Will there be a new Mac Pro in 2015?

[DEMinSoCAL]

The 2015 Mac Pro will be thinner and lighter, and will be slightly faster. It will be called the Mac Pro S.

 

[burnsranch]

It will also come in white.

 

[Thunderbird]

...and gold.

 

[AidenShaw]

...and gold.

And perhaps other colors...

 

[Derpage]

 

[koyoot]

Xeon E5-1630 v3 Quad Core 3.7 GHz
16 GB DDR4 2133 MHz ECC RAM
256 GB SSD PCi-Ex
AMD Fire Pro D310 based on Tonga with 1792 GCN Cores and 2 GB of GDDR5 Memory
2999$

Xeon E5-1650 V3 Six Core 3.5 GHz
16 GB DDR4 2133 MHz ECC RAM
512 GB SSD PCi-Ex
AMD Fire Pro D510 based on Tonga with 2048 GCN Cores and 4 GB of GDDR5 Memory
3999$

GPU build to order option: AMD Fire Pro D710 based on next gen Fiji XT chip, with 4096 GCN cores and 4 GB of HBM Memory 1000/600$ BTO cost.

I dont believe that AMD would give a gigantic discount for new gen chips to put them into the low-end Mac Pros. Using Tonga chips in next Mac Pro for base is logical from financial standing point.

To be fair, Tonga is pretty decent GPU, still.

 

[deconstruct60]

But TB2 was done in the late 2013 Mac Pro -
....
by integrating 3x Intel Falcon Ridge Thunderbolt 2 controllers (each driving 2 TB2 ports each).

Would't we just need Intel to release TB3 Alpine Ridge controllers?

How do you drive three Alpine Ridge controllers ( x4 PCIe v3 each ) with just x8 PCIe lanes ? In short Alpine Ridge requires new inputs also. If don't have the new minimal required infrastructure just having the controller doesn't do much.


If Apple wants to stick with three then they will need a 'new' source of added PCIe bandwidth. That would come in the form of new CPU package and/or I/O chipset.

Either the CPU package can move to PCIe v4.0 and a new PCIe switch could use x8 v4.0 lanes to cover 4 x4 v3.0 lanes or the lane count on the CPU package goes up another 4. Neither one of those are happening before 2016.
Same issue with the rest of Mac line up. There is no large block of unused PCIe lanes to which to connect the Alpine Ridge controller until the IO chipsets are upgraded to to supply the additional bandwidth.


If Apple plateaued bandwidth then they could backtrack from three TB controllers back down to just 2 ( 4 ports). The C610 chipset supports 6 USB 3.0 ports. Apple could go with a 2 USB Type C ports that were USB 3.0 + Alternate Mode Display port, 4 regular USB 3.0 ports ( so didn't need dongles for everything), and just 4 Thunderbolt ports. The system can still could drive 6 directly connected displays.but backtracking a bit from Thunderbolt.

 

[deconstruct60]

Hopefully we get a mac pro with displayport 1.3 and a displayport 1.3 only 5k cinema display

Display Port 1.3 only went "final" in Sept 15, 2014; not that long ago. The standard needs to be done before can weave into a new GPU design as a visible feature.

HDMI 2.0 went "final" September 2013 and it didn't make any of AMD's 2014 cards. Display Port 1.2 went final in 2009 and didn't make TB controllers until 2013. A bit better , circa 2011, for a GPU card:

"... Congrats, AMD! Your Radeon HD 6000 Series is the first of its kind to achieve DisplayPort 1.2 Source Certification! "
http://hothardware.com/news/AMDs-Ra...layPort-v12-Certification#zfXhZkTMZCRiC5ro.99

If the timing is exactly right a year is around the minimal time to market. There is some pressure because of the kludge to make hiDPI 5K monitors to work, but those monitors don't really drive the whole GPU market.

Again, 2016 is a more likely window for DP v1.3 roll out. If AMD's update come in the relatively short term future, 2-3 months, they their designs were likely frozen before DP v1.3 went final.

There is no sign that Apple is in any hurry what so ever to do a new "Display" (not in the display business ) or even a new Display docking station. Probably more likely to do a 4K docking station, now that 4K prices have roughly stabilized, than a 5K one. 4K docking will be much easier for the bulk of the Mac line up to drive than some hiDPI 5K monster.

Apple doesn't build just displays specific to the Mac Pro class systems anymore.

 

[deconstruct60]

Consider...
CPU: Haswell's key benefit is primarily battery life. Performance benchmarks on desktop CPUs showed an average 3% improvement (with some benchmarks being worse on Haswell than Ivy).

Haswell key benefit is better power management not battery life. In a laptop one of the side effects of that is better battery life. Servers , workstations , and in particularly the Mac Pro doesn't operate off baterry. The do however consume power. So power management still is an issue.

The Xeon E5 space the better power management gets two things. Increase in upper limit in core count. Second, is more flexibility not do have to chop clock as core count goes up. ( may have to pay through the nose for high clock and high core count, but it is possible. )

The problem with the "regression" Haswell E ( Core i7 x930 ) study linked is that Apple probably isn't going to use Core i7 x930's in future Mac Pros. If look at the benchmarks that "regressed" between Ivy and Haswell most are highlighted by Anandtech to be sensitive to clock speed. Intel did drop the clock speed in the i7 5930 which hurts on the single thread programs/critical sections. However, In the Xeon E5 space

E5 1650 v2 3.5-3.9GHz
http://ark.intel.com/products/75780/Intel-Xeon-Processor-E5-1650-v2-12M-Cache-3_50-GHz

E5 1650 v3 3.5-3.8 GHz
http://ark.intel.com/products/82765...-1650-v3-15M-Cache-3_50-GHz?q=xeon e5-1650 v3

The drop isn't the same. Top end went from 3.9 to 3.8 whereas the top end for the i7 x930's went from 3.9 to 3.7. The "regressions" are going to be half as small on the subset and the upsides just as big. There are no regressions from what folks are running in their previous Mac Pros ( before the 2013 model ).

If don't need lots of mutithread/multiapp horsepower there are still 4 core E5's that Apple will probably use that cost substantially less.


The E5 v3 has better long term protection if plan to use code written in the future. AVX2 is substantially more well rounded than AVX was. It covers integers and floats which is going to make it far more useful. AVX2 is the basis for what is coming in AXV-512 (only not quite as wide). It is going to go mainstream for modern apps over next 4-5 years that aren't trying to cling to the past.

Better power management inside of a power capped Mac Pro is important. It isn't like there is gobs of excess power inside the Mac Pro with the current design. If the CPU cools off faster that opens up a wider window for GPU operations ( and vice versa ).

GPU: The available AMD Tonga GPUs are very incremental improvements over those in the 2013 model and we won't see major improvements until AMD moves to 20nm with Caribbean Islands later this year.

In graphics, maybe. Tonga opens up OpenCL 2.0 . The current cards don't. Again it is a factor of what software is going to be relevant in 2-3 years time. OpenCL 2.0 is a very significant difference. The lament your current apps don't do OpenCL has to remain exactly the same for the next 5 years to be relevant for a system will choose to stick with for that amount of time.

SSD and I/O: There's no new SSD or I/O technology available now.

Yes there is. Besides newer SSD controllers (cover in other posts) there is actually more mature M2 SSD designs now. May be false hope but perhaps Apple might go with a non proprietary M2 socket at this point. Apple may be inclined to remain hostile to 3rd party SSDs but for the Mac Pro that is just a bozo move. The SSD is replaceable. If RAM and SSD are remain so over several generations of this system playing with gratuitously proprietary sockets is just goofy.... even for Apple.

For Scrooge Mc Duck Apple, the USB 3.0 controller is in the C610 chipset so Apple wouldn't have to buy discrete controller to cover the USB ports. Same Intel only solution they clung to before the arrival of the Mac Pro. Given there seems to be no hurry on a Display docking port with USB 3.0, Apple is probably in a hurry to dump those discrete controllers.

 

[VirtualRain]

Haswell key benefit is better power management not battery life. In a laptop one of the side effects of that is better battery life. Servers , workstations , and in particularly the Mac Pro doesn't operate off baterry. The do however consume power. So power management still is an issue.

The Xeon E5 space the better power management gets two things. Increase in upper limit in core count. Second, is more flexibility not do have to chop clock as core count goes up. ( may have to pay through the nose for high clock and high core count, but it is possible. )

The problem with the "regression" Haswell E ( Core i7 x930 ) study linked is that Apple probably isn't going to use Core i7 x930's in future Mac Pros. If look at the benchmarks that "regressed" between Ivy and Haswell most are highlighted by Anandtech to be sensitive to clock speed. Intel did drop the clock speed in the i7 5930 which hurts on the single thread programs/critical sections. However, In the Xeon E5 space

E5 1650 v2 3.5-3.9GHz
http://ark.intel.com/products/75780/Intel-Xeon-Processor-E5-1650-v2-12M-Cache-3_50-GHz

E5 1650 v3 3.5-3.8 GHz
http://ark.intel.com/products/82765...-1650-v3-15M-Cache-3_50-GHz?q=xeon e5-1650 v3

The drop isn't the same. Top end went from 3.9 to 3.8 whereas the top end for the i7 x930's went from 3.9 to 3.7. The "regressions" are going to be half as small on the subset and the upsides just as big. There are no regressions from what folks are running in their previous Mac Pros ( before the 2013 model ).

If don't need lots of mutithread/multiapp horsepower there are still 4 core E5's that Apple will probably use that cost substantially less.


The E5 v3 has better long term protection if plan to use code written in the future. AVX2 is substantially more well rounded than AVX was. It covers integers and floats which is going to make it far more useful. AVX2 is the basis for what is coming in AXV-512 (only not quite as wide). It is going to go mainstream for modern apps over next 4-5 years that aren't trying to cling to the past.

Better power management inside of a power capped Mac Pro is important. It isn't like there is gobs of excess power inside the Mac Pro with the current design. If the CPU cools off faster that opens up a wider window for GPU operations ( and vice versa ).




In graphics, maybe. Tonga opens up OpenCL 2.0 . The current cards don't. Again it is a factor of what software is going to be relevant in 2-3 years time. OpenCL 2.0 is a very significant difference. The lament your current apps don't do OpenCL has to remain exactly the same for the next 5 years to be relevant for a system will choose to stick with for that amount of time.



Yes there is. Besides newer SSD controllers (cover in other posts) there is actually more mature M2 SSD designs now. May be false hope but perhaps Apple might go with a non proprietary M2 socket at this point. Apple may be inclined to remain hostile to 3rd party SSDs but for the Mac Pro that is just a bozo move. The SSD is replaceable. If RAM and SSD are remain so over several generations of this system playing with gratuitously proprietary sockets is just goofy.... even for Apple.

For Scrooge Mc Duck Apple, the USB 3.0 controller is in the C610 chipset so Apple wouldn't have to buy discrete controller to cover the USB ports. Same Intel only solution they clung to before the arrival of the Mac Pro. Given there seems to be no hurry on a Display docking port with USB 3.0, Apple is probably in a hurry to dump those discrete controllers.

In summary, if a Haswell refresh comes out soon, it will not benchmark any better than the 2013 system. It may offer some improvements in fringe cases down the road, but the typical workstation user won't benefit by waiting.

 

[fuchsdh]

In summary, if a Haswell refresh comes out soon, it will not benchmark any better than the 2013 system. It may offer some improvements in fringe cases down the road, but the typical workstation user won't benefit by waiting.

I'd say newer GPUs alone would be a reason for a spec-bump. I think there's also the issue with Apple and appearances; even if the updates are minimal, people prefer to see them rather than waiting and hoping and wondering when they might be able to upgrade.

 

[edanuff]

I think the main thing is that Apple will likely not do a new motherboard and backplane in both 2015 and 2016. If they can do a "Late 2015" that includes every possible major new thing such as Skylake, TB3 and/or DP 1.3 (which could be done independently of TB3, since TB3 requires a new connector), etc. and then not do another new motherboard/backplane until 2017, then they would probably do that. But if there's anything that they won't want to punt till 2017 on that just isn't ready to ship in 2015, then it seems like a 2016 model is more likely. By the way, this doesn't mean that a spec bump isn't going to happen in 2015. I consider anything slotted, such as the memory, GPU (D350/D550,D750?), and SSD fair game to be spec upgraded at any time. That spec bump will happen exactly 15 days after you buy a 2013 model in 2015

 

[deconstruct60]

In summary, if a Haswell refresh comes out soon, it will not benchmark any better than the 2013 system.

Not a summary of what I wrote ( at least in this reality ). Even the sentence is filled with skew and spin.

There is no "Haswell refresh" coming because there is no Haswell system out now. Can't 'refresh' what hasn't shipped yet.

There is a new motherboard and internals coming. Whether Apple uses Xeon E5 v3 , v4 , v5 ( whatever long period into the future), the E5 v2 in the 2013 model are dead end. The chipset has been superseded. That means changes and upgrades; not a "refresh".

For a multicore workloads , E5 v3 (or better ) systems will benchmark better. Why folks want to buy a "higher than mainstream" core count ability system and do single threaded drag racing more so begs the question of why buying any Mac Pro.

It may offer some improvements in fringe cases down the road, but the typical workstation user won't benefit by waiting.

The fringe case is far more so the current model. Being stuck sub OpenCL 2.0 and sub AVX2 means a less robust in the variety of workloads it can deal with. Apple is weaving OpenCL into the OS. Likewise AVX was weaved into FCP X years ago at this point. The better virtualization in Haswell is used daily.

The Mac Pro 2013 is a "version 1.0" model. Every "version 1.0" model gets cleaned up with the follow ons. Same thing is going to happen here. Even the new chopped down Mac Mini has more internal storage flexibility than the Mac Pro. [ SATA SSD blade using the completely unused SATA subsystem present would at least bring the MP to same 2 internal drive parity with the other Mac desktops. ]

An upgrade would be Incrementally better. Folks with stuff that is "good enough" right now waiting doesn't hurt. Folks with 2013 models probably should be worrying about other things than trying to poo-poo the next generation. Since riding a relatively new system you don't have to upgrade.

 

[goMac]

If they're going to do Haswell, it should be any time now.

DisplayPort 1.3 won't be enough for a 5k Cinema Display. They'll be needing Thunderbolt 3.

 

[deconstruct60]

I think the main thing is that Apple will likely not do a new motherboard and backplane in both 2015 and 2016.

They don't really need to. The chipset for the E5 v3 and v4 is basically the same. That basically will cover 2015 ( v3) and 2016 ( v4).

If they can do a "Late 2015" that includes every possible major new thing such as Skylake, TB3 and/or DP 1.3 (which could be done independently of TB3, since TB3 requires a new connector), etc. and then not do another new motherboard/backplane until 2017,

E5 v5 (Skylake) likely isn't coming until earliest Q3-Q4 '16 ( pretty good chance it will be Q1 '17 if bump in PCIe version causes a similar major release slide as the last bump. )

If Apple is going to hook Mac Pro updates to Thunderbolt updates then someone in Cupertino is drinking a ton of "reality distortion field" kool-aid. Thunderbolt's update cycle is 2-2.5 years long and is completely unaligned with both the GPU and CPU update cycles. To design a workstation system soley and primarily around a i/O support chip is wandering into bozo land.

If the Mac Pro had just one TB v2 controller than perhaps it would be more dire situation to upgrade as quickly as possible to TB v3. It has three. If they want to keep three then they will have to wait for the infrastructure to support three to come. It isn't coming any time soon for workstation class systems.

 

[edanuff]

DisplayPort 1.3 won't be enough for a 5k Cinema Display. They'll be needing Thunderbolt 3.

Sure it will, DP 1.3 can do 5K. Not clear what the connector situation is though. If it's possible to run DP 1.3 at the current mini-DP form factor, we might see that Apple does something with it in advance of TB3.

 

[koyoot]

Newest info on availability of 4096 GCN core chip called Radeon R9 380X is that full availability of that chip will be on late Q3 of 2015.

What that means?

Apple bought all the chips available. Same thing happened for Tonga iMac Radeon chips. Apple bought all the dies from AMD.

All in all, were very close to update of new Mac Pro.

 

[goMac]

Sure it will, DP 1.3 can do 5K. Not clear what the connector situation is though. If it's possible to run DP 1.3 at the current mini-DP form factor, we might see that Apple does something with it in advance of TB3.

Yes, but the Mac Pro doesn't use mDP as a port, it uses Thunderbolt. And Thunderbolt controllers are not currently compatible with a DP 1.3 source until you get to Thunderbolt 3.

There just isn't enough bandwidth on the controllers to carry a DP 1.3 signal.

----------

Newest info on availability of 4096 GCN core chip called Radeon R9 380X is that full availability of that chip will be on late Q3 of 2015.

What that means?

Apple bought all the chips available. Same thing happened for Tonga iMac Radeon chips. Apple bought all the dies from AMD.

All in all, were very close to update of new Mac Pro.

While i'd love to see them use the 380x, I'd have a hard time believing Apple bought out AMD's entire supply for the Mac Pro.

 

[AidenShaw]

Newest info on availability of 4096 GCN core chip called Radeon R9 380X is that full availability of that chip will be on late Q3 of 2015.

Apple likes surprises - the GM204 chips have good OpenCL performance and sip electricity. The nnMP could have OpenCL+CUDA!

Not a summary of what I wrote ( at least in this reality ). Even the sentence is filled with skew and spin.

You're responding to someone who's been bad-mouthing the Haswell-EP for months.

We have some integer-heavy apps that are seeing 20% improvements on our E5-v3 systems vs E5-v2. AVX2 brings "loop unrolling" into the hardware. It's huge.

Just because software optimized for Nehalem and Sandy-Bridge doesn't see a big jump on Haswell doesn't mean that Haswell sucks. As apps add AVX and AVX2 support the gap between E5-v2 and E5-v3 will get bigger and bigger.

Sandy Bridge and Ivy Bridge are dead. Apple will have to upgrade the MP6,1 simply because Intel will stop producing the old chips. (And don't pretend that they aren't "old chips".)

 

[VirtualRain]

Will there be a new MacPro in 2015 ?

Apple likes surprises - the GM204 chips have good OpenCL performance and sip electricity. The nnMP could have OpenCL+CUDA!




You're responding to someone who's been bad-mouthing the Haswell-EP for months.

We have some integer-heavy apps that are seeing 20% improvements on our E5-v3 systems vs E5-v2. AVX2 brings "loop unrolling" into the hardware. It's huge.

Just because software optimized for Nehalem and Sandy-Bridge doesn't see a big jump on Haswell doesn't mean that Haswell sucks. As apps add AVX and AVX2 support the gap between E5-v2 and E5-v3 will get bigger and bigger.

Sandy Bridge and Ivy Bridge are dead. Apple will have to upgrade the MP6,1 simply because Intel will stop producing the old chips. (And don't pretend that they aren't "old chips".)

Agree with everything you say, but most workstation users who might be expecting a 2015 nMP to perform better than the 2013 model (assuming it's launched in the near future) will be disappointed... It will bench almost exactly the same as the 2-year old model in all the common apps and benchmarks used today.

When will software be optimized for it? Perhaps soon, perhaps by the time an even better platform is available. Buying a system on the promise of unlocking performance through software in the future has never paid off in my experience.

 

[koyoot]

Apple likes surprises - the GM204 chips have good OpenCL performance and sip electricity. The nnMP could have OpenCL+CUDA!

No, it doesnt. GM204 is biggest marketing gimmick I have seen in my whole life.

Yes it does OpenCL good, but not as good as AMD GPUs. When it has to do it, unless there is a TDP cap in BIOS is SUCKS power as hell. It uses only 35W less than R9 290X. If you have a TDP cap it will not use more than 180W, but... will be MUCH slower, cause of nacked TURBO state of GPU. They sip electricity only on average. Peak and max power draw are the ones to look.

Funniest part, AMD in their marketing materials show max TDP of their cards. Nvidia for Maxwell shown average, and that is the difference, that people dont see.

I wonder what will happen, when R9 380X will be made on 20 nm process, and will draw around 200W, and give 7-8 Tflops? So far, ChipHell slides show that TDP of FijiXT should be around 200W.

Secondly, why would Apple use CUDA, if Mantle in future will be much more than CUDA is? Mantle will replace DX, OpenCL, OpenGL, and CUDA. And it will be one simple platform.

I really sugest to read carefully reviews on sites, and start paying attention to tech forums, not only Macrumors.

 

[AidenShaw]

I really sugest to read carefully reviews on sites, and start paying attention to tech forums, not only Macrumors.

Mantle and CUDA are apples and oranges.

Mantle is a 3D rendering API that competes with Direct3D and OpenGL.

CUDA is a parallel computing platform.

Both Mantle and CUDA lean on the same backend, the floating point vector processors located inside of modern GPUs. Whereas Mantle competes with Direct3D and OpenGL to create 3D scenes, CUDA competes with OpenCL and DirectCompute to perform parallel computation.

It is entirely possible to lean on the CUDA/OpenCL/DirectCompute APIs to assist with certain game calculations. CUDA is used by NVidia's PhysX middleware, and certain games have used OpenCL to perform certain physics and filter computations. DirectX 11 allows for Direct3D and DirectCompute to be used at the same time and within the same context.

http://www.tomshardware.com/answers/id-2376771/mantle-cuda.html

Overall, while NVIDIA can’t win every compute benchmark here, the fact that they are winning so many and by so much – and otherwise not terribly losing the rest – shows that NVIDIA and GM204 have corrected the earlier compute deficiencies in GK104. As an x04 part GM204 may still be first and foremost consumer graphics, but if it’s faced with a compute workload most of the time it’s going to be able to power on through it just as well as it does with games and other graphical workloads.

http://anandtech.com/show/8526/nvidia-geforce-gtx-980-review/20

http://wccftech.com/nvidia-maxwell-gtx-980-opencl-benchmarks/

Comments?

 

[koyoot]

Comments?

Yes.

All I will say right now is that, what we see from Mantle right now is only being Low-Level API for gaming.

Future is long...

About the performance - I have already said what I have seen on reviews, and on user experience from using those cards.

http://www.tomshardware.com/reviews/nvidia-geforce-gtx-980-970-maxwell,3941-12.html

Windforce does not have TDP cap - it draws almost 300W in GPGPU constantly. Reference does have TDP cap, but also it has way lower TURBO state clocks. Thats why it draws only around 180W of power.

Imagine now. 280W TDP Windforce GTX980 delivers 5Tflops of power. AMD releases 200W with 8Tflops version of their GPU as a reference and it blows away not only the 180W chip but also 280W version of GTX980.

Even more! 125W Apple version of R9 380X should bring around 7Tflops of power. And also give double precision, which Maxwell GPUs don't have, and never will have.

EOT.

 

[deconstruct60]

Sure it will, DP 1.3 can do 5K. Not clear what the connector situation is though. If it's possible to run DP 1.3 at the current mini-DP form factor, we might see that Apple does something with it in advance of TB3.

TB v3 doesn't particularly solve the DP 1.3 problem. TB v3 still only deals with DP 1.2. Note the DisplayPort version in leaked Intel's slide:

http://arstechnica.com/gadgets/2014...olt-doubles-speeds-but-changes-the-connector/

To get to DP 1.3 sooner rather than later Apple would have to get 'raw' DP 1.3 signal out to the connector. It is unlikely Apple will go to a display port only connector. Just look at the trends on the other Macs ( especially the new MBA 12" one with just a single "kitchen sink" do-everything port ). There are no Mac now that have a display only output besides the HDMI variants on two. ( and that isn't bleeding standards compliant edge either).

Apple could go USB Type C if they can find a DP 1.3 USB 3.0 switch to put behind the connector and the DP "Alternate Mode" also covers DP 1.3 signaling. It would also be consistent if Apple is going to slap Type C connectors on all new Macs ( which wouldn't be surprising).

TB v3 controller may pragmatically be a pre-built USB Type C Alternate mode backing switch. The "Four Modes" , 100W power , and lower z height, suggests that will be the case. To keep costs reasonable though it is unlikely it will be the super duper does everything at the most bleeding levels switch. A TB controller is probably going to be skewed to just be top end TB in support. The others will be the more mature ( and easier to implement because all the kinks have been worked out elsewhere) stuff. When TB came out it supported DP 1.1 and PCIe v2 which were not the bleeding edge at that point.

However, for a 2015 product though, the stumbling block is can acquire a GPU that actually outputs DP 1.3 while designing building a custom GPU card (have to license the reference design) , write OS X custom drivers ( OS X driver availability trails Windows/Linux as smaller market ) , and get access to the volume needed for their production.

AMD released the Tonga based W7100 in August 2014. It wouldn't be surprising to see Apple's customized version of that drop in Feb-March 2014 ( trailing by about 6 months). The driver work gets a modest priority bump since similar tweaks needed for the M295X in Retina iMac also (and can protoype off of standard stuff in AMD's pipeline; e.g., earlier reference boards for mainstream products).

Apple releasing a Mac Pro GPU cards that are bleeding edge of AMD ( or Nvidia) would be a huge departure from what Apple has done over last 5-8 years. Throw Apple custom board & connector designs on top of that and it is just more unlikely.

The Pirate Island 380X stuff is just likely too late to make a 2015 Mac Pro given the additional logistical overhead a Mac Pro GPU card adds.

 

[goMac]

TB v3 doesn't particularly solve the DP 1.3 problem. TB v3 still only deals with DP 1.2. Note the DisplayPort version in leaked Intel's slide:

That's true. I'm assuming TBv3 will be changed to use DP 1.3, because if it isn't, we really all are truly screwed.

I don't know exactly what Apple is planning with USB type C. I can't believe they'd abandon Thunderbolt. Maybe like you suggest they could get it to switch to DP. Maybe another option is that Thunderbolt gets some sort of bypass that when you plug in a DP display directly, it bypasses the Thunderbolt controller.

I'm pretty sure what will happen is TBv3 will get support for DP 1.3 though.