A few things:
- Apple is buying in bulk so they're not paying $200.
Apple is one of the largest NAND Flash buyers in the world ... are their SSD and Phone Memory prices to end users lower than average? Nope.
Apple buys "just in time" and in bulk. For the Mac Pro space "just in time" is gong to dominate. Apple's upgrade prices on Intel CPUs is about 30% mark-up over list price for a relatively pedestrian tray of processors price points listed on ark.intel.com.
- This is an upgrade cost we're talking about, so you have to remove the cost of the baseline card (unless this is both the baseline and max card, which would be a little strange.) Apple's upgrade cost would probably only be around $350, so $500 would be a reasonable upgrade price.
You are drifting in to "top of the line card" talk. That isn't where this started. Clockspeed, memory size, and GGDR5-vs-GDDR5X allows more differentiation than previously with previous generation GPUs.
How would Apple get a broad "Good , Better , Best" out of a 1070's foundation?
- I'm not really talking about Apple's pricing here, I'm talking about the end price to the consumer. The Mac Pro sells to a market that doesn't really care as much about price.
Again... why is the D300 kneecapped if nobody buying Mac Pro's cares about price? Why is the DIMMs slot empty on the entry model? The D500 ( 3GB and tweaked core count ) is a tweaked down W8000. Why weren't the D300/D500/D700 priced at the same level of the W7000, W8000, W9000 if no one cared about price at all? They do. Apple customers are a bit more elastic on pricing but they are far from just don't care.
If we were looking at a more efficient card the story would be different. But if they're going to charge 1070 level prices they need to deliver 1070 level performance.
Not really. One, like the Dx00 series versus the mainstream AMD Wx000 line up, this probably will not be coached as what is the off-the-shelf, Fry's/Newegg/etc discount special cards. Two, if there is not 1070 option then it isn't a directly substitutable option. The Mac Pro is primarily going to be sold again the previous Mac Pros.
92+% of the PC+workstation market wasn't buying Macs before and won't after any new Mac Pro is introduced. That isn't who Apple is primarily trying to sell to. The ultimate, price, power, and/or bulk is no object market. Apple has had an even smaller percentage of that. New Mac Pro isn't going to make a different.
If Polaris was more efficient Apple could at least overclock to try to gain ground back.
For the Mac Pro at best the objective would be to original nominal clock. More like going to need to clock down. IMHO looking at the wrong side of power curve.
Again, I think Vega is the only way out here on the AMD side. Polaris has really bad performance per watt compared to Nvidia which is critical to the Mac Pro.
Comparing to Nvidia is a bit apples to oranges. Polaris is better than where AMD was. That is about all really should expect from them. IHMO, Vega is highly unlikely to be some kind of magic bullet. It seems likely that the AMD would have used mostly the same talent level , resources, and design tools to move Polaris forward from the previous generation as Vega will move from the Fury designs.
Vega is coming so close to Polaris is probably quite unlikely that any major design/skills revolutionary changes got folded back into Vega from Polaris. Vega is more likely an increment of Fury ( HBM 2 for more bandwidth and design target for around 14nm. ). Performance/Watt may increase by cranking Performance up much more than are lowering Wattage down. For the Mac Pro unless the P/W also has a "lower Watt" aspect it isn't going to buy much.
I expect Vega to compete for benchmark throughput more competitively with the top end Nvidia offerings, but not on lower Wattage. Once again it will probably be Perf/$ that is primary target.
Similar to how Nvidia is dropping multiple implementations per named generation, I think AMD is going to have multiple implementation, but be split over different named generations. More rebadging and design/process tweaks in the in between years. Kind of like a tick/tock only not completely across the board: Polaris (lower-mid) ... Vega ( upper mid - high end) ... bump (lower-mid) .... bump ( upper-mid high) .... bump (upper-mid high). Each one of those dropping about every year ( after roll out these first two in a year. ) [ Even Intel hasn't been doing tick/tock across the whole line up. Xeon E5 1600 is just now about to get to v4 Broadwell when the mainstream is going to transition to begin their version "7" in the late Fall. ]
Apple may be able to at one point shoehorn one, optimized ( binned and tweaked) version of Vega into a Mac Pro BTO model, but I doubt they will be able to base the main core of the whole line up on Vega any time this year.
At the very least, you could hope that Polaris is just an early sample and that Vega will be much better. But if this is Apple's top end option, they're in trouble.
As I said before, if Apple is going to sit and disappear down a rabbit hole for 3 years at a time their top end option is toast no matter who they pick. AMD vs Nvidia is just completely missing the issue. Apple doesn't need to create an open market for GPU cards, but they do need to do something on a regular basis. Mac Pro design can't be so highly customized that they can't roll out an interim GPU card (along with a GPU upgrade service at cost.... not cheap ) on a period that is shorter than 3 years.
If you can get to another card in 1-2 years as technology improve it is not so much of a underwear-in-a-twist issue at the top end. Apple can be behind to absolute bleeding edge if they are:
1. offering better stability / usability / predictability.
2. are progressing ( just time shifted 6-10 months back ).
For most part OS X ( and its drivers ) have been offering #1 and #2 pretty much was the trend with the 2007-2010 Mac Pros. Disappear down the rabbit hole and flakier versions of OS X , and crufty GPGPU infrastructure/tools isn't going to cut it.
