Enjoy the hype? The problem with
Demerjian is that sometimes he gets on the 'hater' hype train and goes too far.
1. Dropping margins
is an option for Intel. Stockholders won't like it, but if your manufacturing is having major problems then it is pure fantasy to say you can keep sky high margins. If Intel datacenter went from 60-70% margins down to 45-50% margins it wouldn't be the end of the world.
Maybe there do have some bozos running the exec tier but it is just a hit stock price. The stock is up about 50% since the start of the year. If most of that is bloat then losing it won't kill the company or the server business.
The strategy of "we have a weak product so let's increase the prices 20% " would be more than odd. It wouldn't be surprising if that increase disappeared in discounts to folks who buy in volume. It could very well be a 'enterprise" sale tactic of crank up the prices and hand out the discounts as though doing someone a favor. The folks who buy in "onesy , twosey" get hit, but that isn't their core customer base.
Sometimes have to pay for your mistakes. Intel has $12B in cash. The server group shouldn't have to make money for the rest of the company. Raising prices is proactively giving away the business. ( if sales were slow raising them is only going to make sales slower. )
2. His QPI bits rates are off. Also completely ignores the internal bus that Intel's larger dies use. When Intel goes multiple die in a package they may or may not be using QPI as the interconnect. Not sure why would use a inter-package network for intra-package traffic if didn't have to. ( sure AMD's supposedly does both equally but that isn't the only path to a solution. Intel has a ring bus that has worked inside a package before. )
3. I'm not sure he is correct on CooperLake being same limitations as Cascade/Skylake . The leaked chart has Cooper barely inside of 2019 and mostly deploying in 2020.
https://www.anandtech.com/show/1311...ap-leaks-cooper-lakesp-ice-lakesp-due-in-2020
As opposed to his comment of
"... Cooper Lake is due out in early 2019, at the moment, a ... "
If back earlier in his rant he was commenting on how Cascade Lake appearing in Q1 2019 was going to annoy the common folks because the select hypercalers go it early. ( which happened for Broadwell , and its precessor I think also. )
There is a fairly decent chance that since CooperLake is larger due to the "3-die water cooled .." package ( for the AP. The non AP could easily be two ) that CooperLake could be a socket shift to what Ice Lake is going to use. Since it is gap filler, it could fill the same gap with the new socket. All Intel would have to do is 'back port' the additional memory controller onto the die without the x86 core upgrades and they'd have the memcontroller bump. [ something somewhat similar happens when Intel pulled back Ice Lake PCH controllers and mated them to current desktps CPU packages. It is more work but it would be tractable if started a while back. ]
The Cooper Lake AP and Ice Lake AP processors are taking the place of Xeon Phi. That's dead-ended. They aren't indicative of the mainstream server offerings Intel has. Phi is being folded back into the server track foundation. ( the mesh network, HBM memory , extreme high package core counts , etc. ). That actually should save Intel some money.
4. ".. OK with the Meltdown and Spectre patches it will slow down a bit, ..."
Errr, Cascade lake should have the fixes for those. So patches won't be needed (or at least 'large' patches). They won't slow down, they'd be a speed increase over he previous offerings. For secure virtualization it isn't solely about speed ( cloud customers don't want their data ripped off either as well as their computations done. ).
Could Intel charge $1000 more for that? No. Will more secure CPUs sell better in a hyperscalle datacenter? Probably yes. ( especially if it is a socket swap and firmware bump ).
5. . don't be surprised when a substantive chunk of Intel DataCeneter meeting is about how Optane can scale "primary" memory many GB bigger than AMD. For folks that are doing memcache and low latency stuff, AMD cores can't win it by themselves. Once blow past the RAM layer and have to hit storage they'll take a hit.
Intel definitely has problems but the doom being spun here is bit overblown.
Finally in terms of Macs.... Scale out is probably not going to be an issue. AMD is still for immediate future trailing on single core. ( which is why haven't seen huge exodus on laptop/dekstop space. ). 2019-2020 will be rough for Intel, but it isn't complete doom for Intel.
Apple should seriously be considering AMD as an option. Make far more sense than the "move the whole Mac line up to ARM" fantasy game. Some of the problems Intel is having is bonehead moves. Tech glitches can be fixed, but if dealing with bozos .... things are likely to get worse over time.
[doublepost=1533668737][/doublepost]
They're not going to release a SKU that requires dual GPUs. They learned their lesson from that and in their interviews they make it clear that some people didn't need them and that was part of the problem of the design.
A SKU is not a product. It is just
one configuration of the product. SKU -- stock keeping unit. It is an inventory index.
It is a big leap between all of the "good , better, best" standard configurations will be dual GPU than to say none of them will.
Apple could do a "good" that was BTO through all the CPU options with single GPU and a "best" that started with a CPU + dual GPU floor and went up. On Apple's current site there are just two standard configs ( jump in and start to configure from there. ). In earlier iteration a generation or so back where there was "plain" Mac Pros and a "server" mac Mac Pro.
As Apple said. Some folks found dual GPUs are very useful and some did not. Neither group necessarily has to 100% dominate the standard configs. There would probably at least as many single standard configs as dual, but dual doesn't have to be zero. Especially, if the "some" favoring dual is up in the "20+ % " range.
The customers who are trying to buy the minimal "bare bones" system from Apple don't need a lot of SKU. They are going to dump 3rd party parts into the box so Apple doesn't have to keep those configs in inventory ( hence no SKU). The folks are more inclined to buy most/all of the parts from Apple are the ones that need Apple SKUs for because those are what Apple is selling directly.
[doublepost=1533669547][/doublepost]
well there is way one that they can do dual gpo and that is an on board basic low end chip just to give TB video and have open X16 slots for the main video card with full out ports not tied to TB.
That isn't the primary point of a dual card configuration. In current Mac Pro one card is the 'Display" GPU. the second is primarily just a compute GPU. if Apple is selling that second GPU it would probably be as a compute. ( as you are suggesting the second one is "bigger").
One, Apple's card could fit in both slots. That means the dual configuration would allow them to sell more video cards (and get to higher economies of scale). All they need is a card that would fit in the open PCI-e slot with the display not portion not connected. ( it would function solely as a "Compute" GPU withouth being a seperate item to inventory).
Second, if seems doubtful that Apple is going to get into the business of selling general PC market GPUs by themselves. They have no interest in battling it out with Amazon , Newegg, and Fry's. They may sell a whole eGPU system at some point that uses a card as part of some total package. But loosey, gooesy single cards that aren't primarily "Mac" GPUs probably won't happen.
Pragmatically, more than a few folks who are going to want a second GPU for video out are likely going to be folks who want Nvidia solutions. If there is a core fued between Apple and Nvidia at the core of why they loose every design bake off, then Apple isn't likely going to want to sell their non Mac cards directly. it would be goofy for Apple to completely block that solution, but they don't have to participate from the Apple store inventory.
