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
i9 has 44 pcie lines, max 128 GB NON-ECC ram
Xeon W has 48 pcie lines, max 512 GB ECC ram
Not a big difference, but Xeon has a bit more IO for one more SSD or TB3, and it supports a lot more memory.
Ok I think that answers it for me, I missed the ram limit somehow. Sometimes you miss the obvious, and I'm totally guilty of that this time. My bad.
But the supported RAM limit on the MP6,1 is 64 GiB. Surely doubling that would be adequate?
But the supported RAM limit on the MP6,1 is 64 GiB. Surely doubling that would be adequate?
Well, I recall when the AMD CPU rumours were getting started, there was all sorts of sabre rattling that 768GB was an affront to all that was holy. Even though the 768 limit was an issue of dimm memory density ( for lack of correct terminology ).
But here, I offer you a cookie and a toast in appreciation of the cheeky and the snarky
.The merits of near-identically spec'd i9 vs Xeon-W are debatable, especially in an iMac Pro form factor where the memory is limited to 128GB regardless.
In any case, it doesnt surprise me that they equipped it with the W given it is position as a workstation product and the pricing for what you get isnt obscene. I've got one on order for use as a footage handling machine in our production studio, replacing a trash can.
In any case, it doesnt surprise me that they equipped it with the W given it is position as a workstation product and the pricing for what you get isnt obscene. I've got one on order for use as a footage handling machine in our production studio, replacing a trash can.
Xeon SP looks designed for workloads that are parallel in nature and benefit from a large number of cores across multiple CPUs. So if Apple adopts this for the Mac Pro, then really it's only going to make sense in a dual-CPU configuration because otherwise the iMac Pro should outperform a Mac Pro with a single Xeon SP processor thanks to the higher base and turbo clock-speeds of the W-Series (yes, there are select Gold series that can compete with the W's, but they have MSRPs $1000 higher).
So if we do see multiple Mac Pro SKUs, I think they could be dual-CPU and quad-CPU configurations with the iMac Pro addressing the single-CPU segment of the market.
Yes, a Xeon W and Core i9 are pretty much identical outside of ECC support and the additional PCIe lanes, though the Xeons will be the "best of the bin" parts (highest-graded during testing and validation).
I think the requirement (within the workstation market) required Apple to use Xeons and ECC memory.
I know the Xeon-W came out after Core i9, but I cannot help but wonder if Skylake-X was meant to be a workstation CPU and the arrival of AMD's ThreadRipper forced Intel's hand to offer a consumer version (the Core i9), as well. The predecessors to Skylake-X - Haswell-E and Broadwell-E - were branded as "enthusiast" CPUs aimed at gamers and were higher-clocked versions of the baseline CPUs with more cores. Skylake-X, on the other hand, has architectural changes compared to the lower-tier Skylake CPUs.
Last edited:
They could do a single socket SP variant, there are some gems in the SP lineup that would work for it. The 6154 has 18 cores at 3.0 Ghz and with a physically larger socket, better heat dissipation than a comparable W that runs at 2.6. So its not like the old model of single or dual CPU daughterboard approach of the cheesgrater days couldn't work.
Our dual 6154 systems we just got in are running on par with the dual 7601 EPYC in rendering applications. It'd be ideal for the Mac Pro.
Our dual 6154 systems we just got in are running on par with the dual 7601 EPYC in rendering applications. It'd be ideal for the Mac Pro.
True, but it does have a $1000 higher MSRP. Then again, Apple could just design this as a "cost is no object" product with a price tag to match, though I get the feeling people are expecting single-core base model Mac Pros to be cheaper then the equivalent iMac Pro due to the omission of the 5K display so if they come in at the same price or higher...
True, but it does have a $1000 higher MSRP. Then again, Apple could just design this as a "cost is no object" product with a price tag to match, though I get the feeling people are expecting single-core base model Mac Pros to be cheaper then the equivalent iMac Pro due to the omission of the 5K display so if they come in at the same price or higher...
From the wishful thinking bucket, I like the Xeon SP
But as pointed out, it is more likely that Apple will reuse the iMacPro engineering for cost reasons. I'd have no problem with the base model being on par cost wise to make up the premium. It reminds me a humorous parable that suggests the key to effective distribution when the field is designed for presupposed ideal number of options ; is to make each choice as sh*tty as the next, avoid any runaway winners.
PS Nugget. I <3 you.
You’re right, I forgot about the 18 core i9. Cache probably depends on the model, but yes, that is true. As someone pointed out, a few more lanes as well. I’d be willing to bet that the Xeon chips are binned to a higher standard as well. And of course, some people need ECC, or at least would like to have it...and I suppose it’s far more fair to blame intel for charging 2-3x as much for a W series as an i9 for minimal features.
[doublepost=1515572251][/doublepost]
True, but it does have a $1000 higher MSRP. Then again, Apple could just design this as a "cost is no object" product with a price tag to match, though I get the feeling people are expecting single-core base model Mac Pros to be cheaper then the equivalent iMac Pro due to the omission of the 5K display so if they come in at the same price or higher...
I’d be willing to bet that the Mac Pro will have 2x the price spread of the iMac Pro, maybe even more so. I’m sure there’s a chance I’m wrong, as we’re all pulling stuff out of our own asses at this point, but I can certainly see them having 2 different CPU configs, that alone is a massive spread. And I think that’s a good thing, it allows the machine to be flexible.
I definitely think that's possible. The daughterboard setup would at least make having different sockets less troublesome.
Again, it all depends on what Apple's thinking the role of the Mac Pro should be now that the iMac Pro exists. Given the $5K starting price it seems to me that you'd want something that fits in underneath it as a base, and that basically requires -W processors, but going higher end past the iMac Pro probably requires dual sockets (aside from the fact that something like 8DIMMs and swappable GPUs would already put it as far more flexible—and expensive—in the high end, even if it remained a single-socket machine.)
Will we?!
[doublepost=1515603069][/doublepost]A lot of movement in the display segment.
8K, 4K HDR in larger formats with TB3 DP1.4, good.
News on the 5K TB3D would be nice too.
[doublepost=1515603069][/doublepost]A lot of movement in the display segment.
8K, 4K HDR in larger formats with TB3 DP1.4, good.
News on the 5K TB3D would be nice too.
Shall we open the betting pool for timeframes on Apple-supplied news?
For a given core count the Intel W and Core i9 offerrings use exactly the same die. "came out after" is purely a marketing launch and production sequence. It is nothing to do with whatever short term tactical move by AMD marketing.
the <codename>-E products were previously all derivatives of the <codename>-EP products. Core i7 9xxxx (extreme) were same die as Xeon E5 1600 for a given core count. So Core i9 and Intel W are two marketing name changes but the strategy is basically the same as previous iterations.
These designs take 3-4 years to roll out start to finish. Intel didn't know exactly what Zen was but there was probably enough leakage of early info ( customer chatter , etc. ) that they knew that AMD was going with smaller, non-max-complexity cores (relative to Intel ) and more core count ( with more affordable cores). [***] If Threadripper didn't arrive because AMD screwed up and not arrived then still probably would have the same core counts for the Intel W (and Core i9 ) products only the prices and/or the volume discounts would be lower.
There are two dies used for the Intel W product line (and Core i9). A "low core count" ( up to 10 cores ) and the "high core count" ( HCC up to 18 cores ). https://www.anandtech.com/show/1155...-core-i9-7900x-i7-7820x-and-i7-7800x-tested/6
Intel is using bigger dies ( which probably cost more ) and the margins on the Core i9 versions are tighter than previous generations of the "extreme" product line. That's why the Core i9 (and Core i7 'extreme' ) products are more widely kneecapped and gimped on I/O. ( chopped down PCI-e lanes, chopped down RAM max, etc.). The stuff that test better at being overclocked gets tagged with "Turbo boost max 3.0" (https://www.anandtech.com/show/1155...-core-i9-7900x-i7-7820x-and-i7-7800x-tested/7) and the unlocked clock and ECC turned off. But it is bascially the same dies coming off the same production line just binned different ( and features flipped off).
the 'HCC' binned and features flipped off to larger extent as probably shares most if not all of the same die as the Intel SP of the same core count class. Stuffed into a smaller package with a lower pin count to the features flipped off. Again not really going to be cost competitive with Threadripper, but then again doesn't have to be.
Looks like both Intel W and Threadripper are going to get a low end process bump. Intel W to 14nm+++ and Threadripper to what is being called 12nm ( but pragmatically a 14nm with some change up s on what is being measured. ). If Intel cuts their prices and AMD can match the move that may turn out OK for Intel. Next major iteration they'll have to figure out something different. But I think that is primarily why they are 'forked' the 1 socket and 2 socket product lines so they can be better optimized for their target markets over longer term.
[***] The smaller cores with higher count is the same tactic being used by the ARM based server folks trying to crack parts of the market that Intel-W would be deployed to. Intel has the Xeon D and Xeon E3 to cover that too but the general strategy of coming after Intel with increased core count can't possibly be a mystery to Intel. The tactic is already being widely used. So AMD following on that same vector would be a reasonable expectation even if Intel didn't have 100% of the Zen design details well in advance.
Last edited:
Does it? First, The PCH chipsets are different between the Intel SP and Intel W. The boot firmware is going to substantive differences. (can cover more complexity but that isn't more cost effective or less troublesome. )
Second, The cooling requirements are in totally different zone. In the previous generations, the "dual" socket processors were generally in the range of about 90-100W while the singles were generally in the range of 130-150W. Generally not 1/2 the TDP but the two together were 25-45% more headroom than the single. the SP processors with the decently reasonable base clock are more closer to the single Intel -W range ( > 100W range). You have significantly bigger sockets so the cooling of socket 1 is far more likely cooling of socket 2 unless you engage significant airflow diversion around the first's blockage zone. ( go look at the depth of the conduit redirect that the HP does for the Z8. http://storage.media.ext.hp.com/HP_Z8_AirFlow_Final_1280x720.webm )
Third, Space wise of the board also if giving 8 DIMM sockets to each processor ( Apple sticking with 4 DIMM for each socket isn't going to draw the "I want a Z800 max box" crowd. ). You can make the daughter board twice as "tall" to bulk out the redesigned Mac Pro's width, but this is going to going to get into the 5U range .... again my guess is that folks trying to 'rack' this are going to moan about that.
Using the space that an extra socket would take up for more than 4 DIMMs slots is better a 'bang for the buck".
The iMac Pro is priced to fit about the iMac BTO options. Crank up all the iMac BTO options to full and add about a $1000 gap and are in the zone of the iMac Pro. IMHO, the primarily thing that Apple is doing with the iMac Pro is pricing it away from the iMac. Pragmatically, it is an extension of the BTO range of the iMac That's why start with 8 cores because the BTO of the iMac is probably going to switch to 6 cores in 2018.
The Mac Pro just has to be "above" the standard configs of the iMac ( not the entire BTO range of the iMac ).
So Intel W and perhaps lower core at the lower entry price point that is higher than top standard configuration of an iMac ( and many of the reasonably priced BTO options ).
However, there is no need for dual sockets because the iMac Pro leaves all of the Intel W performance on the floor.
Intel W with full TDP window is faster than an iMac Pro. There is absolutely zero need to jump to SP processors to get better numbers than an iMac Pro. Even less if stop myopically looking at just the CPU. The GPU (and GPUs plural) can push the performance range past the iMac Pro envelope again with zero leveraging of the SP derivative.
Intel W is quite capable of doing 8 DIMMs all by itself. With 32GB DIMMs that 256GB and when 32GB DIMMs become practically priced and available that would be 512GB. That is a lot. Wouldn't cover "everybody" , but the percentile of real user workloads is extremely high. Again 8 DIMMs Intel -W is envelope that iMac Pro won't cover. ( at Apple prices for Memory, that will probably blow past the iMac Pro. )
The notion that Apple has to gimp the Mac Pro may making it so graciously expensive folks more folks will "have to" buy the iMac Pro doesn't make sense as a strategy. Long term the most likely outcome of that tactic would be the Mac Pro would die on a pricing death spiral. There is a substantial performance envelope flexibility a. Mac Pro could have with simply just another 500-600W of power and some limited internal expansion flexibility
At this point then you're still just one person saying "no, I want my machine" while everyone else wants theirs. You have an equal claim, but there's probably more of the latter than the former. How many Z8 workstations does HP sell versus Z4 and Z6?
While I don't disagree, you are not mentioning how the MacPro will be competitive among equivalent models from Dell & HP. There has been leakage from traditional MacPro users to them, wouldn't Apple have to consider positioning itself to entice these folks ( who weren't for whatever reasons enticed by the iMacPro ) ?
Looking at the Precision lineup ( not shilling for dell ) but the XeonW equipped models are priced at the entry point ( I realize 5000 is not pedantically the entry point, more mid line, but we could then debate if the lesser offerings even qualify as colloquial definitions of 'workstation' class ). Anyways I can't imagine it would be good optics for value seekers.
Other question I wanted to ask, was regarding i9 to W manufacturing. I can understand how there is going to be some selection process to select out 'inconsistencies' and 'rebadge' chips under separate lineups. But it seems odd that the chips selected for i9 would on the one hand have lowered memory ceilings, lower pcie lane count, but higher caches and better overclockability ? Is it such that the manufacture process has a number of specific or favoured failure modes that end up with chips 'gimped' in one certain area, but better performing in others ?
It isn't about being afraid. if the bleeding edge high end of the market represents 0.06% of the potential Mac market then doesn't make any significant financial impact for Apple. Apple is not in the business of selling everything to everybody. They don't have to. Even less so in the Mac market since it is only 8% (or less) of the overall classic "PC" market. So every day 92+% buys something else. That 0.06% jumping off to join that 92% is still going to leave it at 92% when rounded to the nearest whole integer.
Your sample size of one is demonstrative of almost nothing in terms of market size relevancy. Even if could round up a hundred folks ( 2,000 isn't representative of a viable number so substantive R&D differences. Apple's min threshold number is probably closer to 40K than it is to 10K. )
Intel has a workstation targeted CPU called Intel -W . So why would Apple not using Intel's workstation product for their workstation?
The Intel Xeon SP is a data center (workloads for multiple users ) targeted CPU. Apple sells no data center targeted servers. They are not going to either (the XServe was only aimed at a very narrow subset of that and was nuked years ago at his point anyway. ). Apple doesn't sell $65-100K servers and it is not the end of the Mac world.
SP CPUs are not primarily targeted at the workstation market. The vast bulk of HP, Dell, Lenovo and any other vendor with a substantive workstation business is not based on Xeon SP CPUs. They have edges of their line up that is SP based but those are almost always coupled to a line up of servers where they can sell are reasonable number of SP CPUs. The assertion that the SP fanatics have to cover is how is selling SP workstations a viable business when not also selling SP based servers. No other major player is practicing that.