The thing is that the Skylake Xeon situation is still a mess, because the workstation suitable Skylake Xeons aren't coming out in an obvious timeframe meanwhile the consumer i7s are steaming ahead. I can see that if AMDs new Vega and Polaris chips are ready then the next Mac Pro will just be Broadwell chips. X299 being the chipset for the next three generations is nice, but is it worth delaying for another 6 months at this point? They've already messed up consumer relations enough.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
MP 7,1 Waiting for Mac Pro 7,1 A1991 (no more)
- Thread starter Mago
- Start date
-
- Tags
- macpro 2016
- Sort by reaction score
The thing is that the Skylake Xeon situation is still a mess, because the workstation suitable Skylake Xeons aren't coming out in an obvious timeframe meanwhile the consumer i7s are steaming ahead. I can see that if AMDs new Vega and Polaris chips are ready then the next Mac Pro will just be Broadwell chips. X299 being the chipset for the next three generations is nice, but is it worth delaying for another 6 months at this point? They've already messed up consumer relations enough.
At Apple its starting to seem that no matter what the answer to that question is a resounding YES!
I'm barely up to date on what the situation is, but it sounds like Intel's development has splintered off into many directions. For Skylake it's meant to Purley, which will be Sky to Kaby to Cannon Lake, but Broadwell is still Grantley (X99.)
Well, that's Intel's roadmap for you.
Moral of the story being that Apple can't afford to miss revisions now, because who knows when the next Intel microarchitecture will show up.
Thats a worry now that they have abandoned the tick tick pattern...
[doublepost=1478142135][/doublepost]Wasn't that the major reason they ditched the PowerPC? For the ablility to craft and produce at their own pace...
It was because the Intel chips were much better and at the time had a superior road map. The PowerPC wasn't living up to it's expectations.
I agree. I was waiting for my PowerBook G5
Aren't you still?I agree. I was waiting for my PowerBook G5
At the end we 're going to miss the Gx era, at least then we had some meaningful updates.
It's all wishful thinking that there's no new Mac Pro because Apple wants to wait for the latest chipset. Apple messed up enough indeed. Waiting three years or more for a refresh with no ability to update components is ridiculous. Apple could easily offer new video card updates for the 2013 nMP, if they wanted to.
Last edited:
The thing here is that X99 (or better C612) is not suitable for the nnMP.
too limited when it comes to spare lanes.
And X299 would in fact solve that problem, and SKL CPUs also is better equipped in that department.
Seems a perfect solution for the nnMP.
But it's not available in the near future and that's a problem.
Since Vega is still not out yet as well, all the waiting could be for all the stars to align. Or not.
It would however be a perfect machine.
too limited when it comes to spare lanes.
And X299 would in fact solve that problem, and SKL CPUs also is better equipped in that department.
Seems a perfect solution for the nnMP.
But it's not available in the near future and that's a problem.
Since Vega is still not out yet as well, all the waiting could be for all the stars to align. Or not.
It would however be a perfect machine.
Aren't you still?
At the end we 're going to miss the Gx era, at least then we had some meaningful updates.
I'm still nostalgic about my MDD G4!
There is a misconception about native cpu lines and system overall performance.
The factor that actually its an bottleneck isn't the PCIE lines available at cpu socket or at the PCH, the true performance bottleneck is the available ram bandwidth (which is in function of bus width X nBuses X busSpeed), from that the cpu has to feed the not always slower pcie buses (standard speed), and often peripheral attached to PCIE channels can't handle the full bandwidth with 100% efficiency, that's true even for GPU.
Look at the benchmark of those systems with dual GPU in a single card and those with two GPU on different cards, even some apps are faster on a dual GPU card than on dual cards, while the latest had a theoretical bandwidth advantage of 2x than the one sharing the same slot the truth is that most cpu-gpu traffic are textures (single direction) or crude data-processed data (asymmetric dual direction), and unless you're crunching something very optimized to share compute from cpu and gpu, the most likely scenario will favor the dual GPU card.
I don't know what will sell Apple inside the next Mac Pro, given they surprised providing full Thunderbolt 3 ports to the nMBP very likely they are planning a similar move on the nnMP with 6 or 10 TB3 ports either they lo need to multiplex the cpu lines or sacrifice the 2nd gpu slot for a single dual GPU solution plus 5x TB3 Headers for 10 TB3 Full bandwidth ports plus single PCIE NVMe SSD (and all other peripherals attached to the PCH - dual lan, WiFi, most, etc).
While a dual x16 gpu plus 8 PCIE3 lines multiplexed to 5 TB3 + LAN headers plus rely on PCH for dual ssd requires minor redesign but won't be as optimal as dual GPU on x16 (or high powered single GPU) Plus 5x TB3 full bandwidth (10 ports) plus NVMe, it requires a revision of the thermal core but not as complicated as many may think, the biggest revision is on the back plane busses from 8x pcie3 lines to 20 plus 10 DP 1.2 it's where resides nnMP challenge, even besides being more powerful a single slot dual GPU (or single powerful gpu) it's cheaper to manufacture and CTO.
Mago, interesting speculation on dual GPU cards. If we continue down this pipe dream, then gamers rejoice! You might see support for multi GPU cards and SLI / Crossfire outside the box as well.
Trouble is, they are all fire breathing beasts. Doubt we see this happen (at least not anytime soon with AMD gear).
Trouble is, they are all fire breathing beasts. Doubt we see this happen (at least not anytime soon with AMD gear).
There is no need for Crossfire/SLI.
Metal uses Mantle as a base. If you want to see benefits of Mantle feature set: http://radeon.com/en-us/deus-ex-directx-12-mgpu/
100% scaling of performance.
Whole point of Mantle feature set was to be able to create multiGPU scaling on eGPUs, so you could have perfect scaling over external expansion.
This is of course implementation on AMD GPUs.
Like I said, I think (and hope) Apple is waiting for SKL and X299 to avoid all that mess.
Those statements are true, but the current limitation on the nMP is indeed PCIe lanes good enough for all we want or need.
But with SKL we get 44 lanes PCIe 3 out of the CPU, and another 24 PCIe 3 from the PCH. DMI (which would also be now v3 with X299) will be a bottleneck, and RAM as you say, but there is no way around that yet.
44 lanes on the CPU would be good for both x16 GPUs and dual SSDs. 5 TB3 controllers would fit on the PCH and leave some for the rest, maybe 10GbE.
Sounds good?
[doublepost=1478257575][/doublepost]Now, this would look good with the nMP...
http://wccftech.com/apple-keyboard-concept-features-touch-bar-immensely-slick/
Those statements are true, but the current limitation on the nMP is indeed PCIe lanes good enough for all we want or need.
But with SKL we get 44 lanes PCIe 3 out of the CPU, and another 24 PCIe 3 from the PCH. DMI (which would also be now v3 with X299) will be a bottleneck, and RAM as you say, but there is no way around that yet.
44 lanes on the CPU would be good for both x16 GPUs and dual SSDs. 5 TB3 controllers would fit on the PCH and leave some for the rest, maybe 10GbE.
Sounds good?
[doublepost=1478257575][/doublepost]Now, this would look good with the nMP...
http://wccftech.com/apple-keyboard-concept-features-touch-bar-immensely-slick/
Glue it on top of a Microsoft Sculpt keyboard and then we can talk... been using ergonomic keyboards for more than 2 decades now and my body doesn't want to go back.
Yes but as I said, unless Apple Repeat the Dual GPU Card Approach, for macOS ecosystem purposes its more convenient a Single Powerful 270W-class GPU as Pascal GP100 or Vega 10, also long time ago macOS isn't optimized for dual GPU and unless you have an App supporting specifically this peculiar dual GPU setup (2nd is only available for compute, and while has integrated SLI wGPU1 macOS didn't know what it is.), so would be much more convenient a Single Powerful GPU than two weak GPUs, thats also mandatory to drive 5x 5K Displays.
Whatever I'dont believe the upcoming (hopefully) nnMP to include SKL, but 2 GPU + 2 NVMe (main socket) and 5 TB3 + dual 5Gb Eth (sorry Apple wont adopt 10 GbE, as the industry they are aligned to the new 802.11bz std I read it somewhere, even future Timecapsule/Airport will arrive with this new std for WAN Ports).
Sounds good too, but I fear Apple-AMD alliance if fruitful, will imply the nnnMP (not next, but 2018's) will run on AMD Zen APU or CPU.
Last edited:
But to have 5 TB3 and dual 5Gbps on the PCH with current gen hardware is not easy. Lot of switching to do...
http://www.fudzilla.com/news/processors/42010-macs-with-10nm-intel-in-late-2017
Good editorial on what might be going on with upcoming Intel CPUs and Apple.
One more thing. Because Apple might have partnership with AMD, does not mean we will see them turning away from Intel CPUs. Apple and Intel also have partnership.
Good editorial on what might be going on with upcoming Intel CPUs and Apple.
One more thing. Because Apple might have partnership with AMD, does not mean we will see them turning away from Intel CPUs. Apple and Intel also have partnership.
They may go Zen one day, when x86 performance stagnates enough that there's few differences between chips, but I doubt they'd commit a workstation to AMD anytime soon.
That's bit too long, don't you think? I wonder if Apple is deciding to go "already existed" chips.
Nice mockup, but the artist made a mistake; he used the wired USB keyboard with numpad, and no way Apple would do that.