MP 7,1 Waiting for Mac Pro 7,1 A1991 (no more)

[Mago]

The best thing from for the iMac pro for the mMP is having the memory unaccessible (not 'user' upgradeable), coz having the iMac Pro DIY friendly internals it would tax the mMP sales at the low range (despite Apple interest in their CTO Tax) making most low range Pro's (DIY upg enthusiast) wait for something they can upgrade/taylor with Apple's blessing, else they will be happy with Basic iMac Pro (with Vega 64 GPU) adding RAM and Upgrading CPU as they like not caring of the mMP.

But as the mMP makes presense, the iMac Pro (a temporal 'patch' product IMHO there wont be a 2,1 iMac Pro) will appeal only to those enaged with the AIO form factor,a minuscle user base in the Pro Crowd, also later the mMP configured with non-Apple not TB3 display (cheap 4K displays) will take the natural non-AIO picky iMac Pro Market (those pro that dont care on the Form Factor but on having the machine they need to dotheir work).

So I suggest to the Apple Collectors to buy a 13,000.00$ iMac Pro and file to sell in 20 years from now, as the tcMP it will be a one-off edition along the Cube iMac and the 20th Anniversary Mac Pro.

 

[beaker7]

If they made a good dual Xeon SP tower, with 2-3 x16 slots for GPU, NVME, 10Gb, and a few TB3, for $15-25k, I’d buy a pallet of them. That’s what we pay for Dell and HP now, since Apple stopped accepting our money. It’s really not that hard.

 

[Martyimac]

Interestingly, the iMac Pro has been thoroughly adopted by the iMac forum. Rightly so in my opinion.

I spend lots of time on the iMac forum... That forum has not THOUROUGHLY adopted by the iMac forum. I would even venture to say that over 50% of the posters have issues with the iMac Pro for many reasons. Those include, price, throttling, potential heat issues, no simple upgrade path, AIO issues, and there may be other reasons.

 

[mkjj]

Boom!

 

[fuchsdh]

Intel and AMD has split off single and 2+ socket development. Those markets have some widening gaps that customers are driving two distinct pools. The Z840 is aimed at the 2 (and up) crowd. Apple is probably done with the 2+ ( the 1 socket had the driving volume all along). Folks could more heavily lobby for 8 DIMM slots in a new Mac Pro but greater than 1 socket is probably a lost cause.

I'm a bit confused by that part…

The Wikipedia list for Xeons says -W processors support up to 8xDIMMs, but the corresponding entry for individual processors says 4x memory channels, as do the ark.intel pages: https://ark.intel.com/products/125038

And these are 1S setups, so I have no idea where the 8 DIMM stuff comes from. Using just the 4 channels for 2 DIMMs a piece?

I spend lots of time on the iMac forum... That forum has not THOUROUGHLY adopted by the iMac forum. I would even venture to say that over 50% of the posters have issues with the iMac Pro for many reasons. Those include, price, throttling, potential heat issues, no simple upgrade path, AIO issues, and there may be other reasons.

I don't really get that though if you're an iMac customer. If you can't afford an iMac Pro, you can't afford an iMac Pro. But no easy upgrades, potential heat issues, AIO issues—those exist buying a regular iMac. The iMac Pro is just "hey pay more money and get better innards", it's not a change in any calculus.

Intel and AMD has split off single and 2+ socket development. Those markets have some widening gaps that customers are driving two distinct pools. The Z840 is aimed at the 2 (and up) crowd. Apple is probably done with the 2+ ( the 1 socket had the driving volume all along). Folks could more heavily lobby for 8 DIMM slots in a new Mac Pro but greater than 1 socket is probably a lost cause.

While not an xMac , the Mac Mini going vertical has decent possibility. The desktop footprint wouldn't change much. ( the Mac Pro has a smaller one. 6.6' radial as compared to the 7.7" squared for the Mac mini. Vertical could be slightly smaller but use more vertical space on the desk.


An mini xMac that is primarily aimed at being a iMac fratricide device is probably definitely a no go. However, a BTO option of a vertical Mac Mini would probably overlap with some of the lower half of the 21.5" iMac range. Nothing in the 27" range though. Those BTO Mac Mini could creep into the $1500 range ( although probably driven by large SSD ).


xMac as a headless mid-upper range iMac? No. Way too much fratricide potential there with relatively highly price sensitive buyers. However, a headless/monitor-less and keyboard/touch less entry level MBP 15" possibly.


In the $4999 and up zone of the iMac Pro there isn't a much of the highly price sensitive fratricide potential. Folks have money but more highly specialized needs. So more so buying what they need to fit their market rather than primarily just "cheaper".

The Mac mini getting turned into a taller box versus getting shrunk into a NUC is interesting. It's another area where the Mac mini is so many things to so many people that any deviation from its current form is going to dismay some users.

 

[ManuelGomes]

4 channels, 2 DIMMs per channel. Hence 8 DIMMs
[doublepost=1513701550][/doublepost]I highly doubt Apple will go back to a tower design at this point.
It will be similar in concept to the nMP.

 

[mattspace]

I highly doubt Apple will go back to a tower design at this point.
It will be similar in concept to the nMP.

Here’s the thing, Apple devoted a big part of the iMac Pro launch to VR, according to Panzarrino and Gruber on the latest talk show. The only credible way to package a VR development macine is to have x16 connected GPUs that can be upgraded as fast as the GPU makers update them, because that’s the standard of bandwidth and pace of change the consumers of VR software / solutions will be buying / upgrading their (overwhelmingly PC-based SteamVR) systems.

GPUs for VR developers are like iOS devices for iOS developers - you HAVE to stay up to date with the hardware you’re going to be deploying on, and no one in their right mind is going to suggest an annual purchase cycle for a $5k+ machine.

 

[JesperA]

Just give me a Mac Pro with custom Intel chips in a big.LITTLE configuration, 4-6 cores at 4.5ghz and 10+ cores at 2.5-3ghz

 

[deconstruct60]

I'm a bit confused by that part…

The Wikipedia list for Xeons says -W processors support up to 8xDIMMs, but the corresponding entry for individual processors says 4x memory channels, as do the ark.intel pages: https://ark.intel.com/products/125038

And these are 1S setups, so I have no idea where the 8 DIMM stuff comes from. Using just the 4 channels for 2 DIMMs a piece?

A memory channel is a single bus that DIMMs are connected to. If look at the current iMac 27" processors page will see two channels (since the mainstream processor design) so those machines max out at 64GB. The number of DIMM sockets sitting on a channel is typically called rank ( so can have 1 , 2 , or sometimes 3 or 4 ) all sitting on one controller.

You can kind of indirectly see it on the summary ark summary pages in the ratio between absolute maximum RAM supported and number of memory channels. If DIMMS top out at 32GB ( or previous years 16GB) you to have to have more to get to the very high multiple.

So for example the i5-7500 in a mid range iMac 27". https://ark.intel.com/products/97123/Intel-Core-i5-7500-Processor-6M-Cache-up-to-3_80-GHz .
Two channels and maxes out at 64GB ( using 16GB DIMMs ).

whereas the W 2145 https://ark.intel.com/products/126707/Intel-Xeon--W-2145-Processor-11M-Cache-3_70-GHz
Four channels and maxes out at 512GB. ( that is either a limit rank of 3 for 32GB or limited to 2 but there is a mapping for some 64GB DIMMs in there which don't really exist yet. Probably is 8 DIMMs as outlined. )

A Xeon Gold 6144 (similar 8 core as the 2145) https://ark.intel.com/products/124943/Intel-Xeon-Gold-6144-Processor-24_75M-Cache-3_50-GHz
Four channels and maxes out at 768GB ( 4 x 3 x 64GB = 768GB ) .

The Mac mini getting turned into a taller box versus getting shrunk into a NUC is interesting. It's another area where the Mac mini is so many things to so many people that any deviation from its current form is going to dismay some users.

Conceptually, it is orienting the box ( literally turning it ) more so that pushing it out of the NUC category. It still would be about NUC sized in volume ( a bit bigger, but certainly would run cooler). It is just whether the long side is down touching the desk or sitting up in the air. If trying to keep it very cool and quiet (with active cooling) tilting it up and put as large a fan as can get away with (but still smaller than footprint) fan on top probably works better.

To get to the larger fan the Mini sits on top of it. And then the flow is turned 90 degrees to go out. If instead you just flow with the natural flow then not at odds with super sharp turns. You should be able to crank the RPMs down a notch since not swimming in "wrong" direction.

If made it more like rounded corner rectangle folks could still place it long edge down (for like rack/shelf placement ) and cooling would go "old bottom " to "old top" (but essentially be front-to-back). Widening the base ( height in "old" orientation) menas there is even less pressure to get rid of the 2.5" SATA drive. ( Apple is a bit conflicted they want everyone on Flash storage but also charge much higher than average for Flash storage. High $/GB pushes folks on a tighter budgets to HDD with have much better $/GB stats. If they want folks off so bad bing the Flash $/GB price down. ).

That would be waaaaay better than simply just trying to make the mini an even flatter pancake ( slavishly following the every thinner Mac laptop trend line. ). Following the laptops for internal is one thing. Taking on all of their physical trends is misguided. 0.02lbs lighter sitting on a desktop is not making a difference. neither is 1.1-1.2 versus 1.4 inches tall.
[doublepost=1513717200][/doublepost]

Just give me a Mac Pro with custom Intel chips in a big.LITTLE configuration, 4-6 cores at 4.5ghz and 10+ cores at 2.5-3ghz

Pragmatically already do that with the desktop processor design. Only it is the iGPU that is the "max core count at much lower GHz".

Intel could conceptually do that as they move the high core count processors to EMIB , but it wouldn't be one die.

https://www.anandtech.com/show/1174...-stratix-10-fpga-live-blog-845am-pt-345pm-utc


And I wouldn't hold my breadth. I doubt that would work even with separate dies in the same package at the power levels 50-60+ W power levels talking about here. The heat from the 4.5GHz stuff is going to leak over and vice versa it all bundled in the same package.

Big.Little makes sense in devices that spend at least half the day in sleep (or non interactive ) mode. That isn't the Intel W class mainstream workload.

 

[wallysb01]

whereas the W 2145 https://ark.intel.com/products/126707/Intel-Xeon--W-2145-Processor-11M-Cache-3_70-GHz
Four channels and maxes out at 512GB. ( that is either a limit rank of 3 for 32GB or limited to 2 but there is a mapping for some 64GB DIMMs in there which don't really exist yet. Probably is 8 DIMMs as outlined. )

A Xeon Gold 6144 (similar 8 core as the 2145) https://ark.intel.com/products/124943/Intel-Xeon-Gold-6144-Processor-24_75M-Cache-3_50-GHz
Four channels and maxes out at 768GB ( 4 x 3 x 64GB = 768GB ) .

Ark has the XEON SP series as all 6 channel memory.

 

[askunk]

Vega 20 (running on PCI-e 4.0) should arrive by Q3 Q4 2018. Could Q1 2019 be a probable window for the mMP?

 

[beaker7]

Vega 20 (running on PCI-e 4.0) should arrive by Q3 Q4 2018. Could Q1 2019 be a probable window for the mMP?

Lol

Why not? Should just push to 2023. Make it an even decade between updates.

 

[askunk]

lol right. Why not wait for Vega 200, instead of compromising 😀

 

[Flint Ironstag]

Nearly 300 pages. All I can say is that if they were stuffing things into a plain box, we wouldn't still be waiting. Wonder if it'll be anticlimactic?

 

[Jaho101]

Just give me a Mac Pro with custom Intel chips in a big.LITTLE configuration, 4-6 cores at 4.5ghz and 10+ cores at 2.5-3ghz

I feel like that arrangement makes sense when you're running off a battery, but the server chips inside workstations are designed for pretty constant throughput while being power efficient, not battery saving.

 

[AidenShaw]

The only credible way to package a VR development macine is to have x16 connected GPUs that can be upgraded as fast as the GPU makers update them, because that’s the standard of bandwidth and pace of change the consumers of VR software / solutions will be buying / upgrading their (overwhelmingly PC-based SteamVR) systems.

This.

Today (meaning currently) posters here have said that GTX 1080 Ti is just above "entry level" for VR. When Volta hits the top mainstream "GTX xx80 Ti" market - suddenly those 1080 Ti cards will be obsolete. This will happen for Volta and the next generation - as VR apps ramp up companies aren't going to focus on old Pascal cards. It will be focused on Volta - with a nod to Pascal saying "1080 Ti at a max of 8 FPS".

If Apple makes an mMP that is stuck for 5 years with proprietary Vega cards - they'll be kissing the VR market goodbye. It needs PCIe 3.0 x16 slots that will support whatever Nvidia and ATI announce next spring.

VR will consume every bit of GPU power that is available - and if Apple doesn't realize that and support generic dual-slot PCIe 3.0 x16 GPUs the mMP will be "Cube 3.0".
[doublepost=1513736147][/doublepost]

Wonder if it'll be anticlimactic?

It will be climactic - the mMP announcement will be like:

"I felt a great disturbance in the Force, as if millions of voices suddenly cried out in terror and were suddenly silenced."

if the mMP is "Cube 3.0".

 

[namethisfile]

Nearly 300 pages. All I can say is that if they were stuffing things into a plain box, we wouldn't still be waiting. Wonder if it'll be anticlimactic?

It won't if it looks anything like this!!!

 

[mattspace]

VR will consume every bit of GPU power that is available - and if Apple doesn't realize that and support generic dual-slot PCIe 3.0 x16 GPUs the mMP will be "Cube 3.0".

My most optimistic thought is that Apple got blackswanned by VR, and by the time they realised it was a Product and a Tool that people were already being paid to do work in, that already had an ecosystem in SteamVR in which the operating system is relegated to a glorified BIOS, the iMac Pro was a locked down product and strategy, and they just have to plough ahead, and make the best of the situation by presenting it as VR Capable.

Makes a freaking mockery of their claims at WWDC that the 580 Pro regular iMac was "VR Ready" when one of the major apps they were demoing at the iMac Pro launch, Gravity Sketch, is Vega64 minimum in macOS - which rules out the lower config GPU iMac Pro.

It wouldn't surprise me if someone in marketing has been conflating 360 Video with VR.

 

[deconstruct60]

Ark has the XEON SP series as all 6 channel memory.

I'm sorry ( i added that last example quickly and didn't double check). Thanks for catching that. The older Xeon E5 26xx went rank three. To get better bandwidth they've capped all the new stuff at rank 2 and bumped the number of channels. So probably 6 x 2 x 64GB = 768GB. ( and with two CPU packages can push into the 1TB RAM range if have a wheelbarrow of money to spend. )


Back in Mac Pro context, not only are the Xeon XP sockets a larger footprint, but the DIMM layout will tend to consume more space also. Bigger infrastructure leads to higher entry price. Populating 6 DIMMs (usual Mac Pro class modus operandi for Apple ) pushes up entry price also. [ and yes the entry 2013 MP only had 3 which is a deviation but also points out there are there is not perfect price elasticity for the Mac Pro. .... too much and more than a few won't pay. ]

 

[flat five]

Nearly 300 pages. All I can say is that if they were stuffing things into a plain box, we wouldn't still be waiting. Wonder if it'll be anticlimactic?

since j.Ive is now back managing the design team, maybe they somewhat re-started again since his return.

(fwiw, the last pro desktop design Ive was largely responsible for was G5/cMP).
[doublepost=1513747022][/doublepost]

It won't if it looks anything like this!!!

View attachment 742821

uh oh..
got the straight edge out.
nice
🙂

 

[deconstruct60]

Vega 20 (running on PCI-e 4.0) should arrive by Q3 Q4 2018. Could Q1 2019 be a probable window for the mMP?

PCI-e v4.0 , Not likely. That diagram missing the realities of AMD's track record of late. A 2015-2016 roadmap would have put Vega 10 in Q2. it didn't. This diagram reflection the adjust of reality but is a bit vague. The 'release' was Q3 but the reality of substantive shipments for Vega was Q4. So really the right hand side of those card icons (not the middle). While PCI-e v3 passed 2010 it wasn't until very late in 2011 that testing/validation stuff settled down so that real volume didn't appear until 2012 ( was part of the slide of Xeon E5 v1 series. ). There may be some highly limited distribution, early access Xeon by late 2018 that do PCI-e v4 but that will be the Google/Amazon/Microsoft/FB cloud folks and the super computer vendors. Not Apple. ( Google and that crowd had the Xeon SP for 6-7 months before anyone else. Those folks with data center AI stuff will get first crack at PCI-e v4. And folks buying early IBM Power9 systems. )


Vega 20 is likely to be a 14nm+ process bump. Better optimization, better overclocking characteristics. The iMac Pro will need it more than the Mac Pro would if Apple gives the Mac Pro a bigger thermal budget to work with ( for iMac Pro it is a better fit under the 500W cap). ( even if AMD but nominal support into the card for v4.0 if the CPU package doesn't have it then not really a differentiator. Intel's 2018 bump is a "coffee lake" type bump to 14nm++ ( or 14nm+++). very likely largely just better clocks for them too. )

If primarily a clock bump then getting drivers onto macOS probably would insert the typical delay from Windows (and some Linux) moves. Apple is in for a world of PR hurt if they get to the 5th anniversary of Mac Pro 2013 release. ( https://www.macrumors.com/2017/12/19/trash-can-mac-pro-is-four-years-old/)
If the Mac Pro and the iMac Pro share basically same core major drivers and major subcomponents (but at different feeds and speeds ) then should be able to 'walk and chew gum at the same time' and do both but mid-late December 2018.
(presuming Intel and AMD deliver on time in 2018. )

 

[iPadified]

I spend lots of time on the iMac forum... That forum has not THOUROUGHLY adopted by the iMac forum. I would even venture to say that over 50% of the posters have issues with the iMac Pro for many reasons. Those include, price, throttling, potential heat issues, no simple upgrade path, AIO issues, and there may be other reasons.

I am afraid you missed my point. The iMac Pro belongs to the iMac line not the MP line, - two different uses, two different customers. I read iMac forum as well and sure there are the usual naysayers about AIO but there are also lots of people who think it is a good/perfect machine. I have not counted by my impression is that there is not 50% complainers.

How the complainers can discuss performance without any comprehensive stress tests and benchmarking is beyond my understanding.

 

[JesperA]

I feel like that arrangement makes sense when you're running off a battery, but the server chips inside workstations are designed for pretty constant throughput while being power efficient, not battery saving.

Such a configuration in a desktop has nothing to do with power-efficiency, its about performance. There is still tons of apps and OS tasks/processes that benefits from fever (4-6 cores) but faster cores and then there is apps/tasks that is heavily parallelized where you would benefit from many but slower cores. So for us that are constantly working with a mixed workload on a daily basis, we had to compromise what workload we wanted to optimize our hardware for, if we go with maximum core count (12+ cores) then our 4 core (or less) optimized workload would suffer and vice versa.

Turbo boost is one solution but it has alot of drawbacks, even if we had a 10+ core CPU that could turbo boost to 4ghz+ on 2-4 cores, in practice, its very few times it would actually boost up to those speeds and some processes are very short lived so it would never turbo boost that high anyway and for me that do alot of batch/background processing while still working on my main task it would basically never turbo boost.

So my proposed configuration would be a really fast responsive system (most tasks on fever but faster cores) but still be a beast at batch processing and/or parallell workloads.

Like one other user mention, heatsoak would be an issue maybe but one solution could be a 2 socket asymmetrical system, one socket with fever but faster cores and the other socket with a 8176 CPU🙂)), not sure if that would work though since it couldn't share CPU cache and a standalone controller would be needed since QPI/UPI probably don't support asymmetrical socket configs.

But my proposal was just a fantasy wish, it would be awesome for me but very slim chance that would ever happen.

 

[singhs.apps]

Such a configuration in a desktop has nothing to do with power-efficiency, its about performance. There is still tons of apps and OS tasks/processes that benefits from fever (4-6 cores) but faster cores and then there is apps/tasks that is heavily parallelized where you would benefit from many but slower cores. So for us that are constantly working with a mixed workload on a daily basis, we had to compromise what workload we wanted to optimize our hardware for, if we go with maximum core count (12+ cores) then our 4 core (or less) optimized workload would suffer and vice versa.

Turbo boost is one solution but it has alot of drawbacks, even if we had a 10+ core CPU that could turbo boost to 4ghz+ on 2-4 cores, in practice, its very few times it would actually boost up to those speeds and some processes are very short lived so it would never turbo boost that high anyway and for me that do alot of batch/background processing while still working on my main task it would basically never turbo boost.

So my proposed configuration would be a really fast responsive system (most tasks on fever but faster cores) but still be a beast at batch processing and/or parallell workloads.

Like one other user mention, heatsoak would be an issue maybe but one solution could be a 2 socket asymmetrical system, one socket with fever but faster cores and the other socket with a 8176 CPU🙂)), not sure if that would work though since it couldn't share CPU cache and a standalone controller would be needed since QPI/UPI probably don't support asymmetrical socket configs.

But my proposal was just a fantasy wish, it would be awesome for me but very slim chance that would ever happen.

This. A perfect compromise between faster but fewer cores vs slower but many cores is a dual socket system : 8-12 cores per socket x2 doubles the potential speeds+performance . It also has the added benefit of doubling the PCIe lanes + ram capacity.

The hardware requirements for today and the near future has never been higher since about 10 years ago when it started to draw level with the requirements of the then workflow.

VR won’t really hit their stride until we start seeing 16k res @ 60 FPS per channel at a minimum. Then there is the added expectations of sound ( feasible on today’s hardware ) + touch and (potentially smell ), last two yet to be addressed . In other words engage all the senses.

And no 350 video isn’t VR no matter which way the marketing wants to spin it, but imo it’s where most vendors will want to support. It’s a much easier target.

AR is a lower hanging fruit but also has larger potential to make difference in many areas. I suspect Apple will want to target the latter first.

 

[Mago]

It won't if it looks anything like this!!!

View attachment 742821

I buy your concept, with few corrections:

Fan On Top (ala Trash Can, for rightness thermodynamics), Wider CPU/RAM/Storage/I-O Block

And about the Specifications:

CPU: AMD Epyc SP 8-32 core count.
Memory: 12 Slot, upto 384GB By Apple CTO
Storage: 4x NVMe Upto 16TB ssd Raid 10,1,0
GPU: 1 or 2 Cartridge GPU+PSU AMD or nVidia, from Single RX570 to dual nVidia GV100 upto 400W TDP/Cart/
I/O 4xUSB3.2 6x TB3, 2x HDMI2,1, 2x NbaseT
Expansion: Available General Purpose Expansion Cartridge with 150W PSU + x16 PCIe (std) Slot Half Length.

Price from 2,500$ (1TB+8c+Rx570) to 30,000$ (16TB+32C+dual nVidia GV100)

Add this to the Signature: More Innovation: MY ASS!!! p.s.
Name it: m³ Mac Pro

PD. I'm In Love with this concept.

Possible variations:
three linear horizontal volume mid-Tower with Horizontal Cooling Flow from front==>>backward
each top volume a cartdrige

||_____________________________||
fan2+PSU 2 | GPU 1 or GP PCIe Slot
fan1+PSU 1 | GPU 0_____________
fan0+PSU 0 | I/O|Storage|Ram| CPU
||================================||