Perhaps they're embracing Trump's "chaos" model - no clear strategy, just do stuff.
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 don't know many people who bought iMac Pros, but the ones I know who did came from Mac Pros. I don't know of any that traded up from an iMac.
The price premium, and the components don't really position it for someone who just wants a more powerful iMac. If all you want is a high end GPU it's very much not cost effective. And the core count doesn't really make much sense unless you were in the workstation market anyway.
The people I know who are buying it are Mac Pro users who wanted the power of a Mac Pro, but only used Apple displays anyway and didn't care about upgrades. Once you exclude those people... Who's left for the next Mac Pro?
[doublepost=1522461477][/doublepost]
which one of these two were the plan and info because they are relatively inconsistent. If the MacPro 2013 volume was too small to be viable then there is little rational reason to believe that the iMac Pro volume is any more viable. It has basically the same market limitations.
a. Thermal corner. Same approximate thermal constraints. A bit better because only one GPU standard and the coolers separated, but the GPU is hobbled by a significant percentage and CPU base clock set a bit lower than normal (turbos up to almost range cap so not to big of a hit).
b. no GPU upgrades . Same.
c. singular SSD storage drive . Same
d RAM upgrade . Worse .
e. bundled monitor. Worse for those don't want it.
f. Angry Nvidia CUDA mob . Just as pissed off.
g. no std PCI-e slots. Same .
How is that going to possibly sell better than the MP 2013 did in to the same general workstation market? The folks who didn't like the MP 2013 aren't going to like this anymore. Might even loose some of the folks who actually liked the MP 2013. So where is the major uptick in sales going to come from?
It has newer stuff but they could have put newer stuff in a MP 2013 design. It does share some cost factors with iMac. ( screen , probably wifi implementation, basic case design slightly different holes in case ). But as far case custom internal logic board is has major changes from the iMac baseline.
I think there is a difference between priority and pemanent top end of Mac scale. I do think wanted to get a clean picture of just how much of the old Mac Pro market they could get with the iMac Pro before putting a Mac Pro on the top. If it was a high enough percentage then they'd could skip a Mac Pro. I don't think that it was a "go" if the percentage was going to be too low though ( if most top BTO iMac users would shun it along with the Mac Pro 2009-2013 crowd also shun).
I suspect the iMac Pro gets enough coverage to get out of the door. However, it isn't really high enough. They need something else to spread the R&D costs over and a revised Mac Pro will fit the bill. ( uncapped CPU and GPU of same baseline microarchectuure so can reuse R&D; firmware boot and drivers largely don't care what the clock speed is. )
In short, I think there was a plan to look before they leaped this time. The iMac Pro is basically part of the look and it is the fall back to the new top if can't work something out.
Last edited:
The last thing I want in a "Pro" Mac is an attached display. Hence, I'm not an iMac Pro customer.
Perhaps Apple should make two different boxes, one for a single socket and one for a dual socket system. Give both 4 PCI slots, user upgradable RAM and storage, and enough power to run 2 substantial GPUs.
And for ****s sake, just put it in a tower.
Perhaps Apple should make two different boxes, one for a single socket and one for a dual socket system. Give both 4 PCI slots, user upgradable RAM and storage, and enough power to run 2 substantial GPUs.
And for ****s sake, just put it in a tower.
The iMac Pro is pretty much the successor to the 2013 Mac Pro. But I think one reason Apple probably got negative feedback is because it fixed none of the issues from the 2013, and even doubled down on them. (There is no way at all the iMac Pro GPU could be upgraded, as opposed to the "in theory" of the 2013 Mac Pro.)
Ive is on a completely different page. In his mind, the 2013 Mac Pro is a failure because it wasn't elegant enough. The iMac Pro is his version of a dream workstation, and there will be some people who get on board with that. But the customers that never got on board with the 2013 Mac Pro probably let Apple know that the iMac Pro wasn't going to bring them back.
The iMac Pro is also probably easier to maintain. The internals are wildly different, but the case design is somewhat similar, maybe could share tooling, and the display panel is the same. It is an easier machine for them to maintain, but it runs into not being what customers want.
Apple really wants to keep their high end customers. I don't know why, considering how eagerly they've been dropping everything else. But they really want the high end creatives to stick around.
There is a pretty big leap of an inference. Every single GPU on that list has previously appeared soldered to the motherboard of some Mac in one configuration or another (some mobile stuff clocked a bit lower but same baseline GPU implementation; just a different functionality setting. )
There is little here in terms of "new card" coverage other than bigger thermal window. Apple made the same GPUs they already had drivers for work. There is still open question is whether they can get something they haven't already done supported.
"... Also note that macOS High Sierra 10.13.4 doesn't support eGPUs in Windows using Boot Camp or when your Mac is in macOS Recovery or installing system updates. ..."
Similarly that switching to Bootcamp Windows appears to still have problems. Apple may clean that up by 10.14 but is the coverage going to get wider?
The "stock" cards are those that Apple has done lots of testing with. Others may or may not work depending on tweaks. The specific vendors is more of a narrowing than stock.
...$100K for a restored toyota landcruiser from IKON is something normal non-reptilian humans do.
This would be the first time Apple is supporting GPUs that are not part of their "Made for Mac" program. Before the only GPUs that Apple supported had to be Apple certified. These GPUs they are recommending are not. All I'm saying is they're giving up control over GPU upgrades where they hadn't before. They're not making the hardware vendors go through them for their blessing any more.
The Boot Camp issues have to do with dual GPU Macs. eGPU under Windows wants access to the integrated Intel graphics, Apple's Boot Camp firmware loader disables integrated graphics.
I haven't heard what happens on machines with integrated graphics only, but because it's a problem with Apple's custom GPU mux'er, hard to apply that across Apple's entire line, or figure out if 10.14 would even be able to fix that.
You miss:
h. despite being 5yr Old, the Hated Ill tcMP still outperfoms it at twice fast at FP64 calculations (with D700's).
Its clear the iMP dint sell as good as the tcMP did it first quarter, anyone supporting the iMac Pro "success" its naive at least.
Why Apple didnt launch an mMP along the iMP? one (or both) of two reasons:
- They want to migrate as many PRO users as possible into a new Mac category the AIO PRO Workstation: iMac Pro...
- The mMP uses a meaningfully different hardware architecture not available the same time as the iMP was conceived (as a interim product IMHO) and the actual reasoning why there is no mMP: switching to AMD CPUs.
Right, so how could the iMac Pro outsell the MP 2013? Part of the decline from the 2009 (and before) era of classic MP sales was folks transitioning to other machines. If folks left the Mac Pro for a top of the line iMac then those same folks probably aren't going to buy a iMac Pro. Especially, after the iMac probably picks up 6 core capability this year with its update. (similar issue for MBP 15" if it picks up 6 cores also. )
I also forgot to mention that the base price of the iMac Pro is also $2,000 higher than the original base price of the Mac Pro 2013. There is no way it is going to outsell the MP 2013. It is priced not to interfere with that exodus to the iMac.
The goosed up price may help with the initial buyer bubble demand curve but the year forward propects are likely to decline faster than the MP 2013 did. (and the MP 2013 initial demand bubble was likely better. )
1. I don't Ive is involved much in post sales launch market research. I wouldn't be surprised if he had some influence in prioritizing the iMac Pro as the priority follow on to the MP 2013. One of the objectives here seems to be use being a literal desktop machine as a design constraint. It has to sit on desktop not consuming much noticeable space and be relatively quiet sitting relatively close to the user.
2. I would bet he moaned and groaned about the need to cut visible holes along the back bottom edge of the iMac case to increase air flow. The darker color makes the contrast along those holes and case a bit lower.
3. Not only the customers, but the any competent folks in marketing should have also known that the iMac Pro wasn't getting those folks back either.
several reasons why.
1. Open the Mac App store and look at the Top Charts. Go to the "top paid" section. See those two Apple programs in the top five. They are important to growing Apple's services revenue. Analyst are looking at that more than Mac Pro sales trends. The Mac Pro isn't completely necessary for Logic and Final Cut Pro but it is useful in keeping those in the top 10-20.
Some folks won't buy a Mac Pro but if think their Logic/FCP workload is going to grow to "Mac Pro big" in the future they'll buy into Logic/FCP at a lower level Mac. If you locked in but there is a broad spectrum of platforms/products then don't feel so locked in.
2. The high end customer are not as price sensitive. So they buy more RAM , SSD , and CPU with larger profits than the MBA customers do. PC vendors partially balance off their loss leaders "race to the bottom" systems with the "Pro" line systems with much fatter profit margins. Apple's margins are pretty high across the line up. I think the higher margins are going to be used to cover lower run rates. I suspect they are going to play a game of "chicken" with a "death spiral". That they won't raise it so high that the loose a critical mass.
[ I think they are wrong. Another substantive contributor to "Mac Pro" sales decline has been the increase in the entry base price. If they hadn't established a track record of doing whole lot of nothing for 3-4 years at a time that might work. Higher prices and likelihood going to disappear down the Rip van Winkle hole again probably won't work. ]
3. It isn't just the "high end customers". it is the high end customers who are left. Two 3-4 years cycles of Mac Pro updates and there are a bunch of folks who have just left for Windows ( Linux or maybe Hackintosh). The only upside is that the ones that are left have a high "update pain" tolerance. If Apple goes down the Rip van Winkle hole for another 3 years after the next Mac Pro updates they will probably wait it out.
If Apple did substantive iMac Pro updates in odd years and Mac Pro in even years that would make that crowd happy.
2007 iMac Pro 2018 MP 2019 iMac Pro 2020 MP .....
Even if Apple did something like
2018 MP ( and iMac Pro to synch up). 2020 MP / iMac Pro 2022 MP/ iMac Pro
that high pain tolerance crowd would probably be happy.
I think one thing that put the Mac Pro back into the development resource allocation is that Apple got surprised how many folks stuck around during their long naps and are willing to give them another chance.
4. It isn't the somewhat narcissistic small scale threat ( if I don't get a new Mac Pro I'm not buying a iPhone (or MBP) anymore). That the Guy Kawasaki era "evangelists" holding their family members hostage ( get me a Mac Pro or I'll make my family drink the Jim Jones kool-aid). High end in terms of yearly spend. If your $XX Million dollar a year customer says that would like a Mac Pro in the mix and there is a number of sales folks whose commission/jobs depends on keeping that customer then it bubbles up to the top.
[doublepost=1522477479][/doublepost]
That is a corner case where can push load onto both GPUs. The MP 2013 CPUs is substantially outclassed by the iMac Pro. Single GPU it is a clear win even with the Vega 54. That is a mixed bag.
stop with "Apple watch is a failure" hooey. The iMac Pro did sell. there is little to indicate Apple isn't heading toward breakeven. The reviews from people who actually have used/own them is generally good/positive. Even if Apple wanted to they probably couldn't sell more than they had contracted for with AMD in advance. There are few "spare" GPUs floating around.
1. They already were migrating. This was just matching a trend that customers were already doing.
2. That is alot of hand waving. There are lots of indications that Apple was not working on the mMP at all prior to mid-2016. It is far more likely it is lack of resource allocation inside of Apple that is the hold up than technology outside of Apple being unavailable. The iMac Pro had to finish to release state before next Mac Pro got fully resources to proceed.
What Apple needs is something that has a very low risk of being screw ( again). That means not taking a huge left turn from the rest of the Mac line up. Intel W CPUs would work just fine and in fact would probably improve the margins on both the iMac Pro and the Mac Pro. systems ( which are probably under duress due to lower run rates .) AMD doesn't magically save those systems ( and especially not the Mac Pro individually ) from that run rate problem that is a critical issue.
[doublepost=1522479646][/doublepost]
Where "us" ( in the let's ) means Apple impose a segmentation onto the market to pro "desktops". Yeah that isn't really true. However, it really isn't Apple driving the segmentation. It is more so the customers. So having two "desktop" to address self segregated customer groups isn't necessarily bad.
The Mac Pro of 2006 era encapsulated a different set of users than the Mac Pro of 2013. Those two groupings contain different sub-segments. Those two are also different from the current Mac Pro active user base ( which consists primarily of a blend of 2008-2012 Mac Pro users and Mac Pro 2013 ones ).
Errr no. The Mac Pro 2013 is waaaaaaay overdue for replacement. The iMac is squarely aimed at a huge chunk of them. Getting those folks moved off the Mac Pro 2013 and onto the current and incremental iMac Pro's over the next 3 years so Apple can sail the MP 2013 into retirement in 5-6 years is very much the objective. I think there is gross bias among large segment of the pre MP 2013 market that "nobody" bought those systems and it is super minuscule set of users. It isn't that small and technically they are part of the Mac Pro userbase.
iMac Pro is also aimed at pruning off some of the 2010-2012 hold outs whose workload levels have plateaued. The iMac Pro again is an upgrade target for them. In 2016-2017 there are some folks who would have moved (on 5-6 year cycles) to the MP 2013 but it appeared dead.... so they waited. (the plateaued workload makes waiting quite tolerable). A substantive fraction of those are going iMac Pro also. It is a system they can sit on for the next 6 years and not exit its performance envelope.
They announce another Mac Pro because there are multiple segments of the old market. Not just one uniform one.
The bet Apple keeps making is that people want less complicated pro hardware, with less options (see also: The MacBook Pro.) It's the group inside Apple that thinks these should be purity of design boxes that are never opened and never touched. And tied to that is that businesses typically lease hardware, so they're on a regular change out cycle anyway.
iMac Pro? Great machine for folks who love that. But that's a subset of the market the Mac Pro was in. But the iMac Pro is Apple thinking we're actually the idiots and if only we'd lease our hardware and appreciate better design we'd all buy in. And I do know people who buy into that philosophy and want a clean desk and a workstation. But again, I think Apple was surprised that not everyone bought into that.
I'm not sure at the executive level they even started doing market research until they realized the iMac Pro wasn't going to cut it.
I'm not sure the people in charge knew why they had those folks to begin with.
Nearly every executive who understood the pro market is gone, including Steve himself. I mean, Steve had a iffy relationship with pros, but still better than Cook.
Apple's had no problem cutting categories before, and I'd bet FCPX sales are a shadow of FCPS sales. But whatever the reason, I'm happy they're sticking around.
Not sure what they'll do this round.. The problem is the Mac isn't as sticky as it used to be. If you were a FCPS shop, you weren't going anywhere. But these days people are more willing to jump ship to Creative Cloud and cheaper and more powerful Windows PCs. I think they realize that.
When the Mac Pro first came out, it was an easy sell. A price competitive, easily upgradable, one-of-the-fastest-workstations-out-there box. I knew people who bought them and just ran Windows on them full time.
Looking at the language Apple has been using at roundtables, I think they understand moving away from at least a few of those things was a mistake.
That's my worst nightmare. Apple comes out with a Mac Pro, promises there will be updates, I buy one and... nothing.
Again, if they allow for standard PC upgrades that pain is substantially less. But otherwise why not buy an iMac Pro.
Anyway, I hope Apple wants to bring back the people who left. With Windows being put on the back burner by Microsoft it really is wide open for Apple. They're just letting Microsoft run the field by not even showing up.
I haven't heard any whispers like that. They really thought after they failed to update the Mac Pro that they could ship out an iMac Pro and everyone would be just as happy.
I don't know the specific reason why they decided to put it back into development, besides it was so sudden that people didn't even know the announcement was coming. It just seemed like a big surprise. All I can figure is that something happened that surprised execs, and the best guess I can make is they showed it to big customers who said if this is the replacement we're all moving to Windows.
What I had heard long ago is that Apple didn't a huge number of cMP's, but they were so cheap to upgrade and make that it was hard not to make money on them, and the high end customers stayed happy. Apple basically was able to upgrade the cMP in their sleep, so even if they didn't want to put very much time into it, it was no problem.
The tcMP was NOT that at all. Upgrading it was expensive and time consuming, and they couldn't just buy some off the shelf parts and have an intern throw new configs together. And in the end they never were able to upgrade it.
That's the sort of thing I think about when I read what Apple's VPs say about the next Mac Pro. They're looking at the cMP, understanding that standard components meant more upgrades with less resources. I don't think they want to increase the amount of effort they put into the Mac Pro. They want everything to be less custom so they can ship more Mac Pros with the same or less effort.
This whole thing about custom form factors and having service at the Apple Store etc etc... I mean, it could happen. But I think it goes against Apple not wanting to have to babysit the Mac Pro. Really what they want is a machine where an upgrade means they throw on a new CPU from Intel, get AMD to rustle up some new reference GPUs from the warehouse, maybe do a firmware upgrade, and then ship. Only doing a new board design every 3-4 years. Literally so easy they could outsource most of it and have just a few people spend a few months every year doing the actual engineering.
And if users can do their own upgrades and third parties cover many of the gaps? Even less work and time and money they have to spend on it, and their users are just as happy.
Apple couldn't even bring themselves to sell new SSDs for the 2013 even though it was a user serviceable part. I think people really overestimate how much Apple wants to be involved in Mac Pro after purchase upgrades.
Do you even read ? Nowhere was there any mention of Crossfire/SLI.
I mentioned Nvlink because that way the vram of two cards could be pooled together ( instead of duplicated on Vram across two or more cards vis SLI/Crossfire. A pointless exercise ). Two cards with an nvlink type of bridge would instantly double the vram available for rendering. If the nMP could use two internal vega64 with 16GB each ( and the navi for the next iteration of the nMP..if it come that is ) connected by some type of nvlink bridge, it will double the capacity of the cards to fit data onboard ( instead of having to stream them off the RAM/SSD if it cannot fit them on the vram )
Reg Multi eGPUs, since Apple is supporting them ( and expects the user to plug each box directly to the computer instead of daisy chaining them - a much touted feature of TBS - which suggests one more inelegant solution to workstation class systems.), how would the nMP differentiate itself from the imac pro ( never mind the AIO display issue ) if both of them support multi eGPUs ? More RAM ? SSD ? Slightly faster CPUs ?
You need 4 eGPUs ? Plug four boxes into your nMPs/iMac Pro. you now have a total 8 cables to deal with. In the meanwhile we will continue touting the nMPs shiny design features.
[doublepost=1522484446][/doublepost]
[doublepost=1522485044][/doublepost]
This. By the time the mea Culpa happened, the iMac pro was possibly already in the pipeline and locked down in terms of its main feature sets ( Time cook did mention earlier in Jan of that year that they have great plans for the iMac - which to me suggested we will be seeing a pro iMac ). And it appears some customers gave it a thumbs down when they saw the prototypes. Hence the meeting. Also if Apple was just concerned about it's corporate customers, it could just have had an internal memo shared with these customers in private without publicly announcing they were gong to make a nMP
Last edited:
Doesn't sound that strange from the richest corporation in the world ?
the tcMP is restricted to PCIe2 x4 SSD, notwithstanding there are 3rd prty solutions with NVMe oe even m.2 adapters, I think Apple is aware this.
SLI/Crossfire are one thing, and NVLink/Infinity Fabric(amd's eqv) is another totally different thing, do not mix apples and pears.
The tcMP has Crossfire but only works in bootcamp/windows, macOS never supported it.
SLI/Crossfire in general has as purpose to link the GPU's framebuffer output, and its bandwidth constrained at about 900MBps its equivalent to PCIe3 1x.
Crossfire is being replaced by Xconnect/XDMA an all PCIe solution, AMD supports Xconnect on macOS from current release (key to enable eGPU), instead link framebuffers it allows one GPU to render into anotherś framebuffer, so what you dream is already enabled for rx5xx/VegaXX gpus now onwards .
While nVlink/Infinity Fabric are meant as data carpool to avoid PCIe bottle necks, with nVidia CUDA you actually dont need nVlink for your apps to handle multiple GPUs as a single logical unit, I'm not deep into AMD development but the few I've read they atleast pursue a similar approach.
Last edited:
[doublepost=1522506294][/doublepost]
Not the vram of those multiple GPUs.
Then why not use M.2?
Apple used a custom keyed version of M.2, but they intended for the Mac Pro to be supported by third party upgrades? That makes no sense.
They don't even have any certified third party drives. And no one even offered upgrades at launch.
The whole 2013 MP smells of Apple having a vision of them offering upgrades, and then not caring enough to follow through.
[doublepost=1522524822][/doublepost]
And again, Apple doesn't make enough money off of the Mac Pro to justify spending large amounts of money on it.
--------
One thing I was thinking about is the idea that Apple can't control the Thunderbolt controller, and they can't control the design of Vega. In fact, they deeply have their hands in Intel's Thunderbolt controller development, and they have strong influence over the Radeon Technology Group.
It would be entirely possible Apple would sit down both Intel and AMD and say "We want the next generation Thunderbolt controllers to accept DisplayPort over the PCI bus, and we want the next generation Radeon GPUs to support directly addressing a Thunderbolt controller over the PCI bus. Oh, and in since this is an industry wide problem we're not going to pay for it, but you'll make all your customers happy."
Intel's next gen Thunderbolt controllers would work with DisplayPort signals routed over the PCI bus (or some other similar thing), all the new Radeon GPUs available to everyone would support it. Mac Pro customers could buy retail AMD GPUs off the shelf as long as they were of the latest generation or newer. Nvidia could get in on the party if they want to. And it would be an industry standard because, just like Thunderbolt, Intel will push it as a standard industry wide.
Mago's prediction, and his idea of the technical constraints, is limited because he's trying to make everything work with off the shelf Thunderbolt controllers and off the shelf GPU components. What is being missed is that Apple has a lot of influence over the next generation designs of both those things.
That could be why we're sitting here waiting. Apple has sent AMD and Intel off to go figure out Thunderbolt and discrete graphics on workstations, and the Mac Pro won't ship until "fixed" next generation Thunderbolt controllers and GPUs are available.
Last edited:
Maybe it didn't sell enough to warrant caring. Apple clearly hoped it would sell better than the 2012 model and maybe it did for a time, but if they were expecting a ten-fold increase and it was only a two or three-fold (in it's best quarters)...
And now, years on without an update, it sells under 30,000 units a quarter per the April 2017 Summit compared to around a million units a quarter for the iMac line. No real surprise they made an iMac Pro before a new Mac Pro.
Wrong, requires re-definition of the PCIe std (a 16x channel serial bus), if you'll need to take 2 of them for DP video you are bootlenecking the system, not to say the complexity to adjust bus timings etc, better solution should be to open the tcMP GPU conector (a custom PCIe x16 interface + 6x DP Interfaces + GPU intercnnect interface -CrossFire=>NVlink/Infinity-)
That means to mess the STD PCIe bus just to mockup a custom GPU as a STD GPU and seel it at Amazon?, better take the MXM std or the tcMP GPU connector and open it.
FYI information Thunderbolt didnt come from Intel or Apple Magicians, what they actually did was to find an application to the latest MAXIM RS485 Transceivers (then capable of 10MBps, current on 40MBps, next gen RS485 said to be capable of 100MBps), its not a trivial development, and actually all the Data I/O industry relies on modified MAX RS485 transceivers, from Intel TB, to Mellanox' Infinyband.
I make my predictions based on what is available to Apple in practice, while they may purchase Maxim and develop its own PCIe Thunderbolt, they wont doit to grace your likeness of a Generic PCIe GPU, this is ridiculous, its like Lamborgini purchasing Ram to put a Pickup Bed on the next Aventador.
[doublepost=1522532990][/doublepost]
Actually I see two possible reasons: after Jobs death, Oversight becomes very relaxed at Apple with cook being more into Liberal Agenda than Management, then Apple's R&D just cared on whats priority: Laptops, iPhones, leaving behind all the rest (even macOS), then possible plans to ditch Intel for AMD CPUs added more drag to the Mac Pro development, then (hopefully) come VR/ML trends to explode and surprise then having nothing in the ecosystem for macOS/iOS developers to build products for Apple and actually having a PROs stampeede, they realized they need an emeregency solution: iMac Pro, to add later an actual ML/VR workstation in leu of the tcMP never airmed at this kind of workloads.
[doublepost=1522533792][/doublepost]
vram FYI is an obsolete concept, modern GPUs use Framebuffers to render display output, vram (actually multi-port ram) is not used anymore, framebuffers are read asyncrous by DP serializer (its like another GPU core but one that at one end has access to the GPU ram, and at the other end has a serial signal output)
[doublepost=1522506294][/doublepost]
Not the vram of those multiple GPUs.
Last edited:
Dubious.
x4 PCI-e v3 --> 32Gb/s https://en.wikipedia.org/wiki/List_of_interface_bit_rates#Main_buses
DisplayPort 8K 60Hz --> 49Gb/s
DisplayPort 5K 120Hz ---> ~45Gb/s https://en.wikipedia.org/wiki/DisplayPort#Resolution_and_refresh_frequency_limits_for_DisplayPort
That's 1.4. The next iteration 1.5 or 2.0 should be higher. I think those for are normal, non-HDR, color also. Add HDR and the overhead also goes up.
One problem with that is that not only sucking up PCI-e bandwidth through the switch in the TB controller you also sucking up very substantive bandwidth in the PCI-e switches in the host system too.
The more significant problem though is that Thunderbolt pragmatically have DisplayPort pass through. If just need high end HDR video it can just be passed through.
" So as these types of monitors become more mainstream and pure DisplayPort monitors shift over to DP 1.4, Thunderbolt 3 has needed to catch up. ...
.. Intel’s Titan Ridge controllers will continue to require an external USB Type-C multiplexer and a PD 3.0 controller. "
https://www.anandtech.com/show/12228/intel-titan-ridge-thunderbolt-3
Your problem is that with Type C alt-mode DisplayPort mode the TB controller would have to tap dance that PCi-e encoded DisplayPort signal back into DisplayPort. Your notion is that some hocus pocus game could be play to mutate the PCI-e data as it is encoded into TBv3 protocol. That doesn't solve all of the requirement. alt-mode DisplayPort is also required.
If screen resolutions and gamuts were frozen at 2012 technology levels it might work. It would work if only trying to display to a laptop screen of 15" (or less ) diagonal and current mainstream dpi and gamut. The major issue those is the Mac Pro pretty much does not focus on driving laptop screens.
Other than lossless or visually lossless encoding, the DisplayPort port signal should be raw until just before it hits the TB transport. No reason to encode and decode on the same host system. It will suck up more power relatively corner case benefits.
FYI you are picking some odd issue never implied in the posts. FYI this is what I meant by NVLINK style dual GPU system as a way to differentiate between single card systems like the iMac pro, multi eGPUs vs a high ‘bandwidth, high throughput’ like the nMP : https://www.chaosgroup.com/blog/v-ray-gpu-benchmarks-on-top-of-the-line-nvidia-gpus
That’s it !
I know it's de rigueur to blame anything people do not like about Apple on Tim Cook, but folks really need to find a new hobby. He worked at Steve's side from almost the day he came back to run Apple to the day he passed so if anyone knew how Steve would have run Apple after 2011, it's him. Hell, we probably have Tim to thank for still having a Mac at all since Steve seemed pretty convinced the iPad was where computing would go and the "truck computer" was an anachronism.
While the solution doesn't have to be over PCI-E, PCI-E is bidirectional, which means that a x8 card has 64Gb/s of upstream bandwidth basically unused.
Thunderbolt 3 also can't do 5k 120 hz or 8k 60 hz so the scenario is irrelevant. You're talking a dual Thunderbolt cable solution which is basically the maximum a Thunderbolt Mac can do right now anyway. So it all still fits, and there would be more than enough bandwidth off a 8x or 16x card.
But again, this is a straw man I'm simply supposing to say that it's a solvable problem if you control all the pieces, which Apple, Intel and AMD do.
We're also on the verge of PCIe 4.
Sure, but it's nothing the switch couldn't handle. I mean, you act like it's a lot of bandwidth. But if I'm doing full screen updates on the display every frame, it's basically the same thing, and if your PCIe bus can't handle that, you have no business driving that display anyway.
PCI-E encoded? I'm not sure what you mean. The data would be packeted, but there wouldn't be any encoding or decoding going on. And it's not much of a problem because DisplayPort is packeted anyway.
I think you're missing something here. I mean, this is all just a straw man. But the DisplayPort data would never be encoded or decoded. It's just wrapped in PCIe packets, just like it would normally be wrapped in DisplayPort packets.
Basically what I'm saying is that AMD could modify their GPUs to support sending a raw DisplayPort signal to a peer PCI-E device. In this case, the Thunderbolt controller. The Thunderbolt controller would have to be changed to expect a raw DisplayPort signal over the PCIe bus from a peer device. But again, Apple has their hands in both those suppliers.
I also think we might see Apple, while supporting Thunderbolt displays, move back to pushing direct DispayPort output for displays as well. It's not like there are any existing 8k Thunderbolt or USB-C displays out there they need to support anyway. And it gets them away from Thunderbolt and USB-C bandwidth being the gating factor for professional displays.
[doublepost=1522543864][/doublepost]
Needs citation. Devices that share a PCIe bus can send any info they want to each other. DisplayPort signals are 1s and 0s just like any other data. So send the DisplayPort signal over PCIe transport just like any other data sent between two PCIe devices.
Bottlenecking what? GPUs don't usually send very much data upstream (or downstream, but that's a different conversation.)
So you're saying Intel has no idea how to alter Thunderbolt controllers? Mmmmmkay.
Last edited:
NO
[doublepost=1522545566][/doublepost]
1st, You dont know a thing on how PCIe works, havin unused Upstream bandwidth doesnt means you have an separate open channel available for downstream, as the same copper line is used concurrently to upstream and downstream using different signaling schemes.
All the stuff you say, still no practical sense just to extend the PCIe functionality to justify PCIEe slots in the mac pro, becasue the TB3 imperatively needs DP signals and you know the mMP wont happen w/o TB3, when the solution comes from the hated tcMP: a custom GPU interface driving DP,PCIe and maybe GPU Fabric interconnect, you dont like the solution coz means not std PCIe slots, but Apple could make this solution Public Domain and thus other hardware mfr deploys its own Workstations or Gaming Rigs whit these TB enabled GPUs, w/o messing the PCIe interface.
You need to go outside Gamers forums, and try to read about Digital Compute Science, lots of wrongs and self assumptions from an uneducated biased approach.
Embrace it, you'll never get a Custom GPU solution w/o a custom PCIe solution, even if it uses a bare PCIe slot what you are defining still a cusmto PCIe gpu more expensive and likely Apple Only, no matter if it uses Bare PCIe connectors this mac wont run on STD gpus and thise mMP-GPUs will never run optimally on a Std PC.
Hopefully having a custom GPU - Mac Only means Cryptocurrency Miners will stay away our GPU supplies.
Last edited:
That's exactly how it works. It's even more than that. The channels are reserved. An upstream channel CANNOT be used for downstream data. In fact, even if you saturate a PCIe card with downstream data, all your upstream channels will sit unused.
That's exactly wrong.
PCIe 1.0 and higher is full duplex. That actual speed of PCIe per lane is DOUBLE the rated speed. Except half is reserved for down and half is reserved for up.
To pull from Deconstruct's example, x4 PCI-e v3 is rated at 32 gb/s, but the actual speed is 64 gb/s total. But in since half is reserved for each direction, the top speed is 32 gb/s in a single direction.
PCie reserves lines for up and down and halves them evenly.
Just pop "PCIe bidirectional" into Google and do some reading.
(You should also realize this is the way PCIe works without doing the reading because this is the way Thunderbolt works, which is basically PCIe over a cable. Thuderbolt 3's top speed in both directions combined is 80 gb/s, but the max speed in a single direction is 40 gb/s.)
Wikipedia also spells this out pretty darn clearly:
Translation: I made a prediction and I'm upset someone dare question it.
I just set up a theoretical for how raw DisplayPort data from a modified GPU could be sent across a PCIe bus to a modified Thunderbolt controller. I think I've explained it pretty well.
I'm not saying Apple will do it. But I'm saying because Apple has influence over Thunderbolt and AMD, they have options that aren't on the market right now, and it would be a mistake to assume their options are limited by what's on the shelf at Frys.
...and all of this, for a goal which at the end of the day, amounts to being able to use one single cable for display and peripheral bus, rather than one cable for each.
*shakes head*
meanwhile, i just spent the day doing concept design in VR for a 20' tall sculpture that's going on stage for a play (Rent), using Tilt Brush (not one of the two utility apps for VR on macOS), using a machine that was significantly faster for 3D than anything Apple makes, cost half as much as the 30% slower Mac-based solution (iMac Pro + Vive), was quieter than the ambient noise of a closed, empty (apart from me), aircon-switched-off library, and the only interaction I had with the operating system ("eww, Windows"), was to copy my file to a different directory once I'd finished for the day.
I can't even...
*shakes head*
meanwhile, i just spent the day doing concept design in VR for a 20' tall sculpture that's going on stage for a play (Rent), using Tilt Brush (not one of the two utility apps for VR on macOS), using a machine that was significantly faster for 3D than anything Apple makes, cost half as much as the 30% slower Mac-based solution (iMac Pro + Vive), was quieter than the ambient noise of a closed, empty (apart from me), aircon-switched-off library, and the only interaction I had with the operating system ("eww, Windows"), was to copy my file to a different directory once I'd finished for the day.
I can't even...