Intel CEO Bracing For EPYC Impact, Aims to Keep AMD Under 20% of Server Market Share

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Zstream

Diamond Member
Oct 24, 2005
3,396
277
136
Don't attribute to malice what can be explained by stupidity or so the saying goes.

I'm by no way defending intel but offering my thought about typical management peoples technical knowledge and intelligence.

Yep! We could not purchase AMD gear, because the architects said no. It wasn’t due to Intel malice, but general stupidity among people. You have to remember, most people shouldn’t be in the position they’re in.
 
  • Like
Reactions: lightmanek

zinfamous

No Lifer
Jul 12, 2006
110,512
29,098
146
Isn't 20% of server market share like more revenue than AMD currently has? And if AMD captures even 5% of the server market profit, that's vastly more than they currently have?

If Intel thinks 20% is reasonably achievable, maybe the AMD stock price is justified after all.

I wouldn't say the stock price is justified at 20% server market share: it's vastly undervalued. It is currently justified at 5% market share, imo, which seems to be the generally optimistic side of predictions as much as a year ago for AMD.

20% is quite insane, within ~1 year from now. Hold on to your butts.
 

dark zero

Platinum Member
Jun 2, 2015
2,655
138
106
I know at work that we have built at least a few custom Epyc servers for clients. Seems to a be a promising product line.
Here in Peru some clients are thinking on go Ryzen as secondary servers due cost and due Intel's potential risks with Spectre (they finally realized how risky is that)... They are maintaining their old servers but is unavoidable to buy newer ones.
 

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
25,478
14,434
136
Here in Peru some clients are thinking on go Ryzen as secondary servers due cost and due Intel's potential risks with Spectre (they finally realized how risky is that)... They are maintaining their old servers but is unavoidable to buy newer ones.
When you say Ryzen, do you really mean EPYC ? Thats the server line.
 
  • Like
Reactions: Drazick

bbhaag

Diamond Member
Jul 2, 2011
6,604
1,990
146
what are you talking about, it doesn't have any basis in reality......

Their first, not even real attempt was in 2001 with athlon MP (17 years ago) , With the first real attempt in 2003 (15 years ago). from 2003 until 2012 AMD had success in the server market ( 2010 Nehalem with QPI saw the start of the main decline). For along time AMD's market limiter was there ability to manufacture. Now Zepplin is comparable to intel in terms of performance but AMD's design decisions weren't about building something on a mature 14nm process, they will really show on the much more expensive to develop for , much lower yielding 7nm processes.

Now how about 2019 for you......

Intel with 28 core 14nm CPU's
Intel with same IPC
AMD will be on 7nm with increased clock at same core or more cores at same power
AMD with either 48 or 64 core (48 was original rumors, CPC later said 64) per processor
AMD will have an increase in IPC
AMD will have ~800mm sq of 7nm per socket of silicon to offer during 7nm infancy thax to the multi die strategy
AMD is shifting products back to TSMC, so AMD will have more 7nm capacity for CPU SOC then they do now for 14nm

Want to revisit @ EOFY 19?
We'll see. I've heard this before. Eight years ago when I joined this forum AMDs new Bulldozer architecture was going to revolutionize pc's with it's core count, then when that failed I heard about how AMD won the console bid. That was going to revolutionize pc's because now everything was going to be optimized for AMD. Then that failed.....Guys like you always look toward the future with 2019 predictions of what might happen.
Like I said AMD has a solid chip with solid performance but gaining market share into a realm that has been dominated by Intel for years will be very difficult.
 

piesquared

Golden Member
Oct 16, 2006
1,651
473
136
We'll see. I've heard this before. Eight years ago when I joined this forum AMDs new Bulldozer architecture was going to revolutionize pc's with it's core count, then when that failed I heard about how AMD won the console bid. That was going to revolutionize pc's because now everything was going to be optimized for AMD. Then that failed.....Guys like you always look toward the future with 2019 predictions of what might happen.
Like I said AMD has a solid chip with solid performance but gaining market share into a realm that has been dominated by Intel for years will be very difficult.

You can't be serious. Are you forgetting that people have and can see the value/performance proposition that Epyc offers and how in many cases it is by far the better option. If anything, those examples you listed show the stark contrast between how much bigger of an impact Epyc is going to be. There is no comfort in sight for intel with AMD bringing out Zen 2 on 7nm next year, there's no way they can fend off 7nm Zen 2 especially when their process lead has evaporated and even went backwards! due to the fail at 10nm. This is nothing like those times, it's clear AMD has a winning architecture and associated IP.
 
  • Like
Reactions: DarthKyrie

AtenRa

Lifer
Feb 2, 2009
14,000
3,357
136
Gaining server market share from Intel will be very difficult and time consuming for AMD the first years of EPYC introduction (2017-2018) and even in 2019-2020. But if AMD manage to get 5% in 2018 it will be a huge huge win for them. 5% of the server market with 40% margins will increase their income substantially.
 

maddie

Diamond Member
Jul 18, 2010
4,723
4,628
136
The future is always like yesterday until it isn't. By then it's often too late. The state of most.
 

bbhaag

Diamond Member
Jul 2, 2011
6,604
1,990
146
You can't be serious. Are you forgetting that people have and can see the value/performance proposition that Epyc offers and how in many cases it is by far the better option. If anything, those examples you listed show the stark contrast between how much bigger of an impact Epyc is going to be. There is no comfort in sight for intel with AMD bringing out Zen 2 on 7nm next year, there's no way they can fend off 7nm Zen 2 especially when their process lead has evaporated and even went backwards! due to the fail at 10nm. This is nothing like those times, it's clear AMD has a winning architecture and associated IP.
Of course I'm serious. What do you think Epyc is? Is it this magic wand that sprinkles market share dust all over and then AMD magically gains 20% market share in a one year period. hahaha
 

maddie

Diamond Member
Jul 18, 2010
4,723
4,628
136
Gaining server market share from Intel will be very difficult and time consuming for AMD the first years of EPYC introduction (2017-2018) and even in 2019-2020. But if AMD manage to get 5% in 2018 it will be a huge huge win for them. 5% of the server market with 40% margins will increase their income substantially.
Not too sure about that. See above post. Reason in this case?

China. The old days of the West dominating everything technology wise is dying.

https://www.wsj.com/articles/amd-to-license-chip-technology-to-china-chip-venture-1461269701
 

AtenRa

Lifer
Feb 2, 2009
14,000
3,357
136
Of course I'm serious. What do you think Epyc is? Is it this magic wand that sprinkles market share dust all over and then AMD magically gains 20% market share in a one year period. hahaha

Nobody said or implied AMD gaining 20% market share in a year.
 

AtenRa

Lifer
Feb 2, 2009
14,000
3,357
136
Not too sure about that. See above post. Reason in this case?

China. The old days of the West dominating everything technology wise is dying.

https://www.wsj.com/articles/amd-to-license-chip-technology-to-china-chip-venture-1461269701

It takes more than a good product to take market share, especially from Intel in the Server segment.
But as I have said, a 5% gain in server market share for AMD is huge for them. Just with 5% of market share which is around 1B of Revenue with 40% margins will get them 400M of income.
 

JoeRambo

Golden Member
Jun 13, 2013
1,814
2,105
136
It takes more than a good product to take market share, especially from Intel in the Server segment.

And unlike certain claims in the thread, from purely performance side Intel has architecture advantages in Server segment. 28 core monolithic chip without NUMA craziness is better in all workloads. INT IPC, vector FP throughput, inter core comms, you name it, Intel has advantages everywhere.
Unlike Opteron days, where Athlons had IMC and great architecture, while Intel was stuck with FSB and throwing cache at problems, Intel has advantages in performance and architecture overall.

The real worry for Intel is margins and gradual loss of their hegemonic position in lucrative server market. Since Ryzen is MCM, obviously there is huge advatange in yields, AMD is content with selling 32 cores worth 768mm^2 of silicon , while Intel has to build ~694mm^2 massive chip with 28 cores.
So 14nm vs 14nm, obvious choice is Intel, unless you can't get good price out of them*. Or maybe Your workload fits EPYC like glove ( like smallish VM, that can be bound to CCX, or some niche I/O or high mem needs ).

We can fully expect that to change on 10nm. For AMD obviously they will be able to build a chip with larger module/chip and provide more cores per socket at uptuned performance, that will get nearer. For Intel? I have no clue what they will do, but obviously there is nothing to stop them from building MCM chips themselves, it's not like Infinity Fabric is magic dust, Intel can take 4 shrunken HCCs core (or 2 XCC), slap them on chip with HMB and sell those ~56-72 cores ~1000**mm of silicon on 14nm, who knows how much silicon @ 10nm.

In fact they might already be doing that, with rumors surfacing about Intel Cascade Lake-AP “Advanced Processor Family” (BGA 5903). so many pins can surely fit a ton of I/O and power delivery to feed 2 or 4 chip MCM.

* server people are not paying list prices
** before you nitpick, please consider reductions from not needing so many IMC tiles and other savings
 
  • Like
Reactions: ButtMagician

scannall

Golden Member
Jan 1, 2012
1,944
1,638
136
It takes more than a good product to take market share, especially from Intel in the Server segment.
But as I have said, a 5% gain in server market share for AMD is huge for them. Just with 5% of market share which is around 1B of Revenue with 40% margins will get them 400M of income.
AMD has the added bonus of all those Meltdown afflicted Intel Xeons out there in the wild. Very expensive parts, that will have to be replaced. Or suffer really bad performance hits. I am skeptical about the %20 thing. But they will increase market share. Lisa Su has been cautious here, projecting 8 to 10 percent over the next couple years. That doesn't seem over the top.

It's kind of a perfect storm for AMD here. Intel's next process is stuck in the mud. Plus the Meltdown disaster gives them a window to slip in through. There is no reason at all for any large server farm to buy Intel *today*, until Meltdown is addresed in hardware. I'm sure Intel will get it fixed at some point. But it's going to cost them some sales, as it should.
 

Candymancan21

Senior member
Jun 8, 2009
278
3
81
With this CPU being 7nm, i worry because were at the egde of die shrinking now... What then, and dont say quantum computers.. no one wants a 200 ton liquid nitrogren cooled cpu thats still slower then silicone
 

dacostafilipe

Senior member
Oct 10, 2013
771
244
116
... Intel can take 4 shrunken HCCs core (or 2 XCC), slap them on chip with HMB and sell those ~56-72 cores ~1000**mm of silicon on 14nm, who knows how much silicon @ 10nm ...

Is it really that easy?

It's not like AMD updated HyperTransport into Infinity Fabric, slapped it on their CPUs and called it a day. For me, the "magic" stuff, is not the interconnection, but the uncore.

I could be wrong, but I don't think that easy.
 

JoeRambo

Golden Member
Jun 13, 2013
1,814
2,105
136
I would so so love to see that. Talk about truly monolithic.

I think You misunderstood, i meant 2 or 4 dies on MCM, connected via secret sauce.

Is it really that easy?

It's not like AMD updated HyperTransport into Infinity Fabric, slapped it on their CPUs and called it a day. For me, the "magic" stuff, is not the interconnection, but the uncore.

I could be wrong, but I don't think that easy.

AMD inter die latency is ~same as Intel inter CPU latency in dual multi socket , so as long as Intel is willing to drop monolithic, they can simply take a 2 dies and connect them with QPIs (maybe rename them with more flashy name ). I am not versed in "EMIB" advantages, but cores being within centimeters instead of decimeters in socket) and communicating in "substrate" would allow for more BW and/or latency redution.

I'd say main challenge is power density and delivery to socket. Inte's AVX512 mode has ridiculous power usage per core and without 400w or so per socket AVX2/AVX512 clocks would drop to the bottom.
 
  • Like
Reactions: Bubbleawsome

jpiniero

Lifer
Oct 1, 2010
14,510
5,159
136
In fact they might already be doing that, with rumors surfacing about Intel Cascade Lake-AP “Advanced Processor Family” (BGA 5903). so many pins can surely fit a ton of I/O and power delivery to feed 2 or 4 chip MCM.

Cascade Lake-AP is an HPC APU though... the earlier rumor had it targeted at Icelake but it's said to be 2x24 cores with 4 AVX-512 units per core. Could they hack something together that's basically Cascade Lake with EMIB aimed at regular datacenter? I suppose. Would that be competitive with something on TSMC's 7nm? I don't know.

Edit: BK did say they were going to release something with EMIB next year. Now the Cascade Lake-AP is likely what he was talking about but I suppose there could be more.
 
Last edited:

AtenRa

Lifer
Feb 2, 2009
14,000
3,357
136
And unlike certain claims in the thread, from purely performance side Intel has architecture advantages in Server segment. 28 core monolithic chip without NUMA craziness is better in all workloads. INT IPC, vector FP throughput, inter core comms, you name it, Intel has advantages everywhere.

Ehmm no, Intel doesnt have advantages everywhere. AMD EPYC has advantage in Memory per socket (2TB for AMD vs 768MB for Intel), higher number of Integrated PCI-e lanes (128 for AMD vs 48 for Intel), higher perf/$ with higher number of Cores 32 vs 28 (better for VMs), less problems with Meltdown/Spectre with higher I/O performance for HDDs etc vs Intel and also plug-in upgrade path for future 7nm EPYC SOCs.


The real worry for Intel is margins and gradual loss of their hegemonic position in lucrative server market. Since Ryzen is MCM, obviously there is huge advatange in yields, AMD is content with selling 32 cores worth 768mm^2 of silicon , while Intel has to build ~694mm^2 massive chip with 28 cores.
So 14nm vs 14nm, obvious choice is Intel, unless you can't get good price out of them*. Or maybe Your workload fits EPYC like glove ( like smallish VM, that can be bound to CCX, or some niche I/O or high mem needs ).

What this has to do with market share and EPYC sales ??
 

zinfamous

No Lifer
Jul 12, 2006
110,512
29,098
146
We'll see. I've heard this before. Eight years ago when I joined this forum AMDs new Bulldozer architecture was going to revolutionize pc's with it's core count, then when that failed I heard about how AMD won the console bid. That was going to revolutionize pc's because now everything was going to be optimized for AMD. Then that failed.....Guys like you always look toward the future with 2019 predictions of what might happen.
Like I said AMD has a solid chip with solid performance but gaining market share into a realm that has been dominated by Intel for years will be very difficult.

That's because those actually turned out to be terrible products and AMD's Zen line clearly isn't. It's interesting to try and compare two vastly different situations where performance between two products is quantifiably different, and claim that the mood around them is the same.
 

zinfamous

No Lifer
Jul 12, 2006
110,512
29,098
146
Of course I'm serious. What do you think Epyc is? Is it this magic wand that sprinkles market share dust all over and then AMD magically gains 20% market share in a one year period. hahaha

well yeah, that's a fair point. I don't see them hitting that in one year. I'm thinking BK was talking more long term than short term, unless he's suggesting more about the internal problems at Intel.

Still, AMD has been there before with Opteron, so it's not like they can't convince the market that their product is competitive again. Intel was, again, struggling with design and node issues in those days and what is unique now, they are also dealing with Meltdown and Spectre, which isn't really a problem for AMD. It comes to a point where the data keeps piling on and simply can't be denied and hand-waved away, but I still agree that it takes more than a great product to get purchasers and managers to make proper decisions.

I think Epyc is the real gold of the Zen line. People look at Ryzen still being competitive/marginally slower than Intel's Gen 7 or Gen 8, in games, and tend to confuse that with the space Epyc competes in. I'm not a server dude or anything like that, but when I look at the paper, I don't see anything Intel really has to counter Epyc for the next 2 or 3 years. Their recent display with that non-existing 28?-core exotically cooled microwave they showed last week is just downright embarrassing. It smells of fear and desperation and I think it belies even more undisclosed issues with their upcoming roadmap.