How to take 50% market share in three years.

Page 3 - 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
#51
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.
You're right, I was wrong. I've changed the post. Thanks.

That said, I consider that roadmap to be blue sky. Intel is a lot further behind than the roadmap indicates. Not that Intel would try and mislead investors ... o_O
 
Last edited:

UsandThem

Super Moderator
Super Moderator
May 4, 2000
10,706
488
136
#52
The actual roadmap shows Intel far behind AMD through 2023.
And Enron's road-map showed them growing as well.

A road-map is a plan, largely for investors, and not results. The only way to know if your bold prediction will pan out, is to come back in 2 years.
 

maddie

Platinum Member
Jul 18, 2010
2,512
418
136
#53
Anyone here knows the actual annual server CPU sales number? In cores?

Charlie at Semiaccurate has strongly hinted that AMD has changed the # of chiplets. They might stay with an 8C basic die unit organized as today and up core counts by increasing chiplets.. For certain, the higher core variants have prices justifying the use of relatively small interposers for advanced topologies as presented by their researchers.

We won't know the truth until next year, but I was always mystified as to how they are progressing to 7nm extremely early and also keeping basic unit size around 200 mm^2 by upping to 16 cores. Staying with an 8C unit means a very small die. Sub 100mm^2 with commensurate yield advantages.

A 0.25 defect density in the die yield calculator gives roughly 467 die/wafer for a 8C Summit Ridge class design. 10,000 wafers/month allows ~14 million 32 core equivalent CPUs annually.
 
Jul 19, 2016
135
0
71
#54
If AMD really is confident with Zen2 and 7nm a would assume they have 15-20m wafers/month as minimum - I bet AMD could even outlpay Nvudia here if they wanted to an if they have such a winner in their pockets as this would benefit AMD on the GPU side as well.

However I strongly disbelieve in the uncore and/or 8c chiplet approach for Zen2 - could AMD actually fit 8x 8c modules on SP3 interposer??
 
Last edited:

Shivansps

Platinum Member
Sep 11, 2013
2,462
194
126
#55
A bit too optimistic considering they had to sell 32C Epycs as TR, more than likely because they were not selling well.
 

NTMBK

Diamond Member
Nov 14, 2011
8,280
247
126
#57
A bit too optimistic considering they had to sell 32C Epycs as TR, more than likely because they were not selling well.
I suspect it has more to do with the 28 core i9 which is due out soon. AMD couldn't let Intel win the core war!
 

maddie

Platinum Member
Jul 18, 2010
2,512
418
136
#58
If AMD really is confident with Zen2 and 7nm a would assume they have 15-20m wafers/month as minimum - I bet AMD could even outlpay Nvudia here if they wanted to an if they have such a winner in their pockets as this would benefit AMD on the GPU side as well.

However I strongly disbelieve in the uncore and/or 8c chiplet approach for Zen2 - could AMD actually fit 8x 8c modules on SP3 interposer??

Gesendet von meinem SM-G955F mit Tapatalk
Easily.

Remember an 8C 7nm die would be at most, 50% of the 14nm size. So the total area will equal the silicon area of the present 4 8C dies. Also important to keep in mind, the routing density available with an interposer. The present 4 die package needs a lot of space between dies due to the organic layer package used. A SI interposer will allow little space wasted between dies. A lot more free space made available.
 
Jul 19, 2016
135
0
71
#59
Easily.

Remember an 8C 7nm die would be at most, 50% of the 14nm size. So the total area will equal the silicon area of the present 4 8C dies. Also important to keep in mind, the routing density available with an interposer. The present 4 die package needs a lot of space between dies due to the organic layer package used. A SI interposer will allow little space wasted between dies. A lot more free space made available.
Ok, thx.
 

BigDaveX

Senior member
Jun 12, 2014
330
44
101
#60
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
I'm actually surprised that AMD's market share tapered off as quickly as it did. The general consensus always seems to have been that the Core 2-based Xeons were still pretty poor products compared to their Opteron rivals and that it wasn't until Nehalem that Intel's server line-up really recovered, but it doesn't seem to have made as much of a difference as you'd have thought judging by that graph.
 

NTMBK

Diamond Member
Nov 14, 2011
8,280
247
126
#61
I'm actually surprised that AMD's market share tapered off as quickly as it did. The general consensus always seems to have been that the Core 2-based Xeons were still pretty poor products compared to their Opteron rivals and that it wasn't until Nehalem that Intel's server line-up really recovered, but it doesn't seem to have made as much of a difference as you'd have thought judging by that graph.
AMD really screwed up with Barcelona, with the TLB bug. I think that killed a lot of the trust and momentum they had built up.
 

ub4ty

Senior member
Jun 21, 2017
749
321
96
#62
There are a couple of perspectives here.
From a shareholder's perspective, it'd be great for AMD to get 50% market share.
From a consumer standpoint, it really doesn't matter and might actually negatively impact the great deals I get on their hardware... I'd expect AMD to gouge just as Microsoft did if they get head room.
From an enterprise standpoint, there are things far and beyond 'best performing' product that win contracts. There's a sound and established strangle hold intel has in the enterprise. This doesn't apply to up and coming tech companies that have the ability to experiment/break the mold. It doesn't apply to established companies that see the value and are rolling out EPYC installs. It also doesn't apply to the ever growing Chinese market that AMD is increasingly establishing long lasting relationships/partnerships and buy-ins with.

If someone is restricted from the EPYC platform due to superficial/legacy/political or personal issues, sucks for them. The Ryzen platform captured me on launch. I went from multiple Ryzen 8 core setups to Threadripper. Next, I will be investing in the EPYC platform. I can't imagine being won over by Intel for some time. Their pricing and artificial segmentation are absolutely absurd. This is after having a decade+ worth of strictly Intel builds.
 

BigDaveX

Senior member
Jun 12, 2014
330
44
101
#63
AMD really screwed up with Barcelona, with the TLB bug. I think that killed a lot of the trust and momentum they had built up.
It looks like AMD lost most of their server marketshare at around the end of 2006 and start of 2007 - which was the time that Intel launched their first quad-core Xeons. For some reason I thought AMD had launched K10 Opterons in the early part of 2007, but it actually seems not to have been until September they did that.

So yeah - even if they were strangling each other through the shared FSB and forced to deal with horrendous FB-DIMM memory latency, four Xeon cores were still going to outgun two K8 cores, and server admins would probably have been reluctant to trust AMD after the TLB fiasco even if their chips did perform better. Makes sense why they lost marketshare so quickly.
 

ub4ty

Senior member
Jun 21, 2017
749
321
96
#64
Pretty active day for me...
I've been keeping a track of how the enterprise landscape is evolving and had some prior predictions leading up until now...

I've largely kept a close eye on the ever evolving ecosystem centered around NVME drives.
One glaring detail people overlook is that an NVME drive, upon inspection, is essentially a multi-core computer on a gum-stick. You have a quad core ARM processor steering traffic to/from flash through controllers hanging off of a PCIE x4.0 interface.

Were essentially already in a new age of computing.
One thing that I questioned was the scaling of this...
A lot of the current systems necessitate custom EPYC/XEON boards interfacing to an NVME array connected by an oversubscribed PLX PCIE switch.

Today I read that Mellanox is integrating this into a new speciality dual port NIC 100Gb NIC.
This and likely new technologies like it will completely divorce expensive AMD/INTEL systems from a storage array with even better performance and lower latencies.

I always felt that the next revolution in computing would center on storage and its happening right before our eyes !

In the years to come, there is going to be huge competition all over the hardware sector for everyone.
I fully expect the concept of 'accelerator' cards and HSA to begin becoming more fully realized as PCIE 4.0 comes out and hopefully plx pcie switches become more common in desktop computers allowing for better scaling. GPUs/CPUs are going to have to fight for and share their slices with a larger number of people.

https://www.servethehome.com/mellanox-bluefield-bf1600-and-bf1700-4-million-iops-nvmeof-controllers/


Thank you ARM ! Welcome to the forefront FPGA !
P.S - I look forward to intel doing something special with their FPGA acquisition to distinguish themselves in the times ahead.. they better and better do it right before someone else does.
 
Last edited:
Apr 8, 2002
40,966
73
126
#65
The genius of Lisa Su.

The creme-de-la-creme of the server market is the big cloud players who buy chips by the millions. That is who AMD has already sampled EPYC 2 to.

AMD is offering a chip, EPYC 2, that will have a !!!!! ~3X !!!!! perf/watt/$ advantage over Intel and a roadmap that will maintain that level of advantage for the next three years vs the now stalled out Intel.

Perf/watt/$ is king. 3X perf/watt/$ is king of kings. That is literally an offer the big cloud players can't refuse. And won't refuse. They will wholesale turn to AMD for their new build outs.

Taking the first 50% of the server market won't be hard, the second 50% will be slower, but by then every OEM will have a full line of AMD servers and be pushing them because if they don't their competition will and take their business.

The further genius was switching Zen 2 to an uncore+chiplet architecture mid development cycle to double the number of CPUs available to service the anticipated demand, substantially cut EPYC 2's cost, seriously advance the performance of the roadmap and, through the uncore, provide for myriad customizations across the CPU market and even within a market.

AMD could have half a dozen custom uncores planned for Zen 2 and rapidly and easily steer their 7nm chiplets during the packaging step to whatever market segment needs more supply.
In the past when AMD was able to sell everything they could produce. They maxed around 20ish % of market share. Has AMD expanded production enough to take 50% market share? I highly doubt it. Having the best part in the market is only part of the puzzle in gobbling up market share. And given there is more to the server market than just the CPU. I am not convinced AMD has the best part anyways.
 

teejee

Senior member
Jul 4, 2013
260
6
101
#66
I am not convinced AMD has the best part anyways.
Well, there are some really good AMD servers out there:

From https://www.servethehome.com/dell-emc-poweredge-r7415-review/

"The Dell EMC PowerEdge R7415 is an insanely good server. In fact, it may be the best single socket server on the planet. The server features a single socket AMD EPYC platform with an enormous capacity for high-speed networking, NVMe storage, memory capacity, cores counts, and just about every other aspect that makes a server good. In our Dell EMC PowerEdge R7415 review we are going to go in-depth to show you why this is a disruptive server regarding price, performance, and capabilities. "
 
Dec 28, 2013
53
0
66
#67
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.
Zen brings AMD to parity with Intel, 7nm Zen 2, already sampled to partners, slaughters Intel.
 
Last edited:
Dec 28, 2013
53
0
66
#68
However I strongly disbelieve in the uncore and/or 8c chiplet approach for Zen2 - could AMD actually fit 8x 8c modules on SP3 interposer??
It's an 8 core CCX, there are two of them on a die, so 16 cores per die. 4 X 16 + 64. So .. no problem.
 

maddie

Platinum Member
Jul 18, 2010
2,512
418
136
#69
It's an 8 core CCX, there are two of them on a die, so 16 cores per die. 4 X 16 + 64. So .. no problem.
I have not seen any evidence, versus speculation, as to the Zen2 CCX having 8 cores. Have you?

If yes, please present.
 
Dec 28, 2013
53
0
66
#70
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.
Do you even READ Intel negative posts???

AMD has a KILLER roadmap for the next five years. Intel has a keystone kops roadmap for the next five years. Intel CHANGEs their roadmap every two months. That instill confidence in Intel's future.
 
Last edited:
Dec 28, 2013
53
0
66
#71
In the past when AMD was able to sell everything they could produce. They maxed around 20ish % of market share. Has AMD expanded production enough to take 50% market share? I highly doubt it. Having the best part in the market is only part of the puzzle in gobbling up market share. And given there is more to the server market than just the CPU. I am not convinced AMD has the best part anyways.
Hence all the rumors and logic of Zen 2 being changes to an uncore+chiplet architecture to take advantage of Intel's 10nm redacted.

It DOUBLES the number of 7nm core dies that can be produced. Add sourcing from two foundries and there will be enough dies to allow a 25% server market share in 2019.




[/b]
Profanity in tech is not allowed.


esquared
Anandtech Forum Director[/b]
 
Last edited by a moderator:
Dec 28, 2013
53
0
66
#72
I have not seen any evidence, versus speculation, as to the Zen2 CCX having 8 cores. Have you?

If yes, please present.
There are numerous reports from normally reliable sources. It won't be long before someone working on an actual sample leaks detailed info. And it's utterly logical AMD went that route.
 

NTMBK

Diamond Member
Nov 14, 2011
8,280
247
126
#73
Do you even READ Intel negative posts???

AMD has a KILLER roadmap for the next five years. Intel has a keystone kops roadmap for the next five years. Intel CHANGEs their roadmap every two months. Do you think this instills confidence in their customers?
Indeed, that's why I put "if" in my post ;) I was just clarifying what Intel is trying to convey in that image. I have little faith that they will deliver on schedule, especially the 10nm parts.
 
Dec 28, 2013
53
0
66
#74
Indeed, that's why I put "if" in my post ;) I was just clarifying what Intel is trying to convey in that image. I have little faith that they will deliver on schedule, especially the 10nm parts.
I hang my head in embarrassment. :oops:
 

PeterScott

Platinum Member
Jul 7, 2017
2,605
227
96
#75
There are numerous reports from normally reliable sources. It won't be long before someone working on an actual sample leaks detailed info. And it's utterly logical AMD went that route.
This is utter speculation on your part. You are building a house of cards on speculations like this and arguing that the final outcome has some factual value. It really doesn't.

I have seen speculation, that the 7nm Zen 2 CCX will have 8 cores, but I have also seen 6 and 4 cores speculated.

I find the 8 core version least likely. Even 4C CCX x2 (8cores) is already kind of overkill for desktop, so jumping to 8C CCX and thus 16 core desktop, just pushes things where they don't need to go, at the cost of a much larger die across the board.

I could see AMD going one more generation with 4C CCX. there is a certain elegance in the 4 unit CCX, and the way each die has the same access speed to any of the cache slices, via a direct access connection. But it gets messier with 6 cores, and messier still with 8 cores.

If they have IPC improvement and clock speed improvements, the could have one more generation of 4 core CCX, and have a very good competitive position, and a very tiny, high yield die to make a lot of profit on.

Or they could go for 6 core CCX, and bump core counts everywhere by 50%, given them a core count advantage that Intel is unlikely to address for a while, and still have a relatively, moderately small die, to make moderate profits on.

8C CCX/16 core desktop is probably more expensive to produce than the current die, while you still need parts at desktop prices, so it isn't really the most profitable outcome, and ultimately profit is what they are after.
 


ASK THE COMMUNITY

TRENDING THREADS