How to take 50% market share in three years.

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Dec 28, 2013
53
0
66
#27
Cloud providers might be a different story
Cloud providers are a different story, They're continuously building out their data centers to the tune of millions of CPUs per year. That's NEW server farms. You really think they won't wholesale switch to AMD for a 3X+ perf/watts/$ advantage for the next two years AND a roadmap that maintains a huge competitive edge over Intel for the following three years?

EPYC 2 will also be by far the most secure server chips in the market.

They'll turn on a dime and tell Intel to take a hike for those kinds of TCO and security gains.
 
Last edited:
Dec 28, 2013
53
0
66
#28
Nvm, we just have a different take on delusion.

I'll let the OP carry on ;)
Don't see anyone contesting the particulars of my reasoning or numbers.

A glib person will pooh pooh long differentiated fact backed reasoned posts without actually addressing a single one of the points raised.
 
Last edited:

epsilon84

Senior member
Aug 29, 2010
931
104
136
#29
Don't see anyone contesting the particulars of my reasoning or numbers, just a lot of glib generalizations.

You know what kind of people pooh pooh long differentiated fact backed reasoned posts with short glib inanities without actually addressing a single one of the points raised?
You base your flawed analysis on the hope that Intel will be using 14nm into 2020. Enough said... Carry on
 
Dec 28, 2013
53
0
66
#30
That being said, I'm sure AMD is gaining ground on Intel in the server space, but it's a VERY long road to a 50% slice of the pie...
As you can see above I did the arithmetic and it turns out the actual figure was a 4.5X advantage.

The path to a 50% slice of the pie is a whole lot shorter if the 10 biggest cloud companies go with AMD for their new build outs.

Lets go super conservative with a 3X perf/watt/$ advantage for two years, which is the absolute minimum Intel will be on 14nm with a monolithic architecture. Add in EPYC 2 will be the most secure server CPUs by a substantial margin.

Which of the 10 biggest cloud companies is going to pass that up that kind of competitive advantage to remain with Intel?

I'll warrant none of them do and they will all switch to AMD based server farms just as fast as they can get their hands on EPYC 2 CPUs. Then EPYC 3 CPUs. Then EPYC 4 CPUs. Intel simply will have nothing in the pipeline that can challenge the EPYC line for at least that long.
 
Dec 28, 2013
53
0
66
#32
Last edited:
Dec 28, 2013
53
0
66
#34
Even Intel said they would try to keep it under 20%, and I think 25% in 2 years is realistic. But not 50%
Brian Krzanich said that and was fired shortly after and at that was whistling past the graveyard with that figure. It's human nature to lowball the he// out of such an admission, double that would be closer to what he was really thinking, He did it on purpose to get fired with the affair just an excuse. Do you really think he he wanted to stay CEO as it became apparent how deep a hole Intel was really in? A hole HE was responsible for?

Due to going with an uncore architecture and using 2 foundries AMD will have sufficient Zen 2 dies incoming to take 25% of the server market by the end of 2019. Zen has blazed the qualification path and shown AMD is a serious contender, Zen 2 will show Intel is hopelessly uncompetitive. With a 3X+ perf/watt/$ advantage over Intel and Zen 2 being far more secure than any Intel chip, there's no way in he// the top cloud providers stick with Intel.

It's not like there's any love or respect for Intel, they all know Intel has been gouging them mercilessly because they could.The super 7 alone account for the lions share of new server farm build outs and they definitely will go all in with AMD.
 
Last edited:
Jun 8, 2003
14,060
93
126
#35
Intel's latest server roadmap shows Intel using 14nm well into 2020. In reality it will be 2H 2021 before that changes. At best. Feeel the burn.

https://wccftech.com/intel-confirms...s-and-hedt-cascade-lake-x-28-core-processors/

Quote:
'Moving on to the Cascade Lake-AP series, the new family will be the first to use the “Advanced Processor” badge. Expected to hit markets in the mid of 2019 and under the ‘Walker Pass’ brand platform, the Cascade Lake Advanced Processor lineup is Intel trying to regain their momentum on the server side.'
"The advanced processor lineup will be something that can put Intel back in the lead with disruptive core count and higher I/O capabilities. Just how would Intel achieve that with a new line of CPUs? The answer is MCM."

"With Cascade Lake-AP, Intel plans to execute their own MCM (Multi-Chip-Module) approach. An industry insider, Ashraf Essa, has previously heard rumors of the Advanced Processor lineup and mentioned that it wasn’t expected until the Ice Lake generation. But now with the EPYC series proving to be crushing Intel in one of their dominant markets, there’s no doubt that they’ll end up using an MCM approach a generation early."

It says by mid 2019 these Intel advanced MCM chips will be released. Sounds like 2, 28 core chips glued together just like Threadripper.
I dont think AMD will be that far ahead by next year.

And one year after this you have Icelake 10nm in mid 2020.

"Later in mid of 2020, Intel would introduce their first 10nm Scalable processor family, the Ice Lake-SP. While it was first expected to launch earlier, the 10nm product woes had resulted in a really bad situation for the entire Intel CPU family which has resulted in several delays and various product cancellations."
 
Last edited:
Dec 28, 2013
53
0
66
#36
But now with the EPYC series proving to be crushing Intel in one of their dominant markets, there’s no doubt that they’ll end up using an MCM approach a generation early."

It says by mid 2019 these Intel advanced MCM chips will be released. Sounds like 2, 28 core chips glued together just like Threadripper.
I dont think AMD will be that far ahead by next year.
You do realize 18 months ago ALL intel architectures in development were monolithic?

It takes years to properly develop a new architecture. That's not what Intel will be doing though, they'll be frankensteining together pieces of monolithic architectures with existing fabrics designed for other purposes on a hastily re-engineered '10nm' node in a desperate attempt to get something out the door they can call '10nm' and 'modular', nevermind it will be hopelessly uncompetitive with the 4th gen EPYC. It will take Intel three years after it's release just to match EPYC 2. By that time AMD will be pushing EPYC 5 on 5nm. That's the magnitude of the hole Intel is in.

2 28 core chips won't be 'glued together just like threadripper'. Zen is a clean sheet architecture designed from the ground up to be modular that will be vastly superior to 2 monolithic chips cobbled together with an fabric not designed for the purpose. A dual 28 core CPU vs a 64 core will have substantially lower clocks and lower performance, be half as efficient, cost twice as much, be far less secure and need a new, and gargantuan, 6000 pin socket.

And you posit "I dont think AMD will be that far ahead by next year" ???
 
Last edited:

Markfw

CPU Moderator, VC&G Moderator, Elite Member
Super Moderator
May 16, 2002
17,087
532
136
#37
Quote:
'Moving on to the Cascade Lake-AP series, the new family will be the first to use the “Advanced Processor” badge. Expected to hit markets in the mid of 2019 and under the ‘Walker Pass’ brand platform, the Cascade Lake Advanced Processor lineup is Intel trying to regain their momentum on the server side.'
"The advanced processor lineup will be something that can put Intel back in the lead with disruptive core count and higher I/O capabilities. Just how would Intel achieve that with a new line of CPUs? The answer is MCM."

"With Cascade Lake-AP, Intel plans to execute their own MCM (Multi-Chip-Module) approach. An industry insider, Ashraf Essa, has previously heard rumors of the Advanced Processor lineup and mentioned that it wasn’t expected until the Ice Lake generation. But now with the EPYC series proving to be crushing Intel in one of their dominant markets, there’s no doubt that they’ll end up using an MCM approach a generation early."

It says by mid 2019 these Intel advanced MCM chips will be released. Sounds like 2, 28 core chips glued together just like Threadripper.
I dont think AMD will be that far ahead by next year.

And one year after this you have Icelake 10nm in mid 2020.

"Later in mid of 2020, Intel would introduce their first 10nm Scalable processor family, the Ice Lake-SP. While it was first expected to launch earlier, the 10nm product woes had resulted in a really bad situation for the entire Intel CPU family which has resulted in several delays and various product cancellations."
You sound like an Intel advertisement, except one comment in your own post "But now with the EPYC series proving to be crushing Intel in one of their dominant markets".

So your really need to believe that Intel is in trouble. Its all about how much they loose and how fast.
 
Jul 24, 2000
104
0
81
#38
50% could be possible if Intel pushes 10nm Xeon back to 2021 and 7nm Epyc turns out to be really epic.
 
Jun 8, 2003
14,060
93
126
#39
You sound like an Intel advertisement, except one comment in your own post "But now with the EPYC series proving to be crushing Intel in one of their dominant markets".

So your really need to believe that Intel is in trouble. Its all about how much they loose and how fast.
It was all quotes from the article the other guy posted.
Not my words.
 
Jun 8, 2003
14,060
93
126
#40
You do realize 18 months ago ALL intel architectures in development were monolithic?

It takes years to properly develop a new architecture. That's not what Intel will be doing though, they'll be frankensteining together pieces of monolithic architectures with existing fabrics designed for other purposes on a hastily re-engineered '10nm' node in a desperate attempt to get something out the door they can call '10nm' and 'modular', nevermind it will be hopelessly uncompetitive with the 4th gen EPYC. It will take Intel three years after it's release just to match EPYC 2. That's the hole Intel is in.

2 28 core chips won't be 'glued together just like threadripper'. Zen is a clean sheet architecture designed from the ground up to be modular that will be vastly superior to 2 monolithic chips cobbled together with an fabric not designed for the purpose. A dual 28 core CPU vs a 64 core will have substantially lower clocks and lower performance, be half as efficient, cost twice as much, be far less secure and need a new, and gargantuan, 6000 pin socket.

And you posit "I dont think AMD will be that far ahead by next year" ???
I read the article you posted, and quoted some parts.
I think they meant 14nm dual die parts next year and then 10nm the year after 2020.
It says they pushed forward the original icelake dual die parts to cascade lake next year mid 2019.

I'm just quoting from the article you posted, I dont have a leg in this, I'm more than likely just grabbing a i9 8 core and overclocking to 5ghz. That's enough for my needs.
 
Feb 23, 2017
446
323
96
#41
This whole discussion seems to be missing a little context.
How large is the annual server farm market? (in terms of number of CPUs sold)
How often, on average, would a server farm completely replace the CPUs they use? (in terms of the lifespan of a CPU)
The reason that I ask is simple; if we're talking about this 50% market share being an annual sales figure, or the actual share of CPUs in use in these server farms.
The former is a less challenging task than the latter, but also the longer term affects of Intel being delayed are less relevant in the latter, hence why the hyperbole of Intel being doomed if AMD reach 50% in the former is just that...a hyperbole.
 

Vattila

Senior member
Oct 22, 2004
363
59
136
#42
if we're talking about this 50% market share being an annual sales figure, or the actual share of CPUs in use in these server farms.
Market share in the context of financial analysis usually means share of sales, in units or revenue. Regarding actual product in use in the market, the expression "installed base" is often used. Except for mindshare, the latter has no direct bearing on the revenue and profits reported by companies, so is largely irrelevant in that context, i.e. it is the sales in the quarter that counts.
 
Last edited:

NTMBK

Diamond Member
Nov 14, 2011
8,171
148
126
#43
In defence of the OP- double digit market share in a couple of years is far from inconceivable. Just look at when Opteron launched:



Within about two and a half years AMD was peaking at 25% market share.

I think 50% in three years is still way too optimistic, but big gains in a short time are certainly possible when AMD has the superior product.
 
Feb 23, 2017
446
323
96
#44
I figured as much, though it clearly highlights the shortcomings of such short-term thinking; without persistently hitting 50% market share, AMD's rise has very little bearing on Intel since one can safely assume that Intel won't have the "inferior" product for a long enough time period to be overly hurt by it.
Are purchasing decisions really made on the basis of short-term superiority? There's a mentality that you won't get fired for buying Intel, acting as a glass ceiling.
 
Jul 19, 2016
135
0
71
#45
Starship was supposed to be 48c which is an increase of 50% over Epyc - yes 50% increase from Gen 1 to Gen 2!
Hello Intel, what did u do the last 8 years??? - sorry can't hear ya while you're drowning in a sea of Zen...

Now in 2018 it looks like it might be a 64c Rome chip - thats a 100% increase in core count on a new node - if this is true AMD has switched to ludacris mode!

64c (probably 4x 16c) would be a rather large die even on the new 7nm node. However given Intels series of mishaps it is likely that AMD could charge a lot for such a monster so that it might be worth it - so this rumor I tend to believe if the process is good enough.

But to reach market share AMD has to meet the demand. With TSMC and later GloFo that shouldn't be such an issues once both are up to speed- that however won't be before 2nd half of 2019 - so 50% market share in 2 years is pretty unlikely in this scenario.

So the to increase the amount of dies and therefore CPUs the author and also AdoredTV suggests that AMD will use chiplets on Zen2 rome with an uncore chip - really??

So after AMD leapfrogged from Faildozer to Zen, AMD switched to ludacris mode for Zen2 and now is supposed to be in insanity mode with a chiplet designed ......that's little too much too digest

I mean data center moves slowly.
And now datacenter customers who just got used to the idea to ditch Intel and consider Epyc should swallow to adopt a whole new CPU MCM chiplet design - just think if how long Epyc validation took !

Sorry but 64c + chiplet + data centers buying this right from the start is just too much leapfrogging for the industry ......
 

wahdangun

Senior member
Feb 3, 2011
993
2
106
#46
The first thing is AMD need more Fab capacity, even if they use TMSC an GF simultaneously, it won't be sufficient.
 

Tarkin77

Junior Member
Mar 10, 2018
5
0
41
#47
In defence of the OP- double digit market share in a couple of years is far from inconceivable. Just look at when Opteron launched:



Within about two and a half years AMD was peaking at 25% market share.

I think 50% in three years is still way too optimistic, but big gains in a short time are certainly possible when AMD has the superior product.
I wonder what would have happened with Opteron market share, if Intel wouldn't have released their Core Microarchitecture Xeons in 2006 ... i assume AMD would have been in the forties by 2007
 
Last edited:

CatMerc

Golden Member
Jul 16, 2016
1,111
36
106
#48
Ignoring everything else (and there's a LOT I need to ignore here), have you considered manufacturing capacity?

AMD physically can't produce enough volume to hit 50% server market share lmao
Intel is called Chipzilla for a reason, and TSMC would need to ramp their operations HUGELY to be able to produce 50% of that while also supplying everyone else.
 
Dec 28, 2013
53
0
66
#49
It says by mid 2019 these Intel advanced MCM chips will be released. Sounds like 2, 28 core chips glued together just like Threadripper.
I dont think AMD will be that far ahead by next year.

And one year after this you have Ice Lake 10nm in mid 2020.

The actual roadmap shows Intel far behind AMD through 2023.
 
Last edited:

NTMBK

Diamond Member
Nov 14, 2011
8,171
148
126
#50
Intel's Cooper Lake, in planning throughout 2020, so in development thru 2021 and actually shipping in 2022, presumably on Intel's 10nm 2.0 (12nm), IF it's ready. Still not a true MCM design. Best case scenario.
I think you've misunderstood the meaning of those categories! Those refer to the current state of the projects at the time that the roadmap was published, hence why everything currently out is labelled "shipping". The date that items appear on the roadmap is the target date for launch, and the color is an indication of how far along the project currently is.

What it does tell us is that Cooper Lake-SP is not as far along in development as Ice Lake-SP (presumably because Ice Lake has been sat on a shelf waiting for 10nm to be fixed). But if everything goes according to plan it should be shipped right at the end of 2019.
 
Last edited:


ASK THE COMMUNITY

TRENDING THREADS