or, are they just rolling it out slowly, version by version, to maximize the money they are making on it?
    Definitely cost.  Never mind the silicon and the connectors, you're looking at a couple of hundred bucks just for a 5 m cable.
  • Reply 265 of 269
    cgWerkscgWerks Posts: 1,605member
    ZestyMordant said:
    Definitely cost.  Never mind the silicon and the connectors, you're looking at a couple of hundred bucks just for a 5 m cable.
    Is that just because it's so specialized, currently? I remember when home audio optical stuff was pretty expensive too. I know optical cables are more difficult to manufacture, but it doesn't seem like it should be that much more expensive. I suppose the conversion of that much data needs a robust process/hardware.
  • Reply 266 of 269
    cgWerks said:
    ZestyMordant said:
    Definitely cost.  Never mind the silicon and the connectors, you're looking at a couple of hundred bucks just for a 5 m cable.
    Is that just because it's so specialized, currently? I remember when home audio optical stuff was pretty expensive too. I know optical cables are more difficult to manufacture, but it doesn't seem like it should be that much more expensive. I suppose the conversion of that much data needs a robust process/hardware.
    Is the high cost of optical Thunderbolt cables related to having the electrical/optical transformer hardware in the connectors and using actual glass fibre? TOSLink/ADAT optical cables don't perform any conversion because the light source and receiver are built into the devices they're connecting, and the ones I have use plastic fibre rather than glass (though I suppose they may be able to get away with plastic because the runs are short -- under ten feet -- and maybe glass would be required for longer runs).

    I wonder if the net cost would be lower if the light emitting and receiving hardware were built into the computer and peripherals they way they are with audio equipment? It would increase the cost of the Mac somewhat, but maybe the economy of scale would make it more palatable?

    If optical does become more common, how do we handle powering peripherals like we do with electrical Thunderbolt cables? Optical doesn't carry power, so do we accept that peripherals will require their own power source, or do we create hybrid cables with optical fibre for data and electrical conductors for power?
  • Reply 267 of 269
    cgWerkscgWerks Posts: 1,605member
    lorin schultz said:
    Is the high cost of optical Thunderbolt cables related to having the electrical/optical transformer hardware in the connectors and using actual glass fibre? TOSLink/ADAT optical cables don't perform any conversion because the light source and receiver are built into the devices they're connecting, and the ones I have use plastic fibre rather than glass (though I suppose they may be able to get away with plastic because the runs are short -- under ten feet -- and maybe glass would be required for longer runs).

    I wonder if the net cost would be lower if the light emitting and receiving hardware were built into the computer and peripherals they way they are with audio equipment? It would increase the cost of the Mac somewhat, but maybe the economy of scale would make it more palatable?

    If optical does become more common, how do we handle powering peripherals like we do with electrical Thunderbolt cables? Optical doesn't carry power, so do we accept that peripherals will require their own power source, or do we create hybrid cables with optical fibre for data and electrical conductors for power?
    All good questions.
    I was primarily trying to figure out why optical (in general) would cost more than a non-optical transmission. The only expensive aspect I'm aware of (on scale) is the actual manufacture of the fibre ends, especially out in the field. But, in this situation, they'd be all factory made. But, I'm also not up on this stuff as much as I used to be.

    My hunch is that it is more due to the small quantities and uniqueness of it currently in the market.
  • Reply 268 of 269
    crowleycrowley Posts: 5,522member
    cgWerks said:
    ZestyMordant said:
    Definitely cost.  Never mind the silicon and the connectors, you're looking at a couple of hundred bucks just for a 5 m cable.
    Is that just because it's so specialized, currently? I remember when home audio optical stuff was pretty expensive too. I know optical cables are more difficult to manufacture, but it doesn't seem like it should be that much more expensive. I suppose the conversion of that much data needs a robust process/hardware.
    Is the high cost of optical Thunderbolt cables related to having the electrical/optical transformer hardware in the connectors and using actual glass fibre? TOSLink/ADAT optical cables don't perform any conversion because the light source and receiver are built into the devices they're connecting, and the ones I have use plastic fibre rather than glass (though I suppose they may be able to get away with plastic because the runs are short -- under ten feet -- and maybe glass would be required for longer runs).

    I wonder if the net cost would be lower if the light emitting and receiving hardware were built into the computer and peripherals they way they are with audio equipment? It would increase the cost of the Mac somewhat, but maybe the economy of scale would make it more palatable?

    If optical does become more common, how do we handle powering peripherals like we do with electrical Thunderbolt cables? Optical doesn't carry power, so do we accept that peripherals will require their own power source, or do we create hybrid cables with optical fibre for data and electrical conductors for power?
    Wouldn't part of the problem for Thunderbolt be that if the light-receiving components were in the computer then you'd lose the ability to interchange between optical and copper cables?  And also you'd then lose cross compatibility with USB 3.0, not to mention you'd need a different connector than USB-C?

    Sounds like there's quite a few downsides. 

    If the transformer hardware is a worry, I wonder if there might be an option to modularise the cable, i.e. have the transformer at each end be effectively a dongle rather than integrated into the cable.  Then if the cable breaks, you only have to change the cheaper bit, not the expensive bit.  Hell, if you did that then maybe the middle section could even be a standard TOSLink cable (probably not, I imagine Thunderbolt has greater requirements)?  Though the modularisation would in itself add to the cost, not to mention be much more fiddly for users (albeit pro users who wouldn't care as much). 

    Tricky situation, and there are competing priorities between consumer and professionals, re. convenience and cost.
  • Reply 269 of 269
    crowley said:
    cgWerks said:
    ZestyMordant said:
    Definitely cost.  Never mind the silicon and the connectors, you're looking at a couple of hundred bucks just for a 5 m cable.
    Is that just because it's so specialized, currently? I remember when home audio optical stuff was pretty expensive too. I know optical cables are more difficult to manufacture, but it doesn't seem like it should be that much more expensive. I suppose the conversion of that much data needs a robust process/hardware.
    Is the high cost of optical Thunderbolt cables related to having the electrical/optical transformer hardware in the connectors and using actual glass fibre? TOSLink/ADAT optical cables don't perform any conversion because the light source and receiver are built into the devices they're connecting, and the ones I have use plastic fibre rather than glass (though I suppose they may be able to get away with plastic because the runs are short -- under ten feet -- and maybe glass would be required for longer runs).

    I wonder if the net cost would be lower if the light emitting and receiving hardware were built into the computer and peripherals they way they are with audio equipment? It would increase the cost of the Mac somewhat, but maybe the economy of scale would make it more palatable?

    If optical does become more common, how do we handle powering peripherals like we do with electrical Thunderbolt cables? Optical doesn't carry power, so do we accept that peripherals will require their own power source, or do we create hybrid cables with optical fibre for data and electrical conductors for power?
    Wouldn't part of the problem for Thunderbolt be that if the light-receiving components were in the computer then you'd lose the ability to interchange between optical and copper cables?  And also you'd then lose cross compatibility with USB 3.0, not to mention you'd need a different connector than USB-C?

    Sounds like there's quite a few downsides. 

    If the transformer hardware is a worry, I wonder if there might be an option to modularise the cable, i.e. have the transformer at each end be effectively a dongle rather than integrated into the cable.  Then if the cable breaks, you only have to change the cheaper bit, not the expensive bit.  Hell, if you did that then maybe the middle section could even be a standard TOSLink cable (probably not, I imagine Thunderbolt has greater requirements)?  Though the modularisation would in itself add to the cost, not to mention be much more fiddly for users (albeit pro users who wouldn't care as much). 

    Tricky situation, and there are competing priorities between consumer and professionals, re. convenience and cost.
    Excellent points. I hadn't thought about how optical might affect USB compatibility.

    Apple used to offer both optical and electrical connection at the headphone jack, so all that was required for optical was a special connector shaped like a traditional 3.5mm audio plug but with an optical opening on the end and no electrical contacts. Maybe something similar could be created using a USB-C plug? It looks like it might be a little crowded, but maybe there's a way.

    It seems like simple, inexpensive, optical Thunderbolt may not be imminent, with good reason. I guess anyone who truly needs it just has to be prepared to pay a hefty price.
Sign In or Register to comment.