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

[Zdigital2015]

First, the basis of why the last of the cMP was a disappointment was because of a relevant lack of hardware updates back in that time period, with an example being that SATA3 (ratified 2008) was never implemented.

Pretty sure that neither the x58 nor the 5520 PCH natively supported SATA 3 or USB 3.0. Motherboard vendors had to use third-party controllers for add this functionality and that is something that Apple does not do. When Intel natively supports it, they support it. USB waited until Ivy Bridge is just one example.

 

[apolloa]

And apparently all it costs is an arm!

So you get the leg as change then? Not too bad lol.

 

[AidenShaw]

Pretty sure that neither the x58 nor the 5520 PCH natively supported SATA 3 or USB 3.0. Motherboard vendors had to use third-party controllers for add this functionality and that is something that Apple does not do. When Intel natively supports it, they support it. USB waited until Ivy Bridge is just one example.

If your claim is true, then why does the MP6,1 have a third party Fresco Logic FL1100 4-port USB 3.0 Host Controller on the board and a third party Intel DSL5520 Thunderbolt 2 Controller? https://www.ifixit.com/Teardown/Mac+Pro+Late+2013+Teardown/20778#s56752

(Just one example of many third party chips on Apple systems - the Imac Pro has a third party AQUANTIA AQtion AQC107-B1-C 10 GbE controller and third party Intel T-Bolt controllers https://www.ifixit.com/Teardown/iMac+Pro+Teardown/101807#s191188 .)

Your claim is easily shown to be false. Apple has a long history of using third party IO controllers on their systems.

Sometimes, though, Apple has been pig-headed and blamed Intel for not supporting the latest stuff - whereas the rest of the industry realizes that the design cycle for platform controller hubs is longer than the cycle for third party controllers, and uses third party controllers to bridge the gap until the PCH supports a new standard.

I don't understand why some Apple fans accept and propogate that deflection.

 

[deconstruct60]

Up to 56 cores!!!! That must be one big chip..

From Intel press kit

The 9200 is the AP package version . It is huge . After putting that on a logic board with 12 DIMM slots clustered around it , a relatively very large chunk of horizontal space will be gone. Not going to see that package normally coupled to several generic standard full length , high power ( volume) PCIe slots . ( probably more often to m.2 , OCP NIC , FPGA card , 1-2 compute card , or bridge to PCI-e cage )

 

[Joe The Dragon]

only 40 3.0 pci-e lanes per socket.

AMD has 128 with one cpu soon to be pci-e 4.0 with rome

 

[Zdigital2015]

If your claim is true, then why does the MP6,1 have a third party Fresco Logic FL1100 4-port USB 3.0 Host Controller on the board and a third party Intel DSL5520 Thunderbolt 2 Controller? https://www.ifixit.com/Teardown/Mac+Pro+Late+2013+Teardown/20778#s56752

(Just one example of many third party chips on Apple systems - the Imac Pro has a third party AQUANTIA AQtion AQC107-B1-C 10 GbE controller and third party Intel T-Bolt controllers https://www.ifixit.com/Teardown/iMac+Pro+Teardown/101807#s191188 .)

Your claim is easily shown to be false. Apple has a long history of using third party IO controllers on their systems.

Sometimes, though, Apple has been pig-headed and blamed Intel for not supporting the latest stuff - whereas the rest of the industry realizes that the design cycle for platform controller hubs is longer than the cycle for third party controllers, and uses third party controllers to bridge the gap until the PCH supports a new standard.

I don't understand why some Apple fans accept and propogate that deflection.

Because at the end of the day, a solution supported natively by the PCH is going to require the least amount of work to support and implement from an engineering standpoint and is going to be the most compatible over the long term since Intel has a vested interest in making sure they get it right out the door. Also, third party controllers are sometimes rushed to market and suffer from major issues once they end up out in the real world which then become Apple’s headache. The whole wide Windows world of DIY PCs is littered with examples of this and how sometimes the device free-for-all advocated ends up being a bad thing, especially for a professional workstation. I suspect that HP’s approach to their Z-Series is a bit more disciplined than with their Pavilion lineup. At least, lets hope so.

When Apple started engineering the 2013 Mac Pro, Intel still did not have support for USB 3.0 in the C600/x79 PCH even though USB 3.0 had been ratified in 2008, with products appearing in mid-2009. The same thing happened with the MacBook Pro in that users had been clamoring for USB 3.0 for a while and while Thunderbolt had been standard on MacBook Pros for 16 months, USB 3.0 still wasn’t supported natively by Intel until Ivy Bridge. Apple added support for USB 3.0 with the mid-2012 non-Retina and Retina MacBook Pros when they updated them to Ivy Bridge. Adding a third party controllers eats into board space and on a MacBook Pro, that space is at a premium.

I am sure Apple was reluctant to choose to use a third-party USB 3.0 controller, but there was no way that they were going to ship the Late 2013 Mac Pro with USB 2.0. Sometimes, those compromises have to be made.

Yes, Apple uses an Aquantia 10GBe controller on the iMac Pro. I guess Intel’s controllers did not give them what they wanted. However, this is not exactly ground breaking or earth shattering. The OP mentioned USB 3.0 and SATA 3.0, neither of which were natively supported and by the time qualified third party solutions were available (2010), it was moot as the classic Mac Pro was in “maintenance mode” by that time since Apple did all the major engineering with the Early 2009 Mac Pro. The Mid 2010 and Mid 2012 Mac Pros were relatively minor updates and were not going to end up getting any additional hardware (board level) enhancements, since Apple’s focus was entirely on the 6,1 Mac Pro.

Sorry, but Intel’s Thunderbolt controllers are not “Third-Party”...nice try, though. Intel promised to integrate TB3 functionality into the CPU die, but here we are in 2019 still waiting.

I am not propagating a deflection, but I am trying to give a rational explanation as to why Apple might not have added those features during the cMP’s (2009-2012) lifetime. Both USB 3.0 and SATA 3.0 would have been useful additions to the cMP, it just didn’t happen. Both of those standards were ratified well after engineering got started on the Early 2009 Mac Pro and usable third party controllers weren’t available when the early 2009 Mac Pro was released. Well qualified and stable solutions didn’t come on the scene until 2010 and even by 2012 when Intel finally integrated it, a few issues still existed.

With few exceptions, Apple waits until Intel adds support to the PCH.

 

[AidenShaw]

only 40 3.0 pci-e lanes per socket.

48 lanes....

 

[apolloa]

From Intel press kit

The 9200 is the AP package version . It is huge . After putting that on a logic board with 12 DIMM slots clustered around it , a relatively very large chunk of horizontal space will be gone. Not going to see that package normally coupled to several generic standard full length , high power ( volume) PCIe slots . ( probably more often to m.2 , OCP NIC , FPGA card , 1-2 compute card , or bridge to PCI-e cage )

WWOOOOOWWWWWWWW!!!!

That’s one huge chip! It will have to offer one hell of a performance advantage for the rack space the machines that will use them will take up!

I can see why their is no chance what so ever any new Mac will use them! Pro or server.

 

[Zdigital2015]

Are we expecting Apple to use Xeon-SP in the Mac Pro, or will they instead go with the Basin Falls refresh of the current W-Series in the iMac Pro?

I have to believe that they are going to go with Xeon-W CPUs at this point. This will limit the CPUs that they are able to use, both in terms of max number of cores (18 right now) and leave out dual CPU configs, but I think they are okay with that. Apple is 7 years out from dual Xeon workstation and the LGA-3467 is not for the faint-hearted...although, with the time it has taken to engineer the new Mac Pro, I have to wonder if they decide to make the investment in engineering a 1, 2 or 4 CPU modular beast to get them through the next 5-7 years. I sure hope they showcase the new Mac Pro at WWDC.

 

[wallysb01]

WWOOOOOWWWWWWWW!!!!

That’s one huge chip! It will have to offer one hell of a performance advantage for the rack space the machines that will use them will take up!

I can see why their is no chance what so ever any new Mac will use them! Pro or server.

I'd imagine that interns of total space per core, accounting for an entire package (case/racks, all of it), the 9200 series will be the best way to go. It better be one really well cooled room though, holy moly. And yeah, Macs aren't likely to use the scalable line regardless (though I wish they would).

 

[CWallace]

I have to believe that they are going to go with Xeon-W CPUs at this point. This will limit the CPUs that they are able to use, both in terms of max number of cores (18 right now) and leave out dual CPU configs, but I think they are okay with that. Apple is 7 years out from dual Xeon workstation and the LGA-3467 is not for the faint-hearted...although, with the time it has taken to engineer the new Mac Pro, I have to wonder if they decide to make the investment in engineering a 1, 2 or 4 CPU modular beast to get them through the next 5-7 years. I sure hope they showcase the new Mac Pro at WWDC.

I am strongly inclined to agree. While Apple like to say "no compromises" in relation to the MP 7,1 design, using Xeon SP is going to (seriously) raise the price on an already likely expensive piece of kit.

And macOS doesn't serve a fair number of markets that need/benefit from Xeon SP performance (partly due to the obsolescence of the MP6,1, partly due to Apple intransigence (no nVidia support, no internal storage bays, no internal expansion/connectivity) and partly because macOS was not a market even when they did have current-generation OEM-agnostic hardware (MP5,1). Xeon W would still provide significant performance for the markets macOS still is relevant for and would make it viable for some markets that had moved on due to the 6,1 no longer being rational from a price/performance basis and whom do not wish to undergo the limits of an AIO (iMac Pro) or SFF (Mac Mini) solution.

 

[Zdigital2015]

I am strongly inclined to agree. While Apple like to say "no compromises" in relation to the MP 7,1 design, using Xeon SP is going to (seriously) raise the price on an already likely expensive piece of kit.

And macOS doesn't serve a fair number of markets that need/benefit from Xeon SP performance (partly due to the obsolescence of the MP6,1, partly due to Apple intransigence (no nVidia support, no internal storage bays, no internal expansion/connectivity) and partly because macOS was not a market even when they did have current-generation OEM-agnostic hardware (MP5,1). Xeon W would still provide significant performance for the markets macOS still is relevant for and would make it viable for some markets that had moved on due to the 6,1 no longer being rational from a price/performance basis and whom do not wish to undergo the limits of an AIO (iMac Pro) or SFF (Mac Mini) solution.

I keep looking at the Xeon Scalable Processor lineup and I think this was mentioned in Anandtech’s coverage of the introduction of the second generation that they all seem to be much more suited to data center and server than to the workstation market.

The sheer size of the LGA 3647 socket seems to preclude anything that compliments Apple’s packaging (or lack thereof, depending on who you ask) skills, not mention the power requirements. I also see them wanting to keep a modicum of continuity between the iMac Pro and the new Mac Pro from a parts perspective since it helps drive down costs. The issue I have is will that be enough, even to serve markets that they have a strong presence in (if any).

The flip side is that they build one Mac Pro use only the 28-core 3175X and then the building blocks are to expand functionality of the base CPU core module and move on. Reality says that is a bad idea, but my spider sense says don’t count it out just yet.

 

[deconstruct60]

only 40 3.0 pci-e lanes per socket.

it is a BGA processor package. Intel pretty much knows exactly in advance which motherboard designs they are composing this MCM (multiple chip module ) for. If only provisioning 1-2 x16 slots (or switches) then have 8-24 left, which is a pretty good budget for most HPC nodes. ( where most persistent storage is 'far' away. ) and still have the PCH lanes for local boot drive. These packages are being made for a specific set of purpose; not Bubba Joe's Frankenstein project.

There probably will be a few "pimp my ride" workstation boards with AMD Eypc doesn't cover .... which is why there is the vast majority of the rest o the SP line up to fill those.

AMD has 128 with one cpu soon to be pci-e 4.0 with rome

It is only 128 if only have one CPU which if AMD tried to copy what Intel is doing here ... won't have. ( 64 of that budget are repurposed to Infinity Fabric in a multiple Package set up. Same way some of that 20-40 on the Intel PCH chips is FlexIO to SATA/USB/etc for certain lanes. )

Intel's PCI-e 4.0/5.0 CPUs are probably not very far away. Their just announced Agilex FPGA line up has PCI-e 4.0 (this year) and PCI-e 5.0 ( the variants probably early next year or very late this one ) connections. And they'll work with the current PCI-e SP /AP processors.

Rome is going to get several months "in the Sun" where AMD will be able to do some chest beating. Is that going to last a whole year worth of months ...... perhaps not. Note that the Agilex FPGA are shipping this year in 10nm. The ramp to volume 10nm isn't going to be multiple more years. Implicitly the ramp is starting about mid 2019 when these FPGA start going out the door. Once the SP's get to 10nm they'll extremely likely match up to these FPGA's.

In the context of the Mac Pro, 128 lanes isn't likely particularly necessary. 48 is enough and that could easily be filled by the Xeon W ( or the 60 of Threadripper. )

 

[edgerider]

lets face it if they give us a 18 core xeonw with 4 pcie slot, m2 support, and thunderbolt 3 most of us will be happy.

I don’t see how we could benefit from scalable xeon from a financial stand point. No way i’m spending 30k on a mac.

 

[wallysb01]

I am strongly inclined to agree. While Apple like to say "no compromises" in relation to the MP 7,1 design, using Xeon SP is going to (seriously) raise the price on an already likely expensive piece of kit.

And macOS doesn't serve a fair number of markets that need/benefit from Xeon SP performance (partly due to the obsolescence of the MP6,1, partly due to Apple intransigence (no nVidia support, no internal storage bays, no internal expansion/connectivity) and partly because macOS was not a market even when they did have current-generation OEM-agnostic hardware (MP5,1). Xeon W would still provide significant performance for the markets macOS still is relevant for and would make it viable for some markets that had moved on due to the 6,1 no longer being rational from a price/performance basis and whom do not wish to undergo the limits of an AIO (iMac Pro) or SFF (Mac Mini) solution.

I haven't seen a whole lot of rumors on the Xeon W 22xx Cascade lakes. Are they likely to get 2-4 core increase and some extra clock rate similar to the scalable line?

Right now the top end iMac has the i9-9900k which is 8 cores for ~$550. The nerfed 2145 in the iMac Pro is 8 cores for around $1000. Presumably they keep the iMac Pro in that range, so could be a 10 core (successor of that 2140B) to differentiate it from the top of the line iMac. But then does the Mac Pro need to start so high? Maybe an 8 core for it (successor of the 2133) or the high clocked 6 core (successor of the 2125)? If so, I'll be interested to see where that puts pricing. If following the iMac - iMac Pro model, it would just clear out of the sane, but top end BTOs of the Mac Mini (top i7 6 core, 32GB RAM, 1TB SSD, 10 Gigabit) is $2600. That would put a potential 6/8 core Mac Pro with similar specs in that $3000 range. Wild guess: 6 core - $3000, 8 core - $3500, 10 core - $4500, 12 core - $5300, 16 core - $6100, 20 core - $7000, without all those options actually realized, maybe? 6, 10, 20 or 8, 12, 20 Might cover the range nicely.

 

[deconstruct60]

48 lanes....

Anandtech article says 40 ( that could be a goof/typo ). Oddly, ark.intel.com is entirely silent on the 9242 and 9294 spec page. There are 4 QPI links coming out of this mash up. There is a pretty decent change Intel is "fixed assigning" some lanes to the OmniParth/PCH chipset use for higher boost network. Since the chip is BGA and logic board largely is licked into an Intel design they probably appropriated those to the PCH chipset (and its configuration) that they selected also. Off package perhaps 20 lanes per die/chip to balance out the I/O load and a decent chunk assigned to 10GbE (or more ) network I/O that have to buy from Intel also. ( and that helps put an upper boundary on the MCM cost also. Some stuff they just don't route out. )

 

[danwells]

The new chips may portend an incoming Mac Pro (WWDC?). I've always maintained that the Mac Pro is either going to be off the shelf Xeon-SP or a variant on Xeon SP. - but it'll use the big socket... Intel can obviously make variant chips relatively easily (and is willing to do it)) - witness the 28-core Xeon Platinum variant that became the "Xeon-W 3175" - they took out the multi-processor capability (and a few other server-only features?), raised the clock speeds a bit and sell it for 1/3 the price).

Apple is a much bigger company than the few small vendors likely to offer Xeon-W 3175 based systems. The 3175 won't work in standard Xeon-SP boards (not sure why) nor does it work in standard Xeon-W boards (wrong socket) - a custom board to support one chip is too much trouble for HP and other workstation vendors, no matter how appealing the chip. The 3175 has a great price/performance ratio compared to standard Xeon-SP, and, other than ridiculously expensive motherboards, it isn't terrible by ultra high-end desktop or workstation standards (no, it's not as good as Threadripper - AMD has a price/performance advantage right now).

If those few little companies can request the 3175, Apple should be able to get a nice line of big-socket chips ranging from 10-12 cores to 28 cores, without the hugely expensive multiprocessing and server features - similar chips to the 3175. They may even be able to get some dual-socket capable chips, without the 4-way and 8-way features - Intel charges for dual socket capability, but they really charge for anything more than that - and nobody uses more than two sockets in a workstation (even the $100,000 version of the Z8 is dual 28 core).

Apple won't get the "single" 56 core chip... Since it's not socketed, it's an entirely separate board design, and Apple won't bother with a different board, plus an extra couple hundred watts of power demand, for one ultra high-end CTO option that might sell a few thousand units per year (essentially all to Hollywood), if that many. They are far more likely to offer a dual 28-core than a single 56-core (the board design is probably actually closer).

Something like this...
12, 18, 24, 28 core options
possibly
dual 18, dual 24 and dual 28 cores at the extreme high end (no point in a dual 12 core).

Those all exist with viable clock speeds in the new Xeon Gold and Xeon Platinum ranges - look for Apple to get variants with the 4 and 8 way multiprocessing stripped out for more reasonable prices.

 

[deconstruct60]

I haven't seen a whole lot of rumors on the Xeon W 22xx Cascade lakes. Are they likely to get 2-4 core increase and some extra clock rate similar to the scalable line?

Except at the very bottom .... probably not on the core count. The SP old/new line up starts at 6 and goes to 28 (fitlering out the 9200's ). Intel didn't increase the core count everywhere in the SP line up... just a very few specific positions in the line up.

https://www.servethehome.com/second-generation-intel-xeon-scalable-sku-list-and-value-analysis/2/
[There is a gen1 versus 2 table there. Where they cranked clock they didn't particularly increase core count. Where they bumped core counts the clock generally does not increase. ]

For high clock targets ones they basically kept the prices high (or higher). The whole W line up is higher clocked ( but more reasonably priced.).

Intel might toss the 4 core model at the bottom. First , they never did cover that with the Core i9 variants with these dies ( because mainstream covered 4 cores extremely well with both Xeon E (formerly Xeon E3) and Core options. Now that those are moving up to 6-8 that 4 has even more problems.

It is likely that Intel will 'light up' more of the L3 cache on some parts of the lower end. The clock speed will get a bump (and the TDP may creep a bit). But because they bumped the clocks the core count probably will not increase. It is a a coin-toss as to whether they adjust the pricing ( they should ... but perhaps in deep, deep denial. ). 6 cores start at $617 ( and about a $200 jump from 4 cores ) when crossed with Ryzen 3 and Threadripper 3 is going to be bad. Not to mention the Core i9's and i7's in the 6+ range. if they pull the 6-18 down $100-150 then have more slots to put in more than one of the >18 models.

Right now the top end iMac has the i9-9900k which is 8 cores for ~$550. The nerfed 2145 in the iMac Pro is 8 cores for around $1000. Presumably they keep the iMac Pro in that range, so could be a 10 core (successor of that 2140B) to differentiate it from the top of the line iMac.

Unless Intel moves the price points of Intel W .... I suspect the 8 core iMac Pro will stick around. I/O and GPU rather than core count is the differentiator. It isn't huge but if Intel doesn't move the W costs down much then cranking the base iMac Pro price even higher isn't going to help much.

But then does the Mac Pro need to start so high? Maybe an 8 core for it (successor of the 2133) or the high clocked 6 core (successor of the 2125)? If so, I'll be interested to see where that puts pricing. If following the iMac - iMac Pro model, it would just clear out of the sane, but top end BTOs of the Mac Mini (top i7 6 core, 32GB RAM, 1TB SSD, 10 Gigabit) is $2600.

Not just what the base CPU, but also the base GPU. If they went with a Polaris ( or Navi if delay is long enough) they could have lower Bill of Material (BOM) costs on GPU also. ( or a Vega 48 instead of the 56 in the iMac Pro if AMD has firesale pricing on those and flushes Apple's quota commitment. ). One of the issues with Mac Pro 2013 for a signficant group was that they didn't need that high of a GPU BOM budget.

That would put a potential 6/8 core Mac Pro with similar specs in that $3000 range. Wild guess: 6 core - $3000, 8 core - $3500, 10 core - $4500, 12 core - $5300, 16 core - $6100, 20 core - $7000, without all those options actually realized, maybe? 6, 10, 20 or 8, 12, 20 Might cover the range nicely.

Even more so if systems have ability of a second (or more ) internal storage and entry level SSD size set to something like 512GB. ( or Apple stops inflating their 1TB quite so high. )

 

[deconstruct60]

Pretty sure that neither the x58 nor the 5520 PCH natively supported SATA 3 or USB 3.0. Motherboard vendors had to use third-party controllers for add this functionality and that is something that Apple does not do.

More along the lines of Apple does not want to do than universally true. For most of Apple's system they are space constrained (laptops versus big box desktop ). So functionaliy in the PCH not only comes for "free" (well ... paid for anyway regardless of what else do) but also typically comes in smaller volume than a PCH + 3rd party chipset.

When Intel natively supports it, they support it. USB waited until Ivy Bridge is just one example.

there is also a bit of lazy there too since don't have to support a wide variety of USB chipsets either out of the box. Part of this is Scrooge McDuck motivated too. On Apple part too though some of this is also a shake of version 1.0 support too ( typically standards have bugs that get worked out in the first year. That's one reason they typically don't flow into the PCH months after the standards pass. ). Apple passes on those also. (e.g., Apple passed on the initial generation of Thunderbolt 3 power "helper chips" that went on to cause problems for most of the other vendors who raced to "ship first" status on TBv3. )
[doublepost=1554329880][/doublepost]

If your claim is true, then why does the MP6,1 have a third party Fresco Logic FL1100 4-port USB 3.0 Host Controller on the board and a third party Intel DSL5520 Thunderbolt 2 Controller? https://www.ifixit.com/Teardown/Mac+Pro+Late+2013+Teardown/20778#s56752

That was also driven in part though by the Thunderbolt 2 dock that were rolling out with USB 3.0. Apple was going to have to pick some 3rd party USB controller to focus on to guide folks doing docks targeted at Mac laptops. That was Fresco Logic. Then also the server/workstation PCH typically lag behind the maintream PCH ( every 18-24 months to new socket change versus multiple PCH for same socket in mainstream). So there was pretty high overlap between the Mac Pro 2013 and what the TB docks needed as far as USB controller support.

The MP 2013 also had lanes and space to give since not a laptop or highly space sostrained. The Mac Mini and iMac where on mainstream PCH so the USB 3.0 controller can for "free".

Although it came much later I'd suspect that the LG Utlrafine 5K had Freco logic too. It wouldn't be surprising if there was some aborted thunderbolt display docking station that never made it out of the Apple labs with Fresco logic USB in them too. ( a 4K to be somewhat contemporary with the MP 2013 ( 2014-2015) that got aborted when Apple latched onto the idea of pushing the iMac 27" to 5K and not having a one cable solution. )

(Just one example of many third party chips on Apple systems - the Imac Pro has a third party AQUANTIA AQtion AQC107-B1-C 10 GbE controller and third party Intel T-Bolt controllers https://www.ifixit.com/Teardown/iMac+Pro+Teardown/101807#s191188 .)

That's seems to be more of a partial Intel 'fail' than an Apple want. Intel kept promising affordable 10GbE but as they went more data center centric focused just couldn't get off the margins of 10GbE. It got incrementally cheaper but never got to affordable. The inertia that wanted 2.5-5GbE over their legacy cabling was something Intel entirely underestimated. 10GbE to mainstream PCH doesn't appear to be anywhere on Intel's path. Pushing more WiFi into the PCH has much higher priority. ( and frankly besides desktop ... Apple is 'done' with Physical Ethernet. )


Thunderbolt controllers. It still remains to be seen those are pragmatically PCH merge-able . it is highly likely this is going to look like Intel's Ethernet and now Wifi "embedded" solutions that actually aren't 100% merged. Both of those require a PHY (physical) chip that is still separate from the PCH. Thunderbolt is even more likely going to be in that zone because there are placement constraints on the PHYS chip to physical socket that aren't going away. What more likley to get is "1/2" of a Thunderbolt you already paid for anyway and an incrementally cheaper PHYS chip to provision the whole solution. There will still be an external (to PCH) discrete chip.

Sometimes, though, Apple has been pig-headed and blamed Intel for not supporting the latest stuff - whereas the rest of the industry realizes that the design cycle for platform controller hubs is longer than the cycle for third party controllers, and uses third party controllers to bridge the gap until the PCH supports a new standard.

In the 2010-2012 era.... there were complaints about the Mac Pro not being upgraded to put a SATA chip on the motherboard ( like HP , Dell, Lenovo). Also complaints that Apple didn't put a "Real Raid" chip on motherboard like some workstations. If other changes were not also coming that would a be "tail wags dog" kind of change.

That issue is more so about iteration periodicity than on putting chips on. From 2009, Apple has substantively been tapping the brakes (at least ) or full on in hibernation mode ( 2015-2016) in Mac Pro space. Yeah that really wasn't Intel.

 

[ssgbryan]

I haven't seen a whole lot of rumors on the Xeon W 22xx Cascade lakes. Are they likely to get 2-4 core increase and some extra clock rate similar to the scalable line?

Right now the top end iMac has the i9-9900k which is 8 cores for ~$550. The nerfed 2145 in the iMac Pro is 8 cores for around $1000. Presumably they keep the iMac Pro in that range, so could be a 10 core (successor of that 2140B) to differentiate it from the top of the line iMac. But then does the Mac Pro need to start so high? Maybe an 8 core for it (successor of the 2133) or the high clocked 6 core (successor of the 2125)? If so, I'll be interested to see where that puts pricing. If following the iMac - iMac Pro model, it would just clear out of the sane, but top end BTOs of the Mac Mini (top i7 6 core, 32GB RAM, 1TB SSD, 10 Gigabit) is $2600. That would put a potential 6/8 core Mac Pro with similar specs in that $3000 range. Wild guess: 6 core - $3000, 8 core - $3500, 10 core - $4500, 12 core - $5300, 16 core - $6100, 20 core - $7000, without all those options actually realized, maybe? 6, 10, 20 or 8, 12, 20 Might cover the range nicely.

6 to 8 cores is 2015 thinking. That is Ryzen territory, not workstation territory.

32 core (64 thread) Eypc, 128Gb ram, 1Tb M.2, WX5100 video card, 1,300w PSU = $6,500 today, not at some undetermined point in the future.

 

[fuchsdh]

I am strongly inclined to agree. While Apple like to say "no compromises" in relation to the MP 7,1 design, using Xeon SP is going to (seriously) raise the price on an already likely expensive piece of kit.

And macOS doesn't serve a fair number of markets that need/benefit from Xeon SP performance (partly due to the obsolescence of the MP6,1, partly due to Apple intransigence (no nVidia support, no internal storage bays, no internal expansion/connectivity) and partly because macOS was not a market even when they did have current-generation OEM-agnostic hardware (MP5,1). Xeon W would still provide significant performance for the markets macOS still is relevant for and would make it viable for some markets that had moved on due to the 6,1 no longer being rational from a price/performance basis and whom do not wish to undergo the limits of an AIO (iMac Pro) or SFF (Mac Mini) solution.

Yeah I kind of feel like the landscape has shifted pretty dramatically since the 5,1 in Intel's offerings. -SP options are generally some very wonky cores versus clock speed configurations that feel very much about datacenters more than a lot of mainstream workstation needs.

It'd be nice to get high numbers of cores for cheaper and more DIMMs, but Intel isn't really offering the kinds of products that make sense there.

 

[wallysb01]

6 to 8 cores is 2015 thinking. That is Ryzen territory, not workstation territory.

32 core (64 thread) Eypc, 128Gb ram, 1Tb M.2, WX5100 video card, 1,300w PSU = $6,500 today, not at some undetermined point in the future.

Don't get me wrong. I don't disagree with you. I have a 20 core, 256GB ram, 1Tb SSD, 40TB RAID10 beast behind my monitor (and it was like 8K total, I think?). I just don't think this is the way Apple is going. I'll be pleasantly surprised if the SP line ends up in the Mac Pro, but I'm not counting on it.

 

[jscipione]

If your claim is true, then why does the MP6,1 have a third party Fresco Logic FL1100 4-port USB 3.0 Host Controller on the board and a third party Intel DSL5520 Thunderbolt 2 Controller? https://www.ifixit.com/Teardown/Mac+Pro+Late+2013+Teardown/20778#s56752

The Mac Pro 6,1 Intel C602J Server/Enterprise chipset has no support for USB 3.0 and thus the FL1100 USB3 add-on was needed. AFAIK even the latest Intel chipsets don't support Thunderbolt and still require a separate controller. The C602J certainly doesn't support Thunderbolt.

https://ark.intel.com/content/www/us/en/ark/products/66243/intel-c602j-chipset.html

 

[ssgbryan]

Don't get me wrong. I don't disagree with you. I have a 20 core, 256GB ram, 1Tb SSD, 40TB RAID10 beast behind my monitor (and it was like 8K total, I think?). I just don't think this is the way Apple is going. I'll be pleasantly surprised if the SP line ends up in the Mac Pro, but I'm not counting on it.

Unfortunately, I have to agree with you. I don't think the next "Mac Pro" will be a workstation (as the workstation market sees them).

In 2019 that means either Nvidia (for GPU intensive tasks) or high cores (Eypc) - neither of which today's Apple seems interested in.

 

[seek3r]

Unfortunately, I have to agree with you. I don't think the next "Mac Pro" will be a workstation (as the workstation market sees them).

In 2019 that means either Nvidia (for GPU intensive tasks) or high cores (Eypc) - neither of which today's Apple seems interested in.

I have a feeling the larger continuing move to cloud or other remote machines for heavy lifting for a lot of their pro consumers is part of that. I routinely use dozens of cores for work, and used hundreds and thousands when I was on the HPC side of things... and none of that's running on my local machine. It's why I can use a laptop with a dock as my primary workstation at all.

Hell, I have a quad 2016 15" MBP for work right now, probably going to have IT replace it this summer and I'm leaning towards a quad 13" instead of the equivalent-to-my-current-machine hex 15" because I'd rather have the portability than the power (I also dont use the dedicated GPU, so for people with GPU intensive local workflows it's a different story).

I may consider the new mac pro for my home desktop to replace my 2010 but I doubt I would get more than whatever the base model is core count wise (especially if it's socketed ).

There are plenty of people who do really heavily use those extra core counts locally, but even among professionals they're getting rarer. Or, put another way: the increase in cores/socket is outpacing the local need for cores/socket. So while the average machine for a pro is hex or quad now maybe (it may even be a dual still) and may become hex, oct, or 12, maybe even 18-24 cores within a few years... during that same time we may be up to 100 or more cores/CPU as common for xeons/epyc.