Ahem. SP3 and SP4 are the socket names for Naples chips
.I do not see any FP64 requirement in software on Macs. But I do see a lot of FP16 emphasize in Metal.
The iMac's are already said to have Xeon E3v6 CPUs. As for GPUs, we shall see.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
MP 7,1 Waiting for Mac Pro 7,1 A1991 (no more)
- Thread starter Mago
- Start date
-
- Tags
- macpro 2016
- Sort by reaction score
.I do not see any FP64 requirement in software on Macs. But I do see a lot of FP16 emphasize in Metal.
The iMac's are already said to have Xeon E3v6 CPUs. As for GPUs, we shall see.
Fixed it for you.
What you'll see is that some of the older instruction extensions will be dropped.
See Intel working on a completely new x86 implementation dropping SIMD compatibility, for example.
I'll rephrase: Fingers are being pointed at AMD. But also now the trash can design.
to quote a source this week:
(edit: hah, the forums asterisk profanity - it was a 4 letter word for excrement.)
make of that what you will. Personally, I'd be very surprised if AMD got to keep their exclusive in the new machine. They simply can't deliver the goods (oh sure, the next version, that'll be the good one, like bluetooth), and Pro computers, as used in the real world, are now effectively life support systems for Nvidia GPUs.
(edit: hah, the forums asterisk profanity - it was a 4 letter word for excrement.)
make of that what you will. Personally, I'd be very surprised if AMD got to keep their exclusive in the new machine. They simply can't deliver the goods (oh sure, the next version, that'll be the good one, like bluetooth), and Pro computers, as used in the real world, are now effectively life support systems for Nvidia GPUs.
Zero need there if iMacs go to Xeon E3. That is the same performance class curve as the current i7-x7xx class stuff used now. They'll probably be looking at a 1S (socket) solution. Intel is going to have a workstation class product offering (Skylake-W). Why Apple wouldn't pick that would be beyond odd. Especially if looking to pick up the pace of updates. The coupled desktop PCH drived chipset will move at faster pace. The -W solution also may get updates a bit quicker than the 2-4 socket solutions.
There would be some overlap (and maybe underlap) in the 4 core zone, but once past that .... it isn't. In fact there are some leak roadmap charts that put a that E3 type x86 core package minus the iGPU into the Skylake-W socket. In that case even closer to overlap at the 4 core offering price point.
If Apple just started and told to do something original.... that probably isn't going to happen in the first half. If no show stoppers pop up shouldn't run into 2019. Could be like 2013 where they are doing it on the last day of Fall (Dec 21) and most shipments slide into 2019.
[doublepost=1491635158][/doublepost]
From the comments Apple made in the session with the reporters/bloggers it is modular flexibility they wont. So the cooling will probably move a step toward de-intergration. (each GPU module gets a major cooling design of its own with some constraints ). That doesn't necessarily mean highly expandable.
The intent is not to paint themselves into a corner. So if GPU solutions go +/- 40-50W off some baseline target they can still get a new card constructed and out the door on a regular basis. Also may be able to contract out the work on the card easily. ( give folks a reference board and some thermal power targets. Subcontractor designs solutions to that and have alot less coupling than to cooler than other folks also have to couple to. )
Modular doesn't necessarily mean commodity, off the shelf parts. But it should allow more loosely coupled teams to do work in parallel. That creates flexibility in both time to market and serviceability/upgrades.
highly expandable seems imply high number of modules. There was little in Apple's comments about the market research that stressed that.
Last edited:
I'd question whether anyone could actually make a financial case for designing and / or fabricating something that is both price and performance competitive with an off-the-shelf Nvidia card, that would be profitable when selling to a market that is, according to Apple, 1% of the Mac's install base - and most of that install base would be older cMPs.
An over-saturated market of redundant suppliers no.
However, Mac Pro GPU cards never were the same price before. Why would they be now?
Like the "Pro" graphics cards in the Windows market the run rate is lower and the average cost is higher. Frankly, it does and is working because that is where Nvidia (and AMD) makes a higher percentage chunk of its profits. The race-to-the-bottom, high volume cards are not really the healthier part of the discrete card market. Volume for volume sake doesn't necessarily mean profits.
Well hopefully the fact that Nvidia are now back to writing Mac drivers for desktop GPUs maybe AMD will do the same. If not I’ll switch.
I hope apple will give the users a broader selection of CPUs and GPUs with the coming Macs. THat's the only way I would consider switching back for my work machine.
Xeon CPUs are great for heavy multithreaded tasks. Sadly a lot of Mac users are creative people using the Adobe package, which only suffer from a magnitude of low clockspeed cores because of the bad support for multiple cores in Adobe programs. While a lot of creative people also do 3d, which benefit greatly from a lot of threads. I decided to go with Broadwell-E in my PC, which is a bit of both worlds. I got 6 cores of 4.4Ghz, so I still get good performance in Adobe programs but the machine is powerful enough for rendering fairly fast. Of course I wish I could double up the cores, but not if that would sacrifice clockspeed. Apple seem bound to stick with xeon processors for pros and the lake serie for consumers - which I think is wrong. ...because a lot of people buying the Mac Pro will work in the adobe CC programs most of the time, which then makes a pro machine with Xeons a really bad investment.
Same goes for GPUs. Dual GPUs are just plain stupid in a pro machine except for Final Cut Pro, you get no use for it. Most places you will benefit from dual GPus are games, ironically. So, letting Pros choose this option or if they want to use Nvidia or AMD would also be good. Its a pro machine after all, so making the choices a bit more complicated shouldn't alienate customers.
Xeon CPUs are great for heavy multithreaded tasks. Sadly a lot of Mac users are creative people using the Adobe package, which only suffer from a magnitude of low clockspeed cores because of the bad support for multiple cores in Adobe programs. While a lot of creative people also do 3d, which benefit greatly from a lot of threads. I decided to go with Broadwell-E in my PC, which is a bit of both worlds. I got 6 cores of 4.4Ghz, so I still get good performance in Adobe programs but the machine is powerful enough for rendering fairly fast. Of course I wish I could double up the cores, but not if that would sacrifice clockspeed. Apple seem bound to stick with xeon processors for pros and the lake serie for consumers - which I think is wrong. ...because a lot of people buying the Mac Pro will work in the adobe CC programs most of the time, which then makes a pro machine with Xeons a really bad investment.
Same goes for GPUs. Dual GPUs are just plain stupid in a pro machine except for Final Cut Pro, you get no use for it. Most places you will benefit from dual GPus are games, ironically. So, letting Pros choose this option or if they want to use Nvidia or AMD would also be good. Its a pro machine after all, so making the choices a bit more complicated shouldn't alienate customers.
But maybe Apple has promised to sign the driver to ease up the anger of the developers / Pro users? Nvidia could release a card for MP 5.1 and a TB2 (dual connector) extension box for nMP. We can dream, right? But will see.
Anyway, Apple's war against Nvidia could be over by now. They've had years to convert Cuda programmers to openCL and Metal, and soon Metal should be mature enough to compete with Cuda. Perhaps with macOS 10.13 already.
Yet again, we'll see.
Alot of my workload is kinda similar, 70% of my work benefits from higher clock speeds and fewer cores but the rest of my workload would benefit from alot of cores, so i have a mixed workload. I have fantasized about a x86 big/little configuration for a while, i think that would be awesome, a Xeon with 6 fast cores (+4-4.5Ghz) and when those get saturated a block of many but slower cores would activate, 12-16 cores at 2.7-3Ghz. That way one could get a monster computer for all the workloads, both in applications that cant spread its load on alot of cores, maybe maxing out at 4 cores, and with applications that can make use of as many threads as you throw at it.
Currently its either or, you can choose fast but few cores or many but slow cores, but this way one could get both if a co-processor isn't an option
Yes, well Broadwell E is the inbetween CPUs. But they dont exist in the mac universe. But they should, because a lot of Mac users use Adobe software. I actually think that for the majority of Mac Pro users Broadwell E is the more right fit than Xeons.
Hopefully some day in the future. Intel or amd understands that this is the case for many users. THe iphone 7 has a quad core system, were 2 cores are high clock speed and two low, for battery optimization. WHy cant anyone invent such a CPU system for the pc not meant for power optimzation but for performance usage. So you have 2 super high clock speed processors and 8 more in low, would be awesome.
Maybe the bigger "Mac Mini" in the pipeline is actually Mac Pro SE? And it's just a modded Trash bin with new thermal core and one cpu and one gpu? If so, hopefully real desktop parts...
Last edited:
The Mac mini, I think, will keep occupying the lower range. It is a versatile, cheap and energy saving server for small businesses and small professionals, other than just a low-range Mac for switchers.
I think that Apple may be talking about it as a useful addendum to Pro stations for Pro users, therefore they will offer an SSD, Polaris and TB3 option that will suit an integration in Pro environments, not that it will evolve much towards the nMP form factor and quality level.
[doublepost=1491660473][/doublepost]Does anybody know the actual roadmap for Intel in 2018?
I got everything said so far about Skylake E5 or the W version. I am pretty sure Apple will stick with Xeons E5 class CPUs. However, if the Skylake version comes up within H2 or even Q3 2017, wouldn't it appear as an "old" CPU when Apple will present the mMP in H2 2018....hopefully, not in Q1 2018?
That's why I'm keen on the AMD option. Naples works easily in a 1S MB, it has more memory channels than Xeons, therefore allowing for a really powerful machine, the way we like it
I'm thinking:
AMD Naples 16/24/32 cores
(or... I guess, Skylake E5 6/8/12 cores?)
2 or even up to 4 NVMe SSDs in RAID 0, offering 5 to 10 GBps transfer rates (user replaceable) Each SSD comes in 256/512/1/2 TB. Minimum config: 2 x 256 GB RAID0. It may be even cheaper than a 1x 512... but faster, wouldn't it?
2 or 3 PCI-E 3.0 slots (let me hope for a future ready 1x PCI-E 4 slot!!): 1 with a GPU, the other available for additional GPU, Xeon Phi cards (btw... would they work with AMD CPUs?) or dedicated vertical expansion cards.
Ports:
6 TB3/USBC 3.1 Gen 2 (up to 4 4K/5K @60Hz/120Hz monitors or 2 8K @60Hz via Displayport 1.3?) (I may be quite wrong about frequencies and DP versions.. help me out here )
4 USBA 3.0
1 HDMI 2.0 (up to 8K?)
1 audio in (please bring back the optical!)
1 audio out (again... optical too!)
BT 5.0
802.11 ac/ad Wifi
1/2 10 GbE
Extended wireless/USB Keyboard with touchbar and touchID
A few more ..requests:
* frontal ports! (at least a few TB3/USBC)
* easy to open, easy to clean, easy to upgrade (SSDs, CPUs, GPUs/cards)
* this is quite a stretch, I know, but I would love to have a small LCD screen with CPU/GPU/Network stats on the body
GPUs... maybe we should discuss them in June when we will know what AMD is up to and where nVidia is going for 2018...
What did I forget? Ah... any suggestions for the TBD? Which ports do you think it will have?
I think that Apple may be talking about it as a useful addendum to Pro stations for Pro users, therefore they will offer an SSD, Polaris and TB3 option that will suit an integration in Pro environments, not that it will evolve much towards the nMP form factor and quality level.
[doublepost=1491660473][/doublepost]Does anybody know the actual roadmap for Intel in 2018?
I got everything said so far about Skylake E5 or the W version. I am pretty sure Apple will stick with Xeons E5 class CPUs. However, if the Skylake version comes up within H2 or even Q3 2017, wouldn't it appear as an "old" CPU when Apple will present the mMP in H2 2018....hopefully, not in Q1 2018?
That's why I'm keen on the AMD option. Naples works easily in a 1S MB, it has more memory channels than Xeons, therefore allowing for a really powerful machine, the way we like it
I'm thinking:
AMD Naples 16/24/32 cores
(or... I guess, Skylake E5 6/8/12 cores?)
2 or even up to 4 NVMe SSDs in RAID 0, offering 5 to 10 GBps transfer rates (user replaceable) Each SSD comes in 256/512/1/2 TB. Minimum config: 2 x 256 GB RAID0. It may be even cheaper than a 1x 512... but faster, wouldn't it?
2 or 3 PCI-E 3.0 slots (let me hope for a future ready 1x PCI-E 4 slot!!): 1 with a GPU, the other available for additional GPU, Xeon Phi cards (btw... would they work with AMD CPUs?) or dedicated vertical expansion cards.
Ports:
6 TB3/USBC 3.1 Gen 2 (up to 4 4K/5K @60Hz/120Hz monitors or 2 8K @60Hz via Displayport 1.3?) (I may be quite wrong about frequencies and DP versions.. help me out here
)4 USBA 3.0
1 HDMI 2.0 (up to 8K?)
1 audio in (please bring back the optical!)
1 audio out (again... optical too!)
BT 5.0
802.11 ac/ad Wifi
1/2 10 GbE
Extended wireless/USB Keyboard with touchbar and touchID
A few more ..requests:
* frontal ports! (at least a few TB3/USBC)
* easy to open, easy to clean, easy to upgrade (SSDs, CPUs, GPUs/cards)
* this is quite a stretch, I know, but I would love to have a small LCD screen with CPU/GPU/Network stats on the body
GPUs... maybe we should discuss them in June when we will know what AMD is up to and where nVidia is going for 2018...
What did I forget? Ah... any suggestions for the TBD? Which ports do you think it will have?
Last edited:
I bet Intel doesn't know the actual roadmap for 2018. June 2016 article about Skylake-X coming in Q2 '17.
https://forums.anandtech.com/thread...-and-skylake-w-processors-in-q2-2017.2475298/
It isn't. Probably Q3 and perhaps Q4 for volume shipments.
The Cannonlake-X follow up is contingent upon a very high yield 10nm process. Intel will be using 10nm in 2018. It doubt even they know though when they are going to get to very high yield on that. Same kind of hiccup happens with 14nm, then could very easily see a slide.
The E5 2+ socket solution probably will go first because Intel is going to use smaller dies and "glue" them back into higher core counts via "mulitchip process" named Embedded Multi-Die Interconnect Bridge (EMIB) ( http://www.anandtech.com/show/11115...n-core-on-14nm-data-center-first-to-new-nodes ).
The Cannonlake-W should use the same socket so it really doesn't matter much. If they built a SL-W version and and CL-W is late then can ship. Waiting until '19 would probably be too painful. If CL-W manages to ship in Q4-18 then they could with some small firmware upgrades as the rest of the board is likely the same ( could just use a PCH chipset that spans the two. )
That's why I'm keen on the AMD option. Naples works easily in a 1S MB, it has more memory channels than Xeons, therefore allowing for a really powerful machine, the way we like it
At what price points? Has mature working firmware that works with Thunderbolt ? Yeah 2018 will be differnt but so will Intel. Naples is being compared to current E5 v4. The E5 v5 Purley platform has a significantly revamped memory subsystem. How much of that is walked back into the -W (basin falls) models is open. But the notion that AMD has a major lead now.... that remains to be seen by actual product ( on both sides ).
You guys are nuts. If Apple does not release a machine with a form factor that accommodates at least 2 full sized pci graphics cards, forget the pro market. Apple will never update the graphics cards at the pace of the graphics card market, nor will most pros be willing to drop 10k on a new machine every year or two just to upgrade some piece of apple **** graphics design.
To be blunt and clear, if their pro doesn't have a form factor accommodating at least 2 full pci graphics cards, say goodby to the vast majority of what is left of the pro market.
Stop the stolkhom sydrome, and demand what is really needed. This is your chance to have an influence with the muppets in charge of the Mac Pro. Demand better. Demand at least 2 full pci card slots in the new new Mac Pro.
To be blunt and clear, if their pro doesn't have a form factor accommodating at least 2 full pci graphics cards, say goodby to the vast majority of what is left of the pro market.
Stop the stolkhom sydrome, and demand what is really needed. This is your chance to have an influence with the muppets in charge of the Mac Pro. Demand better. Demand at least 2 full pci card slots in the new new Mac Pro.
Interesting opening paragraph:
http://appleinsider.com/articles/17...os-and-imacs-apple-building-its-own-gpus-more
Apple to be building it's own GPUs. Could that be for the coming Mac Pro?
http://appleinsider.com/articles/17...os-and-imacs-apple-building-its-own-gpus-more
Apple to be building it's own GPUs. Could that be for the coming Mac Pro?
Those GPUs are for iOS devices, not Macs.
http://appleinsider.com/articles/17...ologies-gpu-technology-moving-design-in-house
That's a very narrow definition of what pros need. There are other pros for whom four GPUs is the minimum:
https://developer.nvidia.com/devbox
Apple Made Me Buy a Windows HP Workstation
Apple May Just Commision (and I bet they will do) Reference Design GPUs with just the PCIe interface and DP output re-routed to the card end, not modding the most important part: GPU-Memory lines and Thermals, thus they will be able to deliver new GPU as soon new refrence design are released since re-route the PCIe and DP buses to a custom connector very close to its original locations its something so-trivial even can be automated at the Eagle Software (the CAD used to design Logic Boards), even I bet Apple Will do This, controlling what PCIe peripheral users can insert gives they more system reliability and less warranty issues, PCIe cards are prone to shortcuts due miss aligned conexion, Apple dont want that, further also they NEED to FEED THUNDERBOLT 3 controllers with DP output, and I doubt they will plug an DP cable from the GPU to the Motherboard, thats dirty.
Modular Mac Pro doesn't mean Commodity Mac Prom forger ISA PCIe GPUs return to the Mac.
AMD has shown to provide a very cheap alternative to i5s and i7s... therefore I expect Naples to be at leat 40% cheaper. Moreover, the comparison with Broadwell were showing 150 to 250% increase in performance... I wouldn't think it will not compete with Skylake Xeons..
Disagree. Just slap a regular card and use the video out from the card. Thunderbolt has failed for this purpose outside of laptops. Let it be the glorified FireWire failure that it has become. USB 3.1rev2 will likely supplant it even for interconnect, particularly since it now runs over the USB-c physical plug.
It complicates stuff on a pro model and removes flexibility for no real gain; another form over function dead end complication. Waste of time for video. Let the video card be video card imo.
Not to mention, I do not want apples abortion reconstruction of a reference card. I want to be able to get whatever card ****I**** want (like a titan) and slap it in without drama. Apples one size fits all is a failure for the pro market. We should be able to slap in any card we like, like in the 5,1 Mac Pro.
I think the trash can Mac has proved itself a failure, I see no reason to go back to any part of a design responsible for such a spectacular disaster in the pro market. As always, ymmv
Last edited:
Apple wont please everybody, I'll keep this post bookmarked.
I 'll comment here something curious, yesterday I received an Invite to participate in a closed market study, I got paid just to answer a questionnaire, and everything has to do with Pro R&D Activities HW and software needs, the questionnaire was long took me almost an hour, but IoT, GPGPU, Artificial Intelligence, VR/AR where the main focus, nothing on video/audio, the anonymous customer wanted to know What The Hell do i Need to Work the next 6-18 months?
Dual GPGPU - Some IoT - FPGA - Swift - Python - C - Objective C - Some MS Windows **** - Most Mac/*nix related Stuff.
Curious the customer to include multitude questions around Swift, Python, FPGA and AI, CUDA, OpenCL, but Metal neither Vulkan...
Maybe Apple was just Profiling their Top Tier Pro Users, Or HP was querying what to sell me to make the jump?