I don't think stabilized Mac Pro production is going to be related to adopting Haswell for the Mac Pro. Sticking with Ivy Bridge doesn't get them closer to stabilizing production.
Neither does moving to Haswell. It has to do with fixing the errors they have now that don't have to do with minute parts characteristics and have far more to do with
correctly forecasting demand, sizing the factories and order flow to that accurate forecast. What Apple is exhibiting right now is clearly not a highly accurate forecast. Unless Apple has planned that the status quo ordering time for a Mac Pro be 4 weeks, what they are in currently is a screwed up situation.
Exit out of the screwed up situation and find what the far more accurate forecast is for this product and then correctly size production for the next generation. That shouldn't take months to figure out but it also going to inhibit a shorter than normal cycle also. [ The jacked up iMac intro didn't get an abnormally fast refresh cycle from 2012->2013. ]
If the line isn't able to stick a Haswell processor in a socket instead of a Ivy Bridge processor, Apple probably has really big problems. They're just parts, not production methods.
If you haven't ordered enough parts and don't have enough production stations to insert the part you STILL have a problem. Socketed or soldered CPU doesn't particularly make a lick of difference.
Haswell is a major revision, which I think means Apple will have to be quicker in adopting it. They've already burned a lot of pro goodwill, and if they take 6-8 months to adopt Haswell, they might lose even more of the pro market.
I didn't mean to suggest it would be 8 month time frame. If Apple merely goes 12.5 months on the revision it will slide into 2015. The overwhelming vast majority of Mac Pro 2013 customers did not receive a device in 2013.
Apple can get back goodwill by just being consistent. Regular, largely predictable, 11-13 months cycles would be just fine. I don't think this "launch with large scarcity" is building large amounts of goodwill. A shortage at launch for weeks has litle impact, but if they are going to consistently roll out with 4-5 month shortages that is going to cause goodwill problems that "E5 v3" isn't magically going to correct.
Especially if both Maxwell and 20nm Hawaii hit in that same time period.
Are there going to be 20nm Hawaii in large enough roll out inventories?
Nvidia is rolling out Maxwell at 28nm. AMD can't produce enough on a mature 28nm process to satisfy Hawaii demand.
http://www.anandtech.com/show/7758/radeon-r9-290x-retail-prices-hit-900
Apple, Qualcomm, and every other low power budget ARM CPU+GPU vendor is going to soak up as much 20nm capacity as they can get in the Q2-Q3 timeframe.
I think Nvidia and AMD will rollout, there is just lots of doubts they are going early and early enough that the trailing edge OS X graphics driver updates are going to roll out early also.
Apple likes smaller teams, but Apple's Mac Pro team isn't the same people sitting on the Mac Pro production line putting these things together.
Never said they were. But the production line folks aren't fixing firmware problems with the current one. They weren't diagnosing/fixing bugs/defects that the beta testers turned up. They aren't fixing the bugs/defects that the large swarm of users are turning up now that the Mac Pro isn't coddled in some top secret research lab.
If some of the same folks are fixing current bugs/defects s are tasked with working on future stuff, then the latter tends to slide. Even if it doesn't slide it extremely rarely speeds up.
The smaller teams extend into the other parts of Apple. The driver updates that Mac Pro 2012 need for newer GPUs slid into early 2013 ( 10.8.3? ). IMHO, I don't see much utility at all at moving the release of a "pro machine" close to the transition point for OS X for its yearly update. The first release out is never highly critical production ready. It typically takes Apple almost 6-8 months to roll out a 10.x.3 version that fixes a significant number of blips/glitches/defects and have solid OS X.
While Apple can ship out without all the Mac Pro bugs fixes on a general roll out the impact is spread out over more, much higher volume, Macs. If want higher quality software for Mac Pro, the prudent move would be to move it away from frantic activity of the rest of the Mac ecosystem.
I'd bet Apple has Haswell prototype Mac Pros in the labs right now, along with possibly 20nm Hawaii GPUs. I don't think there are any samples of Maxwell on the desktop, but if Apple were still considering Nvidia they might have them.
They can have them the missing element is how Apple can move
faster than the other teams that don't have to cover current and future issues. Or t teams which have picked up the samples months earlier than Apple.
But in 2008, 2009, and 2010 they shipped on or fairly close the Apple's schedule.
2010 Apple shipped in August when the Intel CPUs launched in late Q1 '10. IHMO in part because the 3600 line up was woefully incomplete so Apple had to settle for 3500 CPUs for entry-mid standard configs.
Launching a new Mac Pro on the same day Intel launches the workstation class updates isn't absolutely necessary. Several weeks or even a month gap isn't a big deal strategically.
- Haswell-EP is advertised by Intel as officially supporting Thunderbolt 2. Given that Sandy Bridge-EP doesn't officially support Thunderbolt 2,
Where is either of this documented? I heard that Xeon E5 can't work with Thunderbolt several times before Mac Pro design was released. It didn't make sense then. It made even less sense when Apple introduced the new design and folks like HP started showing workstation TB card prototypes.
Necessary connection to GPIO , PCIe yes. but the Xeon package isn't provisioning GPIO and the PCIe is standard (and in Mac Pro design goes through an intermediate switch anyway).
TB v2 was long term targeted by Intel for 2014. So has been E5 v3. It isn't going to be uncommon that reference platforms that Intel has probably put together have both features onboard as they are targeted for the same timeframe.
- Haswell is not electrically compatible, but it's physically compatible with the same socket already on the Mac Pros, so that means the amount of changes Apple has to do to the board is minimal.
The chipset is changing as well as the socket. USB 3 is in the chipset. Apple keep the discrete (and pay for USB 3 for nothing, but keep open bandwith for the SSD ) or dump it (and route up to I/O connector board)?
It isn't building a fusion reactor complicated but there is work to do, validate, and test.
- Haswell runs much cooler, which is very much something Apple would like to enhance the Mac Pro.
- Haswell also gains a higher core count, which is going to put Apple at a strategic disadvantage compared to other competitors.
Pick one or the other.... Haswell doesn't do both. The lower single cores enabled higher core count at the same general TDP levels.
This is the same spin that was throw out for waiting for E5 v2 versus going with E5 v1. It is going to be way cooler. It wasn't and probably is not on this generation either.
The fine grained controls on turbo and the ability to go up/down dynamically will be better. But the top end tolerance that the Mac Pro will have to handle will largely be the same. the 8 core part will be way cheaper. (It probably won't be a flipped 10 core 2600 part ). Likewise 6 core probably will be cheaper ( E5 1620 v3 may start at 6 cores if Intel rains in the clock speed from over 4GHz base. )
- If 20nm Hawaii is ready, that's definitely pin compatible with what Apple has now. It's basically a "for free" win. The GPUs are more powerful, and put out less heat.
Hawaii would have to drop heat just to get back to normal Tahiti range.
The die shrink variant probably will get another name but as noted before new hardware without drivers and software doesn't particularly buy much but bragging rights and new benchmark drag racing demos.
I really don't see a reason why they would hold back on an upgrade at this point. Strategically, they aren't in the middle of a redesign like they were in 2011 and 2012, and the Mac Pro's place is more secure at this point.
This redesign isn't take 2010-2013 to do. Part of the reason the Mac Pro is so late into 2013 is because Apple didn't start earlier. Too many pieces out there:
i. rumors Apple was not requesting access to E5 v1 parts in late 2011- early 2012. [ no rumbings that Apple has E5 v3 is not a confirming sign.]
ii. the EU product drought. Yes that was weaved into the grand strategic plan; not.
iii. TB v2 was long term tracked by Intel to be targeted at 2014 for volume production.
Rolling out the 2012 model with ancient GPUs didn't dramatically enhance Apple's standing. The socket , I/O chipset , and TB v1 were all available in 2012 for this design. There is no radical CPU TDP difference either.
That late start has a ripple effect that isn't magically disappear in one iteration.
I think with the 2014 Mac Pro, you're looking at a machine that looks the same, has the same capabilities (single CPU, dual GPU) but with Haswell, DDR4, and the current generation of GPUs. That's not an extreme upgrade from an engineering perspective,....
Engineering was not a huge bottleneck to getting the 2013's design out the door either. Apple's allocation of resources, the OCD aspects of their design process , protracted graphic driver development cycles, etc. all pushed it later. Few , if any, of those factors have shown huge evidence of being dramatically changed.
