New MacPro predictions? Hopes? Fears?

[Wayne Morellini]

Reputation is the aim I think they should go for. That will extend numbers to tens of millions, for inevitable upgrades. But apart from the Mac Pro, way back a few years ago, Apple has been slipping. Just look at the amount of innovative features that slipped them by over the years, from around the closed loop obsessions of App stores and lack of blu ray (or "blur ray" as my spell checker wants to call it in the 4K bluray day. Like it watches the store demos. Lol!). They should have maintained a fully kitted out top model of each series that had the best spec to be bought at a reasonable price (pretty much 90%+ of whatnthey could do compared to.machines double the price+). Instead even HP sells machines more interesting then an IMac, and the MacPro needs to have a twices as big, twice as good model now (see my previous posts).

So, well see, but I think they need that stsckable mini.blade system for serious user and in house sever, just keep adding blades as you need them).

 

[Jarek Zabczynski]

Workstations are Apple's new "bag of hurt" ...

 

[Cid J Salcido Uyarra]

This thread is hilarious

Just buy a used 2013 Mac Pro they're so cheap and get the job done.

Why do you want a $5000 New Mac Pro with Thunderbolt 3 and no support what so ever?

I use to wait and I got sick of waiting even built a Hackintosh in between, ran it til it died. Then bought a 2012 Tower got sick of lugging around from storage to all my gigs plus no thunderbolt.

So happy with my 8-Core 64GB RAM 1TB SSD Dual D700 all I ever needed and handles Phantom 4K Flex gigs like a champ.

I mean seriously beyond 1000MBs your not gonna need more and with certain RAID configs u can get 2200MBs which is TBolt 3 speeds on a Mac Pro you are using 2 Tbolt2 ports but u can get those speeds.

Besides a Faster GPU they aren't gonna be that much faster CPU wise.

OWC has some solid deals on some 2013 MP.

 

[Antony Newman]

If Apple seeks to sell its products through diffentiation (through software and hardware expertise), and 'delight' people with interaction - it might go some way to help understand what they have created, and what 'pro' gear we might expect (or hope for ).

Existing
+) Two years ago, I was surprised that Apple released a 5K iMac screen with a 27" diagonal. I didn't see the practical benefit, but this was a chance to differentiate.
+) When Mansfield 'rolled in to give project Titan a new direction, (rumour is) he didn't believe that would create a differntiatble product, and so the team was repurposed with a new objective.
+) The nMP had something differentiatable - it was Tiny and Quiet, and tuned to enable 'editing in 4k' in FCPx.
+) Last year Apple introduced P3 mobile displays, this year the iPhone7 displays have purportedly incredible P3 colour accuracy right out of the box. I think this both differntiates from yesteryear products, and create demand for 'real colours' in displays (someone has commented previously they don't agree my perspective on this expectation).

So .. what has to come next, and where is Apple at the mercy of the innovation of others?

AJ

 

[Jeff Kilgroe]

More than differentiate, Apple likes to innovate where they can. Often this innovation goes unnoticed by most consumers or users. Especially since they innovations are coming incrementally or within various components inside established devices. Some of the seemingly small or insignificant changes are often large undertakings from an R&D perspective. The P3 calibrated screens on the latest devices are a prime example of this. Along with the innovations comes refinement. Apple has designed the best and most capable TFT screens. Never mind they're on phones, I would kill a dozen hookers and two clowns right now if I could have that tech and pixel density in a 27~32" desktop display. The 5K iMac display is another example of this. And most people don't understand or see the benefit until they really use it on a regular basis. IMO, at this point it needs to be a 6K display. I would love to see a 30" 6K iMac.

The nMP is a great concept. Somewhat flawed in execution, but the real problem is that Apple made some very poor decisions in terms of hardware specs, pricing and ultimately what did the system in for media pros is the shoddy graphics drivers and support. If Apple were to update the nMP now with USB 3.1 and TB3 ports and better GPUs that would make it more attractive in the current market. But it really wouldn't be a more powerful or capable system. The GPUs would be the biggest factor contributing to performance. However, if they remain with AMD, they're just spinning their wheels. AMD and Apple have done a shit job of providing real software support for these GPUs and I don't see that changing. AMD does this on the PC side -- great hardware that has excellent specs, but terrible drivers and the hardware is never fully realized. Same on the Mac, only 10X worse. I won't be buying a new MP update unless they put some nVidia GPUs in there that actually support the CUDA we need for best performance in many applications and because we know nVidia will provide us with far better driver support.

 

[Antony Newman]

There was a comment about the inability for CPUs to ramp up in power.

Whilst I believe this true until Intel, who need to protect their long term rollout strategy, and (IMHO) provide incremental advances in bandwidth - I don't think that Apple will 'be held back'.

ARM released a coherent mesh for 128 core design with 8 x high speed memory links and a cool 1 TB/s to play with for intermesh bandwidth.

If Apple IPC Already matches Intel (I think they may be 25% ahead with similar processes) - and the CPU cost is 1/10 - 1/40th the cost of buying in Intel ... could THIS be the year Apple switch?

ARM backbone + Nvidia + A10X Fusion (with big.Little) + TSMC Fanout tech + 10nm = something for Guy Forks?

Edit 1 : I see rumours for a 2017 ARM laptop with a Apple Hurricane processor. Mac Air first?

Edit 2 : MacOS has new Apple chip compatibility in header files. Hmm. Skylake MacBook Pro's this year, Hurricaine Airs in 2017 (with bit-code recompile courtesy of Apple)? Not sure where this could leave a new Cylinder ... what about a 64-128 core Hurricane with Dual Volta by end of 2017?

AJ

 

[Antony Newman]

"I would kill a dozen hookers and two clowns right now if I could have that tech and pixel density ..."

Classic

AJ

 

[Curran Giddens]

I won't be buying a new MP update unless they put some nVidia GPUs in there that actually support the CUDA we need for best performance in many applications and because we know nVidia will provide us with far better driver support.

Apple has one last chance to make it right. Add TB3 and external GPU capability...it would be sweet to have 4x nVidia GPUs.

 

[Jeff Kilgroe]

External GPUs have been a non-starter in the PC world for professional work. It's kinda cool on a laptop to throw a 1080 or p-Titan X into a TB3 enclosure and reap some big benefit there over the mobile offerings. Unfortunately in the desktop space or on more powerful systems that can actually drive these GPUs to their limits, the bandwidth choke over TB3 is still a significant bottleneck. Not to mention throwing full size GPU cards into TB3 enclosures and tethering them via an inferior interface that has 100X the latency and only 1/5th the bandwidth of their native PCIe 3 X16, really takes its toll. And takes up space. It's more convenient to build a self-contained tower.

If we get full support for the E5 Xeons or Kaby Lake i7's and supporting chipsets in a new OSX update, you're better off to build a real system and make a Hackintosh of it, if OSX is that important to you. I'm sure that the next iMac and Mac Pro, if they release one, will be awesome systems. I have serious doubts that they will be the systems we want in this community though as we move forward with 8K R3D processing.

 

[Peter Majtan]

External GPUs have been a non-starter in the PC world for professional work. It's kinda cool on a laptop to throw a 1080 or p-Titan X into a TB3 enclosure and reap some big benefit there over the mobile offerings. Unfortunately in the desktop space or on more powerful systems that can actually drive these GPUs to their limits, the bandwidth choke over TB3 is still a significant bottleneck. Not to mention throwing full size GPU cards into TB3 enclosures and tethering them via an inferior interface that has 100X the latency and only 1/5th the bandwidth of their native PCIe 3 X16, really takes its toll. And takes up space. It's more convenient to build a self-contained tower.

If we get full support for the E5 Xeons or Kaby Lake i7's and supporting chipsets in a new OSX update, you're better off to build a real system and make a Hackintosh of it, if OSX is that important to you. I'm sure that the next iMac and Mac Pro, if they release one, will be awesome systems. I have serious doubts that they will be the systems we want in this community though as we move forward with 8K R3D processing.

What you want is compact 1U rack-mount main unit (ala old xServe) which focuses on CPU and memory, while providing all the standard I/O with on-chip graphics PAIRED with 3~4U expansion that is connected via robust pipe (ignore Thunderbolt) to allow full potential of the cards in the expansion. This way user that need pure CPU power and memory can have a dense server farms, which folks in custom scenarios (like us in film production) can fill in the expansion with cards that benefit us. GPU's, high-performance storage, Video I/O, etc... Everyone can then customise the expansion for what they need...

My two yen...

:sifone: Peter

 

[Wayne Morellini]

I agree Peter, I was just going to suggest a mini blade system for seriouse is rs, rather than the aging toy performance.

People ask why Apple still sticks with Intel, maybe because of the Phi promise. But Intel promisees a lit but what happens to them. Remember the 1 watt array of processors. The due date came and went. Remember it became Larrabee and something else with vlwi, where? Now Phi, but will Phu exceed GP-GPU. All they need to do is specialise the GPU and give CPU front end like capability (already Nvidia is debeling RiscV). Legacy x86 support is an issue. Phi sounds like somethng they should aim at for $1000, maybe $2000.

People keep talking ARM, but RiscV is a good successor. Arm s likely to need a radical overhaul so much to compete, will it really be an ARM anymore. The thumb instruction sets go towards this, but how much more do they need?

Looking up the Phi architecture, and omni p, they sound like what I was going to suggest on one hand and the example execute over network the bunch I know was using for a while.

I was going say is that all you needed was software to take advantage of a server farm like scenario, where you insert CPU/GPU blades in a stack, and mayge run linux. Even with windows, only one needs to run windows, and the rest set up to act as resources, but somehow, knowing the history of computing, I doubt its that simple.

Using such a blade architecture with ARM or RiscV Linux/Windows (binary translation) would be great. Eben with Intel, when you upgrade you sell the oreviouse box to somebody else down the chain, or as they fall off the end. Nvidia does leading ARM cuda solutions already. Now, imagine even a $1000 system based on that with x86 binary translation. This constriction stuff is incredibly counter productive.

I'm still waiting for a good general purpose/specialised mass hybrid system.

 

[Wayne Morellini]

If Apple seeks to sell its products through diffentiation (through software and hardware expertise), and 'delight' people with interaction - it might go some way to help understand what they have created, and what 'pro' gear we might expect (or hope for ).

Existing
+) Two years ago, I was surprised that Apple released a 5K iMac screen with a 27" diagonal. I didn't see the practical benefit, but this was a chance to differentiate.
+) When Mansfield 'rolled in to give project Titan a new direction, (rumour is) he didn't believe that would create a differntiatble product, and so the team was repurposed with a new objective.
+) The nMP had something differentiatable - it was Tiny and Quiet, and tuned to enable 'editing in 4k' in FCPx.
+) Last year Apple introduced P3 mobile displays, this year the iPhone7 displays have purportedly incredible P3 colour accuracy right out of the box. I think this both differntiates from yesteryear products, and create demand for 'real colours' in displays (someone has commented previously they don't agree my perspective on this expectation).

So .. what has to come next, and where is Apple at the mercy of the innovation of others?

AJ

The displays were just long delayed progressions. For instance, back then a rec2020 relative 4k or 8k mobile display would have been remarkable enough. This prosumer like increments is not what I buy for when I need 30 inch - 60 inch 8k 16k work areas to do photos or multiple control, 4k and 8k video views. It is all a nature time. And before anybody comes along to tell me you need a super computer or something just to display a desktop and 2D images, your likely a moron whoever you are. (One of the better aspects of potential dementia, expressing it as it is). However, nobody here would likely think that, I'm happy to say.

They are drip feeding cheaper to do improvements, not so much very high end professional use. They would have more respect if they difnthise, and drip fed yjrm down to the existing product lines. After they archievr and refine they can make it cheaper quicker, and license the designs for previous years tech back to the commodity manufacturers. Makes good business sense, but the Istore.

More than differentiate, Apple likes to innovate where they can. Often this innovation goes unnoticed by most consumers or users. Especially since they innovations are coming incrementally or within various components inside established devices. Some of the seemingly small or insignificant changes are often large undertakings from an R&D perspective. The P3 calibrated screens on the latest devices are a prime example of this. Along with the innovations comes refinement. Apple has designed the best and most capable TFT screens. Never mind they're on phones, I would kill a dozen hookers and two clowns right now if I could have that tech and pixel density in a 27~32" desktop display. The 5K iMac display is another example of this. And most people don't understand or see the benefit until they really use it on a regular basis. IMO, at this point it needs to be a 6K display. I would love to see a 30" 6K iMac.

The nMP is a great concept. Somewhat flawed in execution, but the real problem is that Apple made some very poor decisions in terms of hardware specs, pricing and ultimately what did the system in for media pros is the shoddy graphics drivers and support. If Apple were to update the nMP now with USB 3.1 and TB3 ports and better GPUs that would make it more attractive in the current market. But it really wouldn't be a more powerful or capable system. The GPUs would be the biggest factor contributing to performance. However, if they remain with AMD, they're just spinning their wheels. AMD and Apple have done a shit job of providing real software support for these GPUs and I don't see that changing. AMD does this on the PC side -- great hardware that has excellent specs, but terrible drivers and the hardware is never fully realized. Same on the Mac, only 10X worse. I won't be buying a new MP update unless they put some nVidia GPUs in there that actually support the CUDA we need for best performance in many applications and because we know nVidia will provide us with far better driver support.

My nearly 10 year old viewsonic TV cane "factory color calibrated". Looked good, but I noticed that many cheap TV's cane out with similar looking color. After all thos time, I don't think P3 is such a leap
The present P3 TV aspirational standard instead of Rec2020, let's face it, is there so lesser TV's and technology like.OLED can look good. Theybvpyld have also introduced an aspirational Rec 2020 spec for other manufacturers to try to get close to (98%). Why follow yesterday's technology professionally. Things change and 2 years time you definitely want Rec2020 and to stick the P3 somewhere useful. Gentlemen, we think like consumers, let us think like professionals (kick starting professionals).

There was a comment about the inability for CPUs to ramp up in power.

Whilst I believe this true until Intel, who need to protect their long term rollout strategy, and (IMHO) provide incremental advances in bandwidth - I don't think that Apple will 'be held back'.

ARM released a coherent mesh for 128 core design with 8 x high speed memory links and a cool 1 TB/s to play with for intermesh bandwidth.

If Apple IPC Already matches Intel (I think they may be 25% ahead with similar processes) - and the CPU cost is 1/10 - 1/40th the cost of buying in Intel ... could THIS be the year Apple switch?

ARM backbone + Nvidia + A10X Fusion (with big.Little) + TSMC Fanout tech + 10nm = something for Guy Forks?

Edit 1 : I see rumours for a 2017 ARM laptop with a Apple Hurricane processor. Mac Air first?

Edit 2 : MacOS has new Apple chip compatibility in header files. Hmm. Skylake MacBook Pro's this year, Hurricaine Airs in 2017 (with bit-code recompile courtesy of Apple)? Not sure where this could leave a new Cylinder ... what about a 64-128 core Hurricane with Dual Volta by end of 2017?

AJ

AJ, I luv you man. If you were a pretty woman I might have wanted to marry you after saying that.

Yes, if they can introduce something like this cheap, in nice stackable mini blade form that could boot to Windows or Linux with binary x86 transliteration, then that's good enough. 128 cores, 2k cores. Take a look at the power requirements of the current nvidia tx1 8 core. They have maybe two generations to go before they tap out arm energy per cycle savings, which might require a switch to overcome short term.

External GPUs have been a non-starter in the PC world for professional work. It's kinda cool on a laptop to throw a 1080 or p-Titan X into a TB3 enclosure and reap some big benefit there over the mobile offerings. Unfortunately in the desktop space or on more powerful systems that can actually drive these GPUs to their limits, the bandwidth choke over TB3 is still a significant bottleneck. Not to mention throwing full size GPU cards into TB3 enclosures and tethering them via an inferior interface that has 100X the latency and only 1/5th the bandwidth of their native PCIe 3 X16, really takes its toll. And takes up space. It's more convenient to build a self-contained tower.

If we get full support for the E5 Xeons or Kaby Lake i7's and supporting chipsets in a new OSX update, you're better off to build a real system and make a Hackintosh of it, if OSX is that important to you. I'm sure that the next iMac and Mac Pro, if they release one, will be awesome systems. I have serious doubts that they will be the systems we want in this community though as we move forward with 8K R3D processing.

Jeff, isn't it more a matter to build programs designed to maximise GPU performance without much bandwidth.

As far as thunderbolt goes I'm very dissapionted. They promised 100Gb/s but where is it. I worked out they could configure the original port to do it over a special cable. They are just killing the goose and letting an inferior USB spec rule all over again. It was bad enough seeing FireWire gobto nothing. Probably virtually everything USB does they could have made TB do, even with close to low cost low speed periphial support, but people don't want to listen. It should have been made and extendable/multi slot with external transparent GPU support from the beginning. It could now do over 200gb/s in one old port.

However, the future. Intel is pursuing optieoectronics. The amount transported over optical even years ago were in the terabit/s range. However, just dumping the maximum (whatever that is at the moment) on a chipset at the moment is probably too much, but in the future with optical developments maybe big data is possible. However, it does mean, initially, dual sli, itlr an entire PCI card slot cabinet could fit. Just depends on what current design can fit.

 

[Curran Giddens]

If we get full support for the E5 Xeons or Kaby Lake i7's and supporting chipsets in a new OSX update, you're better off to build a real system and make a Hackintosh of it, if OSX is that important to you. I'm sure that the next iMac and Mac Pro, if they release one, will be awesome systems. I have serious doubts that they will be the systems we want in this community though as we move forward with 8K R3D processing.

That is the other option I was thinking. I've never been a Windows user on a day to day basis and I would love to avoid it if at all possible. If I did the Hackintosh route I would need some really good step-by-step instructions. Then just never update anything unless absolutely necessary. If it works, don't touch it. I can live with that.

 

[Antony Newman]

Silicon Photonics are needed where large storage systems are a long way from CPU nodes (minimising latency, and for supercomputers minimising transmission wattage).

+) http://www.nextplatform.com/2016/08/17/intel-leverages-chip-might-etch-photonics-future/

But low latency (Silicon Photonics) IMHO does Not solve on chip traffic congestion, ie the bottlenecks when multiple CPUs contending for data, and inter CPU communication is strangled during concurrent operations. Intels Racetrack (or Ring) solution sort of scales with multiple circuits between the current number of cores - but this is rather two-dimensional.

Intel are (again IMHO) not keen for CPUs in High Performance computers to suddenly become twice as fast. Intel's 10nm node has been ready in sufficient yield to build chips. But these have not been offered - even at a premium, because Intel want everyone to buy chips that are only 5% - 15% faster year-on-year, and desire large companies to be in a 'perpetual evergreen' situation to maximise profit on selling hardware. With no 'AMD' to compete - Intel have been holding back their 10nm chip electronics from the planet ... until their hand is forced. Its a similar story with TB1, TB2 and TB3. Slow deliberate rollout of network speeds to avoid multiple systems from efficiently working together as one homogeneous unit.

Servers

Intel has been able to repel Cavium and Applied Micro from entering 'their' server world so far, as the Enterprise software is still being matured, and the highest performing chips designed by them have lowish IPC (instructions per clock) compared to Intels mature Xeon. However the TC (Total Cost of Ownership) is still tipping towards Cavium under certain loads.

+) https://en.wikipedia.org/wiki/Comparison_of_ARMv8-A_cores

Cavium's 'Thunder X1' wimpy core solution using 2 x 48 core CPUs on (old) 28nm processes - is about 5 times slower than a top end Xeon.
Their next (Thunder X2) will still be 2.5 times slower. Cavium will roll out ThunderX2 with less wimpy cores in 2017 - but even these are about 2.5 times less capable than an top end Xeon (per CPU)

+) http://www.eetimes.com/document.asp?doc_id=1330492&image_number=1

Applied Micro's X-gene3 can scale up to 64 CPUs in a Chip - but will only reach about 60% the processing speed of a top end Xeon CPU.

ARM

http://www.anandtech.com/show/10711/arm-cmn-600-dmc-620-128-cores-8-channel-ddr4

+) With ARM's release of a 128 core x Coherent Fabric that can shunt 1 TByte / second around the 'mesh' - even wimpy cores that are 1/3 as capable as a tope end Xeon will be able to Scale to a performance at Intels Current Top end.
+) This fabric also has a Cache.. and is 5 x the performance of ARMs previous fabric (which was no slouch)
+) This cache will no doubt form part of CCIX initiatives (eg specialist hardware accelerators in which there are NO device drivers, and therefore virtually zero latency)


Apple
+) In Sierra .. Apple have purportedly changed the submission of code to the Mac App store from 'x86' (macOS) ... to a the stage before which is Operating System agnostic .. called 'bit-code'.
+) Apple have added a machine language type in Sierra that supports a non-ARM created (ie CPU designed by Apple) but ARM compatible.
+) Apple A10 Fusion chip is equivalent to Intels Xeons in IPC
+) Apple A10 Fusion 'big' core use less power at TSMC 16nm (Real world 18.3nm) than Intel's 14nm (real world 13.4nm).
+) Apples A10X (they might call it A11X) is on TSMC 10nm (Real world 11.3nm) is very likely to jump past Intel on performance per watt - but not necessarily absolute performance. This seems like a perfect fit for the Mac Air. Less heat to get rid of + No money to pay to Intel + a Safe way to Prove MacOs on Apple ARM .. and for all software submitted in the Max App store .. automatic recompilation at Apple HQ.
+) Apple have a hardware version of ARMs big.Little context switch (ARMs design has an achilles heel in that the OS control the switch).

TSMC had advised that they will be earning money from 10nm node in 2017 Q1 .. and that their 7nm node (real world 8.2nm) will be ready for sample testing later in the year.

+) Mac Air : (speculation) Apple will develop an ARM v8 compatible Mac Air .. to be on sale in 2017 H1, and that ..
+) Mac Pro : (speculation) Apple will develop either a 10nm (64 core) or 7nm (128 core) CPU Mac Pro .. and that Nvdia GPUs will adorn it, supporting Metal (great for the OS), CUDA (for stand alone Apps that need it), and OPEN CL/GL (for Apps that have not yet moved to Metal). If a 128 core Apple CPU chip costs Apple $500 at source and will meet your CPU needs for the until Intel 'Catch up' with Apple, and IF 2 x VOLTA GPUs meet our 8K needs ... I think this could be a winning formula.

AJ

 

[Mike Halper]

I agree with most of this. I don't think there's been a real reason for Apple to update the Mac Pro since the performance increase wouldn't be that great compared to what it is now. Intel has been slow to really improve performance from one CPU gen to the next, and who's going to buy a $10,000 Mac Pro one year and then spend another $10,000 on another Mac Pro a year or 2 later for a measly 10% performance increase at best. I think that, combined with all the AMD GPU issues with the Mac Pro along with AMD not making a newer version of Apple's proprietary GPU card in the Mac Pro, is the reason for the lack of an update. I wouldn't be surprised if Apple is tossing that design out and going with something a little more conventional. Something like rectangular or cube-shaped case with 2-4 PCIE slots, but minus any kind of internal drive slots. They could do this in a way to still make it a lot smaller than an older tower Mac Pro. I currently use a Hackintosh, and would be more likely to buy an actual Mac Pro if the GPU was upgradeable. No upgradeable GPU = no Mac Pro for me.

I think for now Apple would be sticking with Intel CPU's and maybe switch to Nvidia GPU's or give an either or option. Later down the line they could switch to ARM-based when they become at least as powerful as Intel, and probably more powerful. That is unless Apple could drop something like 2 (4 core lite performance, 4 core high performance), 4 (8 core lite performance, 8 core high performance), or even 8 (16 core lite performance, 16 core high performance) A10 Fusions into a Mac Pro. But to do that this soon they'd need to be sure there's a way for all users to run non-App Store apps on it with just as good performance as on an Intel-based Mac Pro. They were pretty successful with this with the switch from PowerPC to Intel, so I don't see it being an issue. Hopefully Blackmagic Design and Adobe can get Resolve and Creative Cloud ported over to native ARM-based quickly.

With Intel's new focus on mobile I think there's a little more evidence of the switch to ARM. Intel probably realizes what's happening or about to happen with Apple's iPhone and iPad devices becoming as powerful as they are. The new A10 Fusion being as powerful as a Mac Pro in single core is pretty telling. And with Intel unable to really advance performance in their CPU's quickly they need to spread out into more areas.

I think the MacBook Air is going away. No reason for those anymore with the 12" MacBook being available. I do think it makes sense for these to be ARM-based much sooner than later. I wouldn't even be surprised if Arm-based CPU's showed up in the MacBook Pros as early as next year too. I think when Apple announces new computers next, the laptops will be just MacBooks and MacBook Pros.

However, the question is if Apple switches from Intel to their own ARM-based, what happens to Thunderbolt? Would Intel allow Apple to include Thunderbolt in their own CPU's?

 

[Jarek Zabczynski]

I agree with most of this. I don't think there's been a real reason for Apple to update the Mac Pro since the performance increase wouldn't be that great compared to what it is now. Intel has been slow to really improve performance from one CPU gen to the next, and who's going to buy a $10,000 Mac Pro one year and then spend another $10,000 on another Mac Pro a year or 2 later for a measly 10% performance increase at best.

I agree on those points. I think the people that are most upset are the ones that need a new machine now, or last year even. When people buy hardware they want and expect it to be current. Anyone forced to buying a MacPro right now is doing so knowing they are being screwed by buying old hardware. Updating the machine every year, even if the updates are minute, keeps the machines current all the time. A processor bump, and more importantly GPU upgrades would've gone a long way for Apple with the MacPro. If I could now have a nMP with a current processor and dual GTX1080 cards, I might not be complaining as much.

I'm plugging away with a decked out 2012 tower, but in another year I'll be wanting a new machine, I kinda want one now but Apple doesn't offer anything better than what I have.

 

[Wayne Morellini]

Silicon Photonics are needed where large storage systems are a long way from CPU nodes (minimising latency, and for supercomputers minimising transmission wattage).

+) http://www.nextplatform.com/2016/08/17/intel-leverages-chip-might-etch-photonics-future/

But low latency (Silicon Photonics) IMHO does Not solve on chip traffic congestion, ie the bottlenecks when multiple CPUs contending for data, and inter CPU communication is strangled during concurrent operations.* Intels Racetrack (or Ring) solution sort of scales with multiple circuits between the current number of cores - but this is rather two-dimensional.

Intel are (again IMHO) not keen for CPUs in High Performance computers to suddenly become twice as fast.* Intel's 10nm node has been ready in sufficient yield to build chips.* But these have not been offered - even at a premium, because Intel want everyone to buy chips that are only 5% - 15% faster year-on-year, and desire large companies to be in a 'perpetual evergreen' situation to maximise profit on selling hardware.* With no 'AMD' to compete - Intel have been holding back their 10nm chip electronics from the planet ... until their hand is forced.* Its a similar story with TB1, TB2 and TB3. Slow deliberate rollout of network speeds to avoid multiple systems from efficiently working together as one homogeneous unit.

Servers

Intel has been able to repel Cavium and Applied Micro from entering 'their' server world so far, as the Enterprise software is still being matured, and the highest performing chips designed by them have lowish IPC (instructions per clock) compared to Intels mature Xeon.* However the TC (Total Cost of Ownership) is still tipping towards Cavium under certain loads.

+) https://en.wikipedia.org/wiki/Comparison_of_ARMv8-A_cores

Cavium's 'Thunder X1' wimpy core solution using 2 x 48 core CPUs on (old) 28nm processes - is about 5 times slower than a top end Xeon.
Their next (Thunder X2) will still be 2.5 times slower.* Cavium will roll out ThunderX2 with less wimpy cores in 2017 - but even these are about 2.5 times less capable than an top end Xeon (per CPU)

+)* http://www.eetimes.com/document.asp?doc_id=1330492&image_number=1

Applied Micro's X-gene3 can scale up to 64 CPUs in a Chip - but will only reach about 60% the processing speed of a top end Xeon CPU.

ARM

http://www.anandtech.com/show/10711/arm-cmn-600-dmc-620-128-cores-8-channel-ddr4

+) With ARM's release of a 128 core x Coherent Fabric that can shunt 1 TByte / second around the 'mesh' - even wimpy cores that are 1/3 as capable as a tope end Xeon will be able to Scale to a performance at Intels Current Top end.
+) This fabric also has a Cache.. and is 5 x the performance of ARMs previous fabric (which was no slouch)
+) This cache will no doubt form part of CCIX initiatives (eg specialist hardware accelerators in which there are NO device drivers, and therefore virtually zero latency)

Apple
+) In Sierra .. Apple have purportedly changed the submission of code to the Mac App store from 'x86' (macOS) ... to a the stage before which is Operating System agnostic .. called 'bit-code'.
+) Apple have added a machine language type in Sierra that supports a non-ARM created (ie CPU designed by Apple) but ARM compatible.
+) Apple A10 Fusion chip is equivalent to Intels Xeons in IPC
+) Apple A10 Fusion 'big' core use less power at TSMC 16nm (Real world 18.3nm) than Intel's 14nm (real world 13.4nm).
+) Apples A10X (they might call it A11X) is on TSMC 10nm (Real world 11.3nm) is very likely to jump past Intel on performance per watt - but not necessarily absolute performance.* This seems like a perfect fit for the Mac Air. Less heat to get rid of + No money to pay to Intel + a Safe way to Prove MacOs on Apple ARM .. and for all software submitted in the Max App store .. automatic recompilation at Apple HQ.
+) Apple have a hardware version of ARMs big.Little context switch (ARMs design has an achilles heel in that the OS control the switch).*

TSMC had advised that they will be earning money from 10nm node in 2017 Q1 .. and that their 7nm node (real world 8.2nm) will be ready for sample testing later in the year.

+) Mac Air : (speculation) Apple will develop an ARM v8 compatible Mac Air .. to be on sale in 2017 H1,* and that ..
+) Mac Pro : (speculation) Apple will develop either a 10nm (64 core) or 7nm (128 core) CPU Mac Pro .. and that Nvdia GPUs will adorn it, supporting Metal (great for the OS), CUDA (for stand alone Apps that need it), and OPEN CL/GL (for Apps that have not yet moved to Metal).** If a 128 core Apple CPU chip costs Apple $500 at source and will meet your CPU needs for the until Intel 'Catch up' with Apple, and IF 2 x VOLTA GPUs meet our 8K needs ... I think this could be a winning formula.

AJ

Also to break large bandwidth off chip, as the number of lanes increase too much to handle, particularly with increased slots, USB3.x and TB ports.* Currently all these are being held back due to a lack of lanes to deal with all of them.

Isn't there some talk of AMD getting early access to low power 14-7nm through tsmc or somebody? I think Intel knows they are stuffed in competing against GP-GPU and specialist processing hardware, so they might as well just let people use those inbtje server soave until they can get Phi better, and as the regular processor replacement.

I think the issue about the can is irrelevant, it is what it is, but they require higher powered solutions. My post before about a bigger can that can take multiple/regular cards is relevant, and more CPU's. However, the expandable blade posts ultimately allow a higher end compact workstation farm. Instead they are stuck at a regular desktop publishing, photo editing, and pro video level, hardly workstation worthy. However, I agree about the custom card, custom driver issue. There are so many books and crannies in these things now days, driver updates are numerous, so it pays to have a common design (but two-foyr of them).

 

[Wayne Morellini]

Somebody was talking about native code versions of x86 programs for Apple Arm. It's not strictly needed, there are allready code that translate x86 to ARM. My Nvidia sheild TV is around half or something of the Xbox one, but maybe a tenth of the size. So, how much more processing can fit in the same volume and still be in the x86 xbox one power envelope. Apple is not the only high performance low energy ARM designer. Waiting to see the next two generations.

So, yes, you should he able to get descent x86 performance on ARM systems with the right spec.

Take note, after 7nm there is not Mich of a path, as power per cycle increases. So at 7nm I'm making plans to buy, and wait for something new.

 

[Frank Glencairn]

, and who's going to buy a $10,000 Mac Pro one year and then spend another $10,000 on another Mac Pro a year or 2 later for a measly 10% performance increase at best.

Sounds totally like something, the typical Apple user would do ;-)

..and has done so in the past.

 

[Mike Halper]

I agree on those points. I think the people that are most upset are the ones that need a new machine now, or last year even. When people buy hardware they want and expect it to be current. Anyone forced to buying a MacPro right now is doing so knowing they are being screwed by buying old hardware. Updating the machine every year, even if the updates are minute, keeps the machines current all the time. A processor bump, and more importantly GPU upgrades would've gone a long way for Apple with the MacPro. If I could now have a nMP with a current processor and dual GTX1080 cards, I might not be complaining as much.

Absolutely true. The current Mac Pros really aren't much faster than the tower Mac Pros. It would be a minute upgrade aside from getting Thunderbolt 2 and USB 3. But you can add USB3 to a tower, and how many people actually use and need Thunderbolt? I was able to add anything I'd need Thunderbolt for to my tower with PCIE cards, and prefer that than a bunch of cables and boxes sitting on the desk. The current Mac Pro could actually be slower in some cases considering you can put a newer GPU in the towers. Intel just hasn't been able to advance their CPUs well enough. Apple appears to have shot themselves in the foot with this new design. Hopefully they're going away from it and that's why there hasn't been any spec bumps in 3 years. If they go back to normal PCIE slots then they could easily meet most users' need and wants.