Intel boss: Expect chip shortages into mid-2019, stumbling server processor sales this year Intel saw some of the shine wiped off its 2018 fiscal year as the chip giant closed out the year with a disappointing fourth quarter and warnings of lean times to come. Chipzilla said for the final quarter [PDF] of last year, ending December 29: Revenues of $18.7bn were up 9 per cent from the year-ago quarter, but short of …
The FANGs already get custom silicon. Google has for years been working with other silicon, including its own "TPU"s, Amazon has just launched its own ARM stuff.
I think the customers here will be the usual data centre crowd, plus companies ramping up their own virtualisation efforts where it is still very much an x86-64 world. The move to the "cloud" has been extensive over the last couple of years, but it will come to an end at some point, as Intel alludes to (presumably from fewer requests and orders).
We will, hopefully, see a competition server market at one point but the barriers to switching architectures will remain high for a couple of years.
It's not an artificial monopoly if you screwed up and can't supply new products that you need to remain competitive. And it's not short sighted if you have been trying to deliver next gen for ~5 years and won't deliver it in a way that affects the market in 2019... I would go as far to say, I expect volume delivery on a 7nm process (no...I'm not suggesting Intel will re brand 10nm) in 2020 before they ever get volume delivery of 10nm. Intel has effectively wasted US$30b on two 10nm fabs (US$10B/fab is in the right ballpark for initial setup costs plus an additional $10B on remedial work to try and get yields upto a level that maybe profitable) what will likely never produce viable product lines and will be scraped... What we will see out of the 10nm is more excuses (PCIe 4 chipset issues being the current reason delaying 10nm)
There is a monopoly aspect to this in that Intel has a single competitor that is in catch up mode at the moment.
The question is whether AMD can execute in 2019 - their samples show a lot of promise, but AMD have shown promise many times in the past, but not always executed in a way that allows them to benefit. AMD likely have 18 months before Intel are back on a level pegging (i.e. both players on 7nm-like processes) and it will be interesting to see how the products compare. Historically Intel have been conservative in their designs, knowing they can make up for it in manufacturing while AMD have had more innovative technical designs that have been let down by manufacturing issues...
You write as if you were completely oblivious of the current situation. Intel screwed up the 10 nm process, but it didn't screw up its position. AMD just made a small dent into Intel sales. Intel was close to a monopoly in its segment and that situation changed only slightly, at least for now.
I'm not completely oblivious to the situation.
AMD hasn't had 5 years of competitive products during Intel's 10nm struggles, and the current Epyc CPU's only hint at what is possible. i.e. at 12nm they are competitive with early memory access issues resolved, at 7nm they appear to be unbeatable until Intel addresses it's 10nm issues but that is likely to be Q2/Q3 2019.
Because of the delays in 10nm, there is a lot of existing demand from 2018 for features that were promised but not delivered and can no longer wait for Intels next generation, but they aren't quite prepared to jump to AMD (i.e. adding more capacity to existing VM farms and hitting VMotion/live migration issues with differing processor architectures). However, the cloud providers are looking very closely at AMD.
Nope... Intel are selling 100% of what Intel can produce or are contracted to provide. Or very close to it.
Traditionally, Intel have run CPU's on their latest process node/newest fabs (N process node), chipsets on the N-1 process node and embedded systems on N-2 with N-3 being refitted for the N+1 process node or mothballed as it has reached end of operational life.
With N=10nm not producing much and parts on N-2 still moving to N-1, and N-3 parts moving to N-2 to allow N-3 end of operational life, this is causing a noticeable capacity constraint at 14nm.
Intel still have to try and produce a "faster" CPU product and to do that with the same process node, they need to tweak the design (done, done and done again..) and offer new chipset features to make the platform faster instead (hence chipsets moving to 14nm to join the CPU's).
If your 14nm manufacturing processes are running close to capacity/theoretical yield limits, demand for top binned products exceeds supply and you are trying to cover the costs of 10nm fabs that aren't producing anything, do you produce slower chips or just produce the top binned items and let existing inventory cover the remainder of the market?
The bigger the die, the great the chance of defects. AMD's strategy of many smaller dies has put pressure on Intel. They are now in a core count race, and losing. A 28 core 3.0 GHz monolithic die will have better performance than an AMD 28 core 3.0 GHz chiplet. But the AMD chiplet means they can sell a processor for significantly less and still make a handsome profit because of yields. Intel's monolithic chip will have ultra low yields; AMD's will have very high yields. Intel will have to go chiplet at some time.
A 28 core 3.0 GHz monolithic die will have better performance than an AMD 28 core 3.0 GHz chiplet
. . .
Intel 28 Core Xeon Platinum 8180 Server/Workstation CPU/Processor
£9,508.49 each.
Item currently awaiting an ETA == vapourware
AMD EPYC 7601 32-Core, 64 thread Processor
£3,999.00 each
Released June 2017 == Buy it now. (literally; it's available on Amazon)
So Intel is hoping that their not quite released vapourware is going to be competitive with AMD's current 18 month old hardware with 4 cores disabled? Yep, sounds about right.
As does the fact that despite having superior and cheaper hardware AMD processors are almost impossible to buy through normal sales channels, which doubtlessly has nothing to do with Intel.
"Intel 28 Core Xeon Platinum 8180 Server/Workstation CPU/Processor"
Definitely not vaporware - we just setup a 4 node ESXi cluster with 2 x 8180's in each node. The customers specs, not mine.
You buying retail or going through an existing vendor? Cisco certainly had them for sale in December (when the order was processed)
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
