Intel admits they're likely going to lose server market share to AMDs Epyc through 2018

[Master Disaster]

23 minutes ago, Deus Voltage said:

I am noticing a very strange trend as I'm searching for direct quotes from Brian Krzanich. News publications, tech or financial (or both), all seem to be quoting Romit Shah as though his words are gospel, but none of them are willing (or can) provide direct quotes from Intel CEO.

 

I'm inclined to, at the risk of seeming conspiratorial, believe that such a meeting was conducted behind closed doors, and much more was said that "could" not be explicitly divulged to the media. 

 

There is more to this story, we're being fed bread crumbs.

And the problem with news by proxy is you never know how much is genuine and how much is embellished.

 

I think as a precaution I'm going to add a disclaimer in the OP.

 

55 minutes ago, leadeater said:

I think he is probably talking about new product purchases rather than market share install base. Not everyone is going to be replacing servers and some people will buy extra and the old will still be running. 15%-20% is only realistic for new sales, can't see total deployed systems being that high there is just way too much out there for that to happen and I don't think AMD could even make and ship that many CPUs and server OEMs put them in products and sell that many.

As per usual, I'd imagine you're 100% correct. Imagining 20% of installed servers magically changing to AMD by March 31st 2019 is silly. Still even 20% of total sales would be a win for AMD and a win for competition.

 

AMD didn't just light a fire under Intel's ass, the purchased one of musk's flamethrowers, set it to full power and shoved it inside the anus.

[mr moose]

3 minutes ago, Master Disaster said:

AMD didn't just light a fire under Intel's ass, the purchased one of musk's flamethrowers, set it to full power and shoved it inside the anus.

Whilst I certainly agree that AMD is now a fire under Intel that is more than likely causing them some concern or desire to restructure their server lineup/business, I'm not sure I would go as far to say the flame thrower is fully inserted.

[cj09beira]

40 minutes ago, mr moose said:

Whilst I certainly agree that AMD is now a fire under Intel that is more than likely causing them some concern or desire to restructure their server lineup/business, I'm not sure I would go as far to say the flame thrower is fully inserted.

when amd starts selling 7nm 64 core cpus that will ust drop into already working servers with a bios upgrade, thats when the penetration will peak

[Sauron]

Well duh, when your competitor introduces a compelling product a market share hit is to be expected. Note that this doesn't necessarily mean Intel will be losing customers or selling less product, it just means AMD is going to sell more product than before.

49 minutes ago, mr moose said:

News stories of Intel losing market share have been doing the rounds since December last year.   It sorta hasn't happened yet so whatever they discussed is not something new or largely significant to the market.  It may just be the case that the story is only bread crumbs and the media are just looking to fatten it up a bit. 

 

EDIT:  or this might just be another case of the media jumping on what Intel "didn't say".  

See above, losing market share does not directly lead to selling less. Regardless, it will take time for the difference to be clearly visible - our best estimates come from official sales figures and stuff like steam surveys, which means they are either subject to temporary anomalies (in either direction) or influenced by nearly two decades of Intel dominance. What matters is that AMD made a decent profit for the first time in years and they have an incredibly competitive architecture which could be what they need to claw their way back to a significant market share across the board.

[Sauron]

7 minutes ago, cj09beira said:

when amd starts selling 7nm 64 core cpus that will ust drop into already working servers with a bios upgrade, thats when the penetration will peak

Enterprise customers don't tend to instantly upgrade to new products, but certainly not having to toss 80% of their hardware every time is a nice bonus.

[mr moose]

5 minutes ago, cj09beira said:

when amd starts selling 7nm 64 core cpus that will ust drop into already working servers with a bios upgrade, thats when the penetration will peak

that's so far away that if Intel aren't prepared for it then they will deserve what they get.

 

2 minutes ago, Sauron said:

Well duh, when your competitor introduces a compelling product a market share hit is to be expected. Note that this doesn't necessarily mean Intel will be losing customers or selling less product, it just means AMD is going to sell more product than before.

See above, losing market share does not directly lead to selling less. Regardless, it will take time for the difference to be clearly visible - our best estimates come from official sales figures and stuff like steam surveys, which means they are either subject to temporary anomalies (in either direction) or influenced by nearly two decades of Intel dominance. What matters is that AMD made a decent profit for the first time in years and they have an incredibly competitive architecture which could be what they need to claw their way back to a significant market share across the board.

I never said Intel will sell less, however for every AMD chip sold is one less bought from Intel, so that's still a lost sale which reduces potential revenue.

[laminutederire]

2 hours ago, mr moose said:

I only want to see his words because it's rare for a CEO to admit they will loose market share to a competitor.  Usually they only make such admissions after the fact when there is nothing to point the blame at.  Claiming they will makes the investors ask hard questions like why? and saying things like: if this is true you shouldn't be the CEO.

Why? A good CEO can't win all the time. Denying you'll be losing market share can be seen from investors that you do not see the obvious problem at hand and therefore bad leadership.

[Tribalinius]

Epyc is getting traction at least and that's good news for everyone in the server market.

As leadeater mentionned, not everyone will change their servers by next year and I would not expect companies jumping platform on a whim just for the sake of it. It's usually a very long process and given Epyc is not even a year old yet I'd assume a lot of companies that would be interested in jumping platform will wait to see their long term reliability and/or their next upgrade cycle before considering it and continue buying Intel platform for continuity/compatibility reasons. You don't want to support 2 platforms at the same time unless it's already planned that way or you really hate system engineers .

 

As for new sales, oh boy that thing is a compelling choice. Intel salesmen will need to work harder this time around to convince people buying into their products. 15-20% is realistic, I'd even push up to 25% if they got enough hardware to sell.

[cj09beira]

22 minutes ago, mr moose said:

that's so far away that if Intel aren't prepared for it then they will deserve what they get.

 

I never said Intel will sell less, however for every AMD chip sold is one less bought from Intel, so that's still a lost sale which reduces potential revenue.

thats probably going to happen in little over 6 months (might not be 64 cores and be 48 but the result should be similar)

[Heesleemer]

Intel - Silly 

AMD - SEND IT 

#FullSend

[mr moose]

7 hours ago, laminutederire said:

Why? A good CEO can't win all the time. Denying you'll be losing market share can be seen from investors that you do not see the obvious problem at hand and therefore bad leadership.

Because as I said, it's rare for any CEO to make such admissions.

7 hours ago, cj09beira said:

thats probably going to happen in little over 6 months (might not be 64 cores and be 48 but the result should be similar)

How many companies do you think already have epyc servers now and are going to upgrade in 6 months? Average lifespan for the average company last I read was about 2 years ( @leadeater can weigh in here), we're only coming up to 12 months now and their uptake in the industry hasn't exactly been stellar.  In 6 months market share figures will change due new purchases. not hardware upgrades.  There just isn't enough epyc servers out there for CPU only upgrades to be a thing for at least another 3 years, and probably not for another 2 on top of that for it to be of any concern for Intel.  And this is assuming the first series of epyc boards actually maintain sufficient feature compatability to make a CPU only upgrade actually worth it (we have to keep in mind that the business mindset is vastly different to the domestic one when purchasing tools). 

[BlueChinchillaEatingDorito]

Time will tell if EPYC is reliable enough for long term server use. Based on the number of seemingly odd problems people have been experiencing with Ryzen, I wouldn't put my hopes up too quickly. 

[cj09beira]

32 minutes ago, mr moose said:

How many companies do you think already have epyc servers now and are going to upgrade in 6 months? Average lifespan for the average company last I read was about 2 years ( @leadeater can weigh in here), we're only coming up to 12 months now and their uptake in the industry hasn't exactly been stellar.  In 6 months market share figures will change due new purchases. not hardware upgrades.  There just isn't enough epyc servers out there for CPU only upgrades to be a thing for at least another 3 years, and probably not for another 2 on top of that for it to be of any concern for Intel.  And this is assuming the first series of epyc boards actually maintain sufficient feature compatability to make a CPU only upgrade actually worth it (we have to keep in mind that the business mindset is vastly different to the domestic one when purchasing tools). 

i didn't say that only upgrading the cpu was the best corse of action for most people, but my point is that servers capable of using that cpu are already on the market which should help companies test out epyc, as if they already have test servers they can change only the cpu and test the new gen of epyc cpus with an hardware that they already put through its paces.

btw what i meant was that the launch of epyc 2 might be in around 6 months

[mr moose]

2 minutes ago, cj09beira said:

i didn't say that only upgrading the cpu was the best corse of action for most people, but my point is that servers capable of using that cpu are already on the market which should help companies test out epyc, as if they already have test servers they can change only the cpu and test the new gen of epyc cpus with an hardware that they already put through its paces.

btw what i meant was that the launch of epyc 2 might be in around 6 months

 

9 hours ago, cj09beira said:

when amd starts selling 7nm 64 core cpus that will ust drop into already working servers with a bios upgrade, thats when the penetration will peak

It reads very much like you are claiming in 6 months Intel will shit themselves because companies won't have to buy whole new servers just some epyc CPU's.

[leadeater]

45 minutes ago, mr moose said:

How many companies do you think already have epyc servers now and are going to upgrade in 6 months? Average lifespan for the average company last I read was about 2 years ( @leadeater can weigh in here), we're only coming up to 12 months now and their uptake in the industry hasn't exactly been stellar. 

Ours all run for 4 years, not too many go shorter than 3 years unless they don't have too many servers. We have no plans currently to put any EPYC servers in for prod use or even buy any tbh, we're basically stuck in what ever Nutanix supports so if they don't support it (they don't) then we can't no matter how much I want to.

 

49 minutes ago, mr moose said:

And this is assuming the first series of epyc boards actually maintain sufficient feature compatability to make a CPU only upgrade actually worth it (we have to keep in mind that the business mindset is vastly different to the domestic one when purchasing tools). 

This basically isn't a thing anyway, no one upgrades CPUs in existing servers. Server OEMs will start selling servers with the new CPUs but no one will upgrade (very few anyway).

[cj09beira]

7 minutes ago, mr moose said:

 

It reads very much like you are claiming in 6 months Intel will shit themselves because companies won't have to buy whole new servers just some epyc CPU's.

i am not great at expressing myself, but i wasn't trying to imply that, what i meant is that intel will be shitting themselves because amd will have many many more cores than them, being able to test/use old servers is just a plus for those who want to do it

[Lurick]

52 minutes ago, mr moose said:

Because as I said, it's rare for any CEO to make such admissions.

How many companies do you think already have epyc servers now and are going to upgrade in 6 months? Average lifespan for the average company last I read was about 2 years ( @leadeater can weigh in here), we're only coming up to 12 months now and their uptake in the industry hasn't exactly been stellar.  In 6 months market share figures will change due new purchases. not hardware upgrades.  There just isn't enough epyc servers out there for CPU only upgrades to be a thing for at least another 3 years, and probably not for another 2 on top of that for it to be of any concern for Intel.  And this is assuming the first series of epyc boards actually maintain sufficient feature compatability to make a CPU only upgrade actually worth it (we have to keep in mind that the business mindset is vastly different to the domestic one when purchasing tools). 

Yah, most installs for Epyc will be new installs as a company grows and expands their data center or servers going in to refresh already aged servers that happen to line up with the new servers coming on the market. I think Dell was the first to announce Epyc servers back in Feb and HP wasn't far behind. Not sure when SuperMicro jumped on board and Cisco launched them two weeks ago so they are still very new in the market and I doubt a lot of refreshes will be happening in the enterprise right now.

[mr moose]

2 minutes ago, Lurick said:

Yah, most installs for Epyc will be new installs as a company grows and expands their data center or servers going in to refresh already aged servers that happen to line up with the new servers coming on the market. I think Dell was the first to announce Epyc servers back in Feb and HP wasn't far behind. Not sure when SuperMicro jumped on board and Cisco launched them two weeks ago so they are still very new in the market and I doubt a lot of refreshes will be happening in the enterprise right now.

I wasn't aware that Epyc uptake was that late in the game.  

[Epic Wink]

As a software developer, there would have to be a huge incentive for me to go off intel. A lot of the developer community would have to make a great push to support AMD, even though they use the same instruction set. Things like JavaScript and (pure) Python shouldn't be too much of an issue, but there is an insane amount of precompiled libraries out there (which I use everyday) which have too many issues with non-Intel. This is ultimately the biggest roadblock AMD will have

[DocSwag]

15-20% would be huge for AMD. I'm guessing that's above 500 million dollars a quarter in revenue, which is over 30% of AMD's total quarterly revenue rn. My stock sure would be happy about that  

[cj09beira]

7 hours ago, DocSwag said:

15-20% would be huge for AMD. I'm guessing that's above 500 million dollars a quarter in revenue, which is over 30% of AMD's total quarterly revenue rn. My stock sure would be happy about that  

na its amd we are talking about, it would probably go down  

[Tribalinius]

15 hours ago, mr moose said:

How many companies do you think already have epyc servers now and are going to upgrade in 6 months? Average lifespan for the average company last I read was about 2 years ( @leadeater can weigh in here),

It really depends of the context. 2 years is very short for an average SMB who leaves it in the closet compared to a datacenter. 4 years is being proactive and 5-6 years is the norm around here. It's an hefty chunk of change when you upgrade an SMB. We just did a quote a couple days ago for a company and we were looking at 13k CAD$ just for a single server. When we finished adding windows/software licenses + our configuration/transfer fees we were at almost 50k CAD$ + an another 10k CAD$ for a brand new phone system. For a 50-75 employee business, 60k CAD$ is a good amount and I clearly understand them to try and spread that on 5-6 years .

That's one of the reason why we are making a push for the Cloud in general. Dealing with hardware/software licenses upgrade is a pita for a normal SMB. It generally means a lot of money and downtimes plus whatever hardware problems that might happen down the line.
 

14 hours ago, leadeater said:

Ours all run for 4 years, not too many go shorter than 3 years unless they don't have too many servers. We have no plans currently to put any EPYC servers in for prod use or even buy any tbh, we're basically stuck in what ever Nutanix supports so if they don't support it (they don't) then we can't no matter how much I want to.

 

This basically isn't a thing anyway, no one upgrades CPUs in existing servers. Server OEMs will start selling servers with the new CPUs but no one will upgrade (very few anyway).

You got a good point, server platforms need to be compatible before you can even think about switching. No one upgrades OEM servers CPUs. You better have a really good reason to do it if you need too. If I don't do it in standalone OEM tower servers (unless it's to drop a second one in it), I hardly imagine anyone doing it in datacenters.

[leadeater]

1 hour ago, Tribalinius said:

It really depends of the context. 2 years is very short for an average SMB who leaves it in the closet compared to a datacenter. 4 years is being proactive and 5-6 years is the norm around here.

We used to do 5 years but our capacity expansion requirements were too fast which then made it hard to ensure the same generation of server could be purchased, not too big of a deal with EVC modes but not great to have to mixed hardware generations within a cluster. Was easier and overall cheaper to drop to 4 year life cycle.

 

One of the down sides to hyper-converged systems (HCI) I've seen is the ability to scale out is relied on too easily so rather than enforcing sensible provisioning and VM life cycles hosts are just purchased instead, that's how we've doubled the number of ESXi hosts in 2-3 years and we're over 80 of them now. Admittedly they are lower spec to what we used to buy, HPE DL360 2x E5-2690v4 386GB vs Nutanix 2x E5-2640v4-2670v4 (don't get me started on the mixed CPUs grrr, zero involvement with that), so the number had to increase but double or more is a bit silly

 

To be honest HCI lives up only to the first 4 letters..... if ya catch my drift .

[Jito463]

17 hours ago, BlueChinchillaEatingDorito said:

Time will tell if EPYC is reliable enough for long term server use. Based on the number of seemingly odd problems people have been experiencing with Ryzen, I wouldn't put my hopes up too quickly. 

Miscellaneous issues aren't unique to Ryzen or even AMD in general.  There's a reason the term "errata" is an industry standard.

[Tribalinius]

48 minutes ago, leadeater said:

We used to do 5 years but our capacity expansion requirements were too fast which then made it hard to ensure the same generation of server could be purchased, not too big of a deal with EVC modes but not great to have to mixed hardware generations within a cluster. Was easier and overall cheaper to drop to 4 year life cycle.

 

One of the down sides to hyper-converged systems (HCI) I've seen is the ability to scale out is relied on too easily so rather than enforcing sensible provisioning and VM life cycles hosts are just purchased instead, that's how we've doubled the number of ESXi hosts in 2-3 years and we're over 80 of them now. Admittedly they are lower spec to what we used to buy, HPE DL360 2x E5-2690v4 386GB vs Nutanix 2x E5-2640v4-2670v4 (don't get me started on the mixed CPUs grrr, zero involvement with that), so the number had to increase but double or more is a bit silly

 

To be honest HCI lives up only to the first 4 letters..... if ya catch my drift .

Obivously you and I are dealing with each end of the spectrum . A 4 year cycle in a datacenter makes total sense but, with a slow moving SMB, 5-6 years is still ok. Some companies I work with can be years without having any need of a new VM while others require that I'd do that on a nearly daily basis. Sometimes, I get really excited when I get to play with a server cluster once in a while just because I'm bored playing with "normal" stuff .