Does anyone know where the nCMP will be made? My guess is the Austin TX site that made the trashcan mac will be making them. But do we know for sure?
That facility may not be all that appropriate. The volume here is probably going to be even smaller than the Mac Pro 2013. The initial demand bubble will probably be a bit different. ( might be a temporary surge that is higher but the steady state run rate is probably lower. )
That specific Austin site may not be the right size. "thousands of cheap Chinese laborers" is certainly not the right size either.
If they are planning to use the giant pile of aluminum cuttings to feed into MacBook Air retina cases then chunks of it will be close to that case building location.
It also wouldn't be surprising to end up in Cork IR to satisfy some tax shelter hocus pocus that Apple wanted to do if they wanted to bring it more "in house".
All Apple has to do is use 2-3.5 months to build an inventory for the intial demand bubble and then build at a relatively low run rate. They can do that with a modest sized facility in a variety of places. there is no "it will only fit here" constraint on this.
Could that be adding to the expense of this model?
No. It really didn't relatively do all that much to the Mac Pro 2013. This 100% increase in base revenue ... even less.
[doublepost=1560114744][/doublepost]
I think although Apple will be getting a good deal on the CPU's they won't pass that on to us and I think the deal is across all the Intel CPU's that Apple uses. I don't think they get 50% off on everything but it probably equates to that level of discount once you normalise all their discounts across the full range of processors, chipsets and thunderbolt chips that Apple is buying.
Doubtful apple is getting 50% off because they use "Just in Time" manufacturing contracts. They set an upper limit of the volume they want to buy at some price but then also monthly/weekly sends notices in of how much to sell. The latter behavior is that of a spot market buyer. Spot market buyers don't get 50% discounts. Apple probably has some "minimal" buy boundaries in there to get some substantive discounts but Intel would have to be quite foolish to give them 50% discounts no matter what. That whole "cut off orders quickly" thing is a shift of the risk back to the suppliers. If Intel had some hyper shorter of parts they could aways diver there is lower risk but
It's not uncommon for Intel to give rebates or discounts to OEM's that don't use AMD parts. They famously gave Dell billions of dollars in rebates to keep them an Intel exclusive shop.
And Intel has been slapped around for anticompetive contracts in the past too. Intel also gives discounts to vendors for slapping the Intel Inside stick on their systems and Apple 100% passed on that too. Apple isn't doing absolutely
everything possible to maximize the Intel discounts they get. That is one reason why they have some leverage to get minor tweaks to the processors ( like the ones in the iMac Pro), because they actually probably are closer to paying "full freight" on the processors and just simply add that to the basic cost of the systems.
As for the T2, it does enable hardware acceleration of H.265 video encoding. But to what degree I'm not sure and usually software only encoding can give better results due to the constant quality enhancements added to the encoder (x265 in this case) which is why people choose not to use the hardware based H.265 encoders present in NVIDIA GPU's - Quality vs Speed. The AVX2 and AVX-512 instructions are perfect for software as they're just ways to do specific math operations very quickly and simultaneously.
There is about zero indication I've seen that this encoding expands past the
Apple camera the T2 is hooked to in most systems. In mac systems with a built-in "FaceTime" camera, that subsystem is hooked directly to the T2 (and the system 'sees" the camera through the T2). This just replicating the same general set up in a iPhone or iPad where the Apple SoC handles the camera data encoding. It also makes for a cheaper camera subsystem because all it has to do is 'burp' raw sensor data off of the subsystem and off to the Apple SoC. If the data link off the sensor is fast enough they don't even need a local buffer to store the data. ( Can use the SoC's RAM as the buffer. )
For the Mac Mini and Mac Pro that probably does a whole lot of nothing because those systems do
not have a "FaceTime" camera. So yes the T2 is capable but if you haven't attached a camera that capability isn't being used.
For the nominal Mac set up the T2 is hooked to the microphones , speaker , fans , camera , TouchID, SMC , etc. all in addition to the SSD duties it has. If apple every moves FaceID cameras to Macs the T-series SoC will process that also upstream of the rest of the system. Same thing with moving on device basic Siri processing. That too can be securely moved to the T series processor. Basic voice specific characteristics never make it off the T2 ( the Mac side is feed already recognized basic parts of speech and not the raw microphone capture). If next T series gets more Tensor cores then more "smarts' can be processed almost in an autonomic fashion just on the T series. It will be limited in scope because only mostly fixed Apple ML models will go into the relatively fixed T2 firmware. More 'smarts" can be layered on top by the host processor.
It doesn't make alot of sense to pump 3rd party camera data to the T2 (or T series ). Most macs have an Intel integrated GPU with fixed function H.265 encoder. That encoder is hooked directly to RAM. Copying it out of RAM to the T2 and back again is going to take time. For discrete GPUs the link to the embedded H.265 encoder is on a x8-x16 PCI-e v3 line. The T2 is on a x4 one. Again slower round trip time.
The T2 makes lots more sense when it is the first in line to get the camera data; not the last.
It's all very interesting. Also that FPGA based accelerator card they released, that is very interesting.
If priced reasonably perhaps. If priced so high that only very high end speiclists need it then not so much.
With it not being an ASIC they can add deep functionality over time. Heck it could even become a H.265 encoder if they wished it and they could keep upgrading its quality over time with software updates as it's not fixed function like an ASIC based card would be.
ProRes will change over time. It may mean just tracking that. Apple running Compressor data flow through it is possible. Maybe some Apple Motion. An even bigger Maybe some LogicX plug-ins.
I will be surprised though if the usages expands past Apple specific FPGA 'programs'/configs. Like a general API to the FPGA for 3rd party users. Apple is most likely going to be doing the 'programming" here and their stuff does change over time ( so don't need new card each time.). But this will be more like "firmware" upgrades that you get periodically over time. Not an "AppStore" for the Afterburner kind of thing.
