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
Muxers
Ok, not really Xeon W. provides 48 PCIe3 CPU lines plus 24 from PCh
PCh (C422) is good for LAN/WAN => 4 for 2 NbaseT, 12 for 3 TB3 headers full speed, 1 for Wlan/Bt , 7 PCIe free.
CPU PCIe i 32 fo 2x GPU, 8 for 2 NVMe, still 8 lines available either for 2 full sped TB3 or more storage (in a normal WS will be used for 2x PCIe x8).
Last edited:
With the current Xeon W there is not "lane problem". There are 48 lanes just on the CPU. The PCH has more than enough if need to float some low usage "extras" for modest bandwidth I/O devices.
From example:
x16 -- Primary Boot/Display GPU
x16 -- open standard d PCI-e slot
x4 -- 1 or 2 10GbE Adapters ( two on a PCI-e switch from x4 v3 ---> two x4 PCI-e v2 )
[ Or leave the second embedded 10GbE adapter off and have an empty x4 slot ]
x4 -- M.2 slot [ Or again a switch and a M.2 and other M2 slots or empty single x4 v3 slot ]
x4 -- Thunderbolt bus 1
x 4 -- Thunderbolt bus 2
Three Thunderbolt buses are in no way shape or form necessary at all. If Apple wanted to shot themselves in the head they could trade the M.2 slot for a third, but that would be extremely dubious. The three on the Mac Pro 2013 were as much driven by wanting to have 6 mini-Display port (i.e., drive 6 monitors with Apple's then standard video connector) than anything specific about Thunderbolt. ( it was the backward compatible, fallback mode that was of higher utility. ). Thunderbolt 3 shifted away from mini-DisplayPort and DisplayPort version > 2 means can run multiple displays on a single wire. The constraints of 2013 aren't the constraints of 2019. Two Thunderbolt pairs is enough. To get 6 displays two additional ports ( 1 mini DP + 1 HDMI or 2 HDMI ) is sufficient to get at least 6 panels being driven with better backward compatible coverage for older monitors.
x4 PCI-e v3 is 32Gb/s . Two 10GbE links fit on that just fine.
The PCH can host T2 ( and x4 SSD ) , bluetooth wifi , USB Type A sockets (and perhaps a USB only Type-C for the front panel). Also even another 1GbE port if Apple wanted to put on three ( or two ; one 10GbE and one 1GbE. One for storage subnet. One for generic Internet ). Or two 1GbE ( if think there are still folks who want to use the Mac Pro as a super expensive bridge router. That too is clinging to a now distant past. )
If willing to thin out the PCH bandwidth there is room on top of all of the above to even put another M.2 slot hanging off the PCH.
[doublepost=1541445643][/doublepost]
Don't need a Xeon SP package for that at all.
pci-e x4 for storage is will suck up the full x4 for pch (now in some server broads you can add more pci-e lanes to the pch)
The iMac Pro doesn't have Xeon SP ( Platinum) processors. Intel W is a different package (with a similar die with some features flipped off. ).
It is highly doubtful it will be an updated version of same design constraints. The single central core doesn't line up with the "two Vegas' approach. Those won't be even TDPs all around and Apple has openly said that core and uneven TDPs isn't optimal. So doubtful they are going that route.
Similar in gross 'spirit' a slim maybe. If they are hooked on it being a literal desktop machine and roughly similar overall volume then they'll probably walk themselves back into a slightly different corner. But they already have a desktop "top end" Pro system; iMac Pro. Why they would need another that was constrained to the literal desktop would be a mystery. The Mac Mini being shifted up in average selling price even more so.
The other design OCD problem is that the GPUs have to perfectly symmetrical twins. Same type, build shape. That is just ludicrously dumb also. Some configs can be twins, but it shouldn't have to be all configurations have to be twins. That is just completely unnecessary. The second 'compute' engine may not even be a GPU. Or even a 'compute' engine.
6 is a fall out from Xeon SP ... which they are probably NOT using. Apple should enable 8 ( 4 with two DIMMs per channel that the Xeon W fully supports ). That is far more cost effective than slapping overly expensive Xeon SP processors in the system. Max would 512GB; which is higher anyway.
6 channels as a nominal default would just drive Mac Pro prices higher with Apple's "fill all primary DIMMs" standard approach to configuration. An entry Mac Pro with 4 8GB DIMMs is going to be more affordable than 6 8GB DIMMs.
Copper won't do anything productive with the root cause issues of high mismatched TDPs. And 600W is probably too low for a Mac Pro budget. Something closer to 800-900 would need to be minimum if targeting two top end Vegas and any serious way. 600W is about what the combined GPU subsystem budget should be all buy itself ( including some slack for some power spikes. ).
Apple doesn't need 6x faster ... then just need anything that is new and up to date. The new Mac Minis are passing up the 6,1 on some benchmarks. The primary root cause problem is 'old age', not some narrow Intel tech porn at the most expensive end of the pricing alternatives.[/quote]
[doublepost=1541447832][/doublepost]
First, that is why the 'secondary" M.2 slot isn't on the PCH ( it is on the CPU ). [ Same if Apple puts a "drive bays" worth of M.2 slots off the main logic board. 2-4 slots hanging off of a switch. ]. the empty x16 slot could be used for full sizsd, max capacity SSD if need bulk and speed out of a SSD more than two GPUs.
Two, the T2 handles more than just the SSD. The camera and mic are also off of it also. T2 also handles some PMIC duties. The latter are all normally hooked to the PCH so that bundle places it there. Yes, you'll get some contention out of the USB gen 1 being used on same DMI link as the system SSD , but probably not that much. But vast majority of USB stuff is more than mundane I/O. ( yeah some are SSD hooked to USB but most of the stuff that users tend to hook up grossly is not same class of bandwidth. ).
Three, the PCH is oversubscribed. SATA , USB (possibly USB gen 2 ) , and a SSD basically can fill up the whole system. But it is a 'cheap' switch because have to buy it anyway. It is already there and you have to pay for it ( system won't work without it.). So if need to oversubscribed something .... might as well oversubscribe what you already paid for. That is WAY more cost effective than drifting into Xeon SP zone, if you want to have any decent single user oriented base clock speeds and efficient use of internal volume/space.
[doublepost=1541448016][/doublepost]
Neither likely does Apple. This is all extremely likely 'stir the pot' , "Dark net guy" hand waving as opposed to credible rumors.
Last edited:
You can add lanes to the PCH
look at the manual block map for this board.
https://www.supermicro.com/products/motherboard/Xeon/C620/X11SPi-TF.cfm
can even link cpu-2 dmi out to an pci-e slot
https://www.supermicro.com/products/motherboard/Xeon/C620/X11DPX-T.cfm
look at the manual block map for this board.
https://www.supermicro.com/products/motherboard/Xeon/C620/X11SPi-TF.cfm
can even link cpu-2 dmi out to an pci-e slot
https://www.supermicro.com/products/motherboard/Xeon/C620/X11DPX-T.cfm
From the article: "'With regards to the Mac Pro, we are in the process of what we call "completely rethinking the Mac Pro'".
"Completely rethinking" the Mac Pro is what got them into trouble the last time.
If they made a Mac Pro akin to the tower design, they wouldn't have enough lanes off the CPU for all internal expansion and Thunderbolt 3 at full speed, but that doesn't seem like a major stumbling block—how many people would have all slots and all expansion busses running simultaneously?
Either way I highly doubt we're going to get another cylinder, insofar as Apple seems to have made it clear that the thermal core, though impressive engineering, ultimately was too limited for its uses. I know people here (myself included) wondered if it was possible to scale up the design, and it might be, but Apple specifically talking about the thermal corner of asymmetrical loads suggest that it's fundamentally not what they're looking at for the successor.
In the US here it was $599. The only $499 models in the US were the original 10,1 and 10,2 and then the 7,1 in 2014.
Either way I highly doubt we're going to get another cylinder, insofar as Apple seems to have made it clear that the thermal core, though impressive engineering, ultimately was too limited for its uses. I know people here (myself included) wondered if it was possible to scale up the design, and it might be, but Apple specifically talking about the thermal corner of asymmetrical loads suggest that it's fundamentally not what they're looking at for the successor.
In the US here it was $599. The only $499 models in the US were the original 10,1 and 10,2 and then the 7,1 in 2014.
Last edited:
Please Craig Federighi read this, please in the name of God...
Not a problem as Apple has no shame to underclock GPU/GPUs to fit its beautiful exotic whims.
It's just following the Mac mini 1x per year speed bump formulae (I was kidding)...
[doublepost=1541517660][/doublepost]
Solution, the all-copper (or all heathpipes) Thermal Core II with double TDP capability (900W) (given how poor was the original thermalcore, is not rocket science to increase it TDP capability 2x or even 3x by just implementing heathpipes or pure copper fins.
Given the pictures of the stack of Minis, and that Thunderbolt has excellent networking capability (I have a hybrid RAID/NAS that hooks up via Thunderbolt (or 10GBe - it has both), but shows up as a network drive even if attached over Thunderbolt), could Apple be working on clustering?
What if the Mac Pro is a taller Mini with a Xeon (something with a bunch of cores >10 and options running well above that), a high-end Vega (non upgradeable or proprietary upgrades at best), plenty of RAM capacity (>512 GB) and big SSD options? It also has 8 TB3 ports on 4 buses and multiple 10 GBe ports... To add power, you plug more Minis or Pros into it, according to your needs. If you have the need and the budget, you can build a supercomputer without much trouble... It wasn't all that long ago that some university cracked the TOP500 supercomputer list with a roomful of Power Mac G5s (it actually impacted early availability of the G5 to anyone else).
Of course, this raises software issues - many rendering jobs run on clusters, but not all do. Final Cut and Compressor aren't there yet, but they have hooks to allow it, and it wouldn't be hard to add. You don't need to run Word on a cluster, anyway - apps like that would simply ignore the extra machines. Overall, I don't think it's that much harder (I'm not a computer scientist - anyone who is, please correct me) to get a big job running on a cluster than it is to get the same job to use a many-core processor (or multiple processors) adequately.
The days of powerful processors without adding cores are over... Intel's been struggling with per-core performance for years, and other manufacturers are hitting the wall as well. The only way left to add speed is to add cores, whether it is on big chips, multiple processors or multiple boxes. Could Apple be doing multiple boxes?
What if the Mac Pro is a taller Mini with a Xeon (something with a bunch of cores >10 and options running well above that), a high-end Vega (non upgradeable or proprietary upgrades at best), plenty of RAM capacity (>512 GB) and big SSD options? It also has 8 TB3 ports on 4 buses and multiple 10 GBe ports... To add power, you plug more Minis or Pros into it, according to your needs. If you have the need and the budget, you can build a supercomputer without much trouble... It wasn't all that long ago that some university cracked the TOP500 supercomputer list with a roomful of Power Mac G5s (it actually impacted early availability of the G5 to anyone else).
Of course, this raises software issues - many rendering jobs run on clusters, but not all do. Final Cut and Compressor aren't there yet, but they have hooks to allow it, and it wouldn't be hard to add. You don't need to run Word on a cluster, anyway - apps like that would simply ignore the extra machines. Overall, I don't think it's that much harder (I'm not a computer scientist - anyone who is, please correct me) to get a big job running on a cluster than it is to get the same job to use a many-core processor (or multiple processors) adequately.
The days of powerful processors without adding cores are over... Intel's been struggling with per-core performance for years, and other manufacturers are hitting the wall as well. The only way left to add speed is to add cores, whether it is on big chips, multiple processors or multiple boxes. Could Apple be doing multiple boxes?
What you write the TrashCan Mac Pro, just squared.
I Fear Apple witl just announce a tcMP II upgraded to handle ~600W total TDP (even if requiring underclock), and the same shape, just offering updated formula as the mini, Dual Vega, ranging from Dual Vega 20 , for almost fan-less operation, to dual NAVI for serious number crunching and mild AI development. Navi GPUs are expected to be at least 4x-5x more powerful than the D700 being competitive with nVidia Pascal GP100, and pricey, a Dual NAVI tcMP loaded with 18 core xeon likely to cost 10-12K US$ (still a bargain compared with dual GP100 setups close to 20K$)
Apple has some contextual issues around the Mac Pro they didn't have before. The Mini has been moved incrementally 'up'. Desktop Processors instead of mobile and wider range of coverage into the mid-performance zone ( it is't being 'held back' to cover folks on super tight budget or very low end server. ). Same thing with the iMac. Apple "rethought" is coming after spec'ed out where the iMac Pro was going to land and probably where the Mini would land.
The iMac Pro is in part a "rethink" of the Mac Pro ( 2013). There are substantive parts of the old 2006 era Mac Pro market that have left and gone into other Mac products. The 'rethink' should be about who is left and where their needs and growth match up with what the iMac and mini options don't cover.
More than like also part of the "rethink" is about how to weave in T2 (T-series). It won't work to completely close off the Mac Pro. However, it also won't work to simply shop for parts at the Sunnyvale Fry's either and along with the strategic objectives that Mac line up as a whole has.
They should be "rethinking" six Thunderbolt ports as being necessary or desirable. Being in a "port count" pissing match with the iMac Pro and Mini doesn't really buy much substantive value. It is extremely unlikely to have zero ( or just simply two), but six ports is disconnected from current market realities.
Most of all though they need to 'rethink' whether they want the Mac Pro to be a literal desktop system or not. That is a basic driver issue that had a substantive number of impacts. ( is the Mini , iMac , and iMac Pro enough for 'full' literal desktop line up? Yes or no. ). In 2011-12 they felt a need to go literal desktop. If they are still in that camp then what Appleinsider article talks about might be more relevant. If they aren't then much of that Appleinsider article is highly likely off base and misguided.
[doublepost=1541527958][/doublepost]
No.
That write up is sure to get a bunch of clickbait views, but it deeply misguided on several points.
First, this is not substantively true.
" ... But, Apple has very carefully crafted a message about what we should expect —and what they aren't saying is as important as what they explicitly are. ".
One of the premises this article is built around is that Apple has talked alot about the design of the new Mac Pro. They did not. The vast majority of what they talked about in those two April sessions was about current products ( hits and misses ) , but not about details about the future products. Stuff along the lines of 'we have a pro targeted iMac coming that we put lots of effort into" is pretty close to saying the sky is blue. It is a Mac. It has an integrated display. It is priced so you need some substantive income/revenue to justify it. There was no detail there. ( anything like moving to E5 / -W class components. higher end desktop baseline GPU. No 'details' such as that).
The future Mac Pro characteristics were just as warm Apple pie and Boy Scout helping old ladies across the street in specificity.
In short, it has basically been "it will be good, trust us it is worth the wait" message. The primarily thing it is carefully crafted around is not saying anything specific, but making it appears to those who want to run off with a whiff of clue to make up their own fantasy. ( "Oh they admitted they were 100% wrong...they are going back to 'cheese grater' ... Nope never said that. )
Apple has been managing expectations. ('won't be this year (2017). It is 2019 product ( won't be 2018 either). ) That isn't details about the product, but more popping unmotivated ballon filled by others ; not Apple.
Second, This one is also deeply unmotivated.
"... The 2019 Mac Pro will be like the current model and the new Mac mini. One system to which you can add what you need. ".
Since Apple didn't talk about 2019 Mac Pro specific how does it have to be like the current model. The mini didn't change physical form so the Mac Pro can't substantively change physical form? That is a whole lot of hooey. Remember Apple did explicitly talk about what did and did not work with the current Mac Pro. The "did work" was not empty. So yes some characteristics of the current Mac Pro will probably end up in the new one ( e.g., Thunderbolt and bumped to v3 and more than just one controller , Dedicated SSD boot drive (HDDs are not primary drives, some customers made good use of dual GPUs etc.). However there was also stuff that didn't work ( Everyone didn't need two GPUs. Thermal core had issues with unbalanced loads and very high coupling with more the a couple high heat sources, perhaps too high a reliance on all connections for storage capacity (one and only one storage drive, etc.).
The latter if addressed will significantly change the baseline design. If multiple fans and thermal sinks that doesn't mesh at all with baseline design they had. It doesn't necessarily mean retreating 100% back to 2010 design either. The did need to do some rebalancing of the design. I don't think Apple is looking to build somethign that is primary a container of other peoples 'stuff', but not opposite extreme of "no stuff at all" really doesn't look at what is left of the Mac Pro market base after the shifts over the last 9 years either.
The Co-location folks and the "stack to cluster" folks were probably very happy with Apple sticking with the exact same overall form factor with the Mini. I highly doubt any rack/subsystem folks are in 'love' with the 2013 basic design.
[ the 2010 model is a bit rack hostile also. ]
There are other parts to the Appleinsider article but honestly it more so appears that he drew some conclusion he wanted to draw and then tried to retroactively wrap some 'facts' around it to justify it. There is a bunch of "Emperor's New Clothes" there.
What a quite timely comment.
Posted yesterday at AnandTech, a test report on some new Intel Xeon "E" series (embargo lifted):
https://www.anandtech.com/show/13526/intel-xeon-e-review-e2186g-and-more-tested
FWIW, I found this interesting in that it appears to confirm what I'd asked a few weeks ago about Intel's CPU architecture designs, such as Xeon potentially just being an "I" CPU with ECC enabled.
Granted, this may not necessarily be the CPU that we think should be in a Mac Pro, but its still another Xeon option to consider ... and samples have been out there for months for the likes of Dell, HP, (Apple) to have made new prototypes around.
I'm afraid that most of the damage has probably already been done, with the remaining vestiges largely IMO being
the smaller businesses and Prosumers. The real question is one of mindshare brought by the influencers, but as others have pointed out, Apple's gone for the sugar rush of easy money in the iPhone space and won't second-guess their destabilization of core businesses until such time that iOS stumbles - - which in the nature of business cycles always is inevitable. Apple seems to be spending lavishly in other areas where they may claim represent diversification, such as in Services, but when those same services are 90% tied back to service on iOS devices, its not really diversification.
Intel isn't as exciting as the AMD TR 2990WX.
Love to see Apple make the move, now that TB3 doesn't have to be tied to an Intel Chipset.
Love to see Apple make the move, now that TB3 doesn't have to be tied to an Intel Chipset.
Depending upon just how badly Intel has screwed up the Intel W roadmap their could be "Threadripper" Zen2 variants out by Q3-Q4 2019 too. All AMD needs in a smaller and simpler IO die. It wouldn't be that hard to take 4 Infinity Fabric connectors off , simplify the internal crossbar switch and drop 4 DDR controllers. It is removing pieces they would have already done at that point.
I' suspect they won't crank the core count as much as drop the power down to where the Intel W is. ( at least if want uptake from Apple. )
[doublepost=1541541603][/doublepost]
Not sure why he would have to read it since he explicitly basically outline 3 issues that would shift the desing already if Apple "fixed' those. Minimal standard configuration of 2 GPUs ... he outlined that was a problematical issue. The grossly simple solution is to just put one in the standard configs. That's isn't rocket science. Right there the design substantively changed. Especially, when coupled to the unbalanced thermal core issues he brought up.
He also mentioned perhaps leaning a bit too much on Thunderbolt for storage ( admittedly a bit vague there but again the simple solution is to have more than one internal drive. ).
Where they have to fit to a enclsoure. A new Mac doesn't have to commit to either exactly matching the 2013 or 2010 enclosure. Thinner isn't particularly a desktop constraint.
Copper will do no such thing. Magical heatpipes and water coolers aren't either given the same physical volume and fan airflow constraints.
Rome is 225-250W TDP
.The design is extremely modular, extremely simple, and extremely scalable. IPC appears to be slightly higher than Skylake, just based on high-level overview of the architecture, and what they have done. The only thing that can scr*** everything up for AMD are cache latancies, but apparently, they have done improvement on this front, as well.
Jesus, this is the biggest advancement in silicon design we have seen in the past 5-10 years.
P.S. Why would they need smaller IO die, if the CPU package is the same for Epyc and Threadripper?
AMD will use the same IO die for every Epyc 2 config regardless of the amount of chiplets.
Rome is 225-250W TDP
The current Intel W is 140W. Even if there is a 36W bloat with a "Cascade Lake" class tweak that is still 175W which is 50W lower than your minimal above. For much slower base clock speeds. For a datacenter server sure. Some small corner case workstations , maybe. But broad usage single user workstation workloads those are the wrong chips.
If AMD can more their 180W solutions down 160W or so they'd have something that would probably hit Apple's radar ( I don't think the Cascade bloat is going to be that high. wanted conservative, even numbers. )
Possible Memory latencies too because there is nothing 'local'. All the cores are going through I/O die.
Oh Please! Stop. just stop. IBM effectively did this years ago with a admittedly larger MCM design for their mainframes.
P.S. Why would they need smaller IO die, if the CPU package is the same for Epyc and Threadripper?
The socket is a difference size for one. It is cheaper for second.
No. It is the same socket size, even 1:1 pin compatible.
Oh Really?
Rome: 8 Matisse 7 nm chiplet + 1 14 nm IO die.
Matisse, AM4 mainstream desktop: 1, or 2 Matisse 7 nm chiplets + 1 14 nm IO die.
APUs: 1 Matisse 7 nm chiplet + 1 14 nm IO Die + 7 nm GPU chiplet + HBM.
Did IBM ever designed something this scalable for ALL of their platforms?
THIS is the best silicon design we have seen past 5-10 years, biggest advancement, and biggest innovation.
Last edited:
Don't get overhyped up on MCM solution, as this is like moving 2 step backward to make 1 step forward. It is just clever workaround to make single processor design to address many different markets. AMD obviously went this direction simply because they have limited resources on R/D and it will provide better margins for them than designing big monolithic chip on 7nm.
There is a reason those companies spent a decade integrating literally everything on CPU die, yet now they have to go backward since process shrinks are getting far more expensive than previous nodes and no longer gives huge benefits previous process improvements provided.
Last edited:
