EPYC wins in the server room for AMD

Subject: General Tech | August 7, 2018 - 02:18 PM |
Tagged: amd, EPYC, Intel, market share, server

AMD is continuing to see success in the server room, grabbing a bit more market share from Intel this quarter.  The estimated revenue was $57.66m in the Q2 2018 whereas Q1 was $36m, as far as actual market share, AMD has increased their slice of the pie by 1.3% versus an increase of 0.5% this time last year.  AMD is hopeful they can reach 5% by the end of the year; The Register notes that 2.1% or so would be more in line with the current trend.  Regardless this is great news for AMD and indicates the attractiveness of EPYC for those companies looking for server upgrades.

View Full Size

"Aaron Rakers, senior analyst at Wells Fargo, has seen the second 2018 quarter numbers. He told The Register: "Intel's server CPU share is estimated to have dec lined to 98.7 per cent vs 99 per cent in the prior period and 99.5 per cent a year ago."

Here is some more Tech News from around the web:

Tech Talk


Source: The Inquirer

Video News

August 7, 2018 | 03:20 PM - Posted by EstimatesAndGuesstimates (not verified)

If Lisa Su/Management-Team is expecting 5% by the end of the year and Low double digits by the end of 2019, then maybe there is some pent up demand that was due to many server customers waiting for the first generation parts to be fully certified before they would finilize their orders.

AMD's CEO would be in the know about any larger orders from any customers that where just wating for first generaton Epyc to clear the long Vetting/Certification process and maybe, just maybe, Lisa Su is privy to more information than those old hacks at El Reg, good hacks that they are on most things!

That's a terrable chart and many El Reg commentards have noted, and really obsfucates things.

The end of the year is two reporting quarters away so we will not know until some months after the new year on that one, unless some very big deal drops in this next quarter that's yet to be reported.

AMD's Foundry Partners for first generation Epyc have been turning out those first Zen/Zeppelin dies at greater than 80% dies/wafer yield. So AMD probably has some excess stock to be able to meet any unexpected demands. Server parts are kept in production/availability for years longer owing to the longer server/parts availabilies guarantees(5+ years) that the server market demands. So with first generation Epyc those parts will be around for years and so will some Ryzen Pro parts for business Laptop/PC OEMs for those large purchase contract enterprise customers's demands.

August 7, 2018 | 04:15 PM - Posted by NoFlimFlammingoutOfThat (not verified)

S/A and Charlie the D has more to say(1) about Intel A-Day!


"Intel has no chance in servers and they know it

Updated: Numbers straight from the horses mouth"


August 7, 2018 | 04:47 PM - Posted by Alb (not verified)

Why do you really read Charlie?? poor you :).

Funny thing, Intel hold the server market and "has not chances". Idiotic article from an old Intel hater with a unsuccesful and dying site of low end analysis.

He assume AND will execute very very well and Intel very very bad. Childish, this will never happen. In the wost case intel will leverage on prices dumping the small competitor.

August 7, 2018 | 06:17 PM - Posted by LipidMegatonz (not verified)

Intel's margins can not fall below 55% or the Wallstreet Quants will have Intel hung out to dry. Those Chip Fabs are a cash bleeding albatross around Intel's neck with upkeep costs as are those fat middle management pay/perks.

Intel is such a Fat lumbering beast so full of useless middle management and was for years tied to that useless most recently departed CEO(fraternization violations they claim). First generation Epyc already represents a better Price/Performance and Price/Feature metric than Intel's highly segemented(For Profit Milking) Xeon overpriced SKU lineup. Intel can not pivot fast enough to stall AMD's return big time to the competative server market place.

Intel was profiteering when they should have been engineering for the future and Intel should have been investing more in their offerings for the server market instead of billions wasted on contra revenue trying to buy their way into the mobile market. Intel sat of their fat wallet for too damn long while AMD engneered their Epyc return and now the Server Cash Cow is headed towards the AMD barn while Intel fiddled with an IOT/Mobile market that Intel can never hope to win, even with barrels of cash incentives.

10nm delayed for years and unwise investors fooled by Intel's dog and pony show and accounting prestidigitation all while the world passed that Fatzilla by. AMD has now come back so far that Intel is still in a state of shock at just how much value and performance can be had at such a fraction of Intel's up front costs. Old Chuckla's head will be taking out a good 20 mile radius as it flys apart, Scanners Style, at the news. That Fatzilla tried to piviot but the magatons of fat can not be changed quickly in the directional vector so Lardzilla is still in the process of beginning to bend that fat neck against that ocean of thick lipids that fills to almost bursting that bloated scaly hide, so weighed down that Lizoblob is in earth's gravity well!

Intel is still stuck in milking mode and the cow is now on well on the way to becoming beef jerky as Intel's server customers have seen what AMD's Epyc can and does offer in the way of a better value proposition.

'My name is Chipzillmandias, king of kings;
Look on my works, ye Mighty, and despair!'
Nothing beside remains. Round the decay
Of that colossal wreck, boundless and bare
The lone and level silica sands stretch far away.

August 7, 2018 | 07:53 PM - Posted by AnonymouslyAnonymous (not verified)

Couldn't tell ya who would deny that EPYC is a good CPU. However of the major OEM's that provide large chassis to provide the most amount of servers, taking up the least amount of space, how many major players support EPYC so they can just slide in and go? AMD had a major announcement with Cisco not too long ago, but was with one of their lowest selling models. AMD has the CPU, they just need to get to the point where companies can replace their stuff with least effort in order to truly profit.

August 7, 2018 | 09:54 PM - Posted by LotsOfEpycPartners (not verified)

Crey has announced Epyc based SKUs, and Dell, HPE, Lenovo, Microsoft, Baidu. AMD lists about 50 design wins and that Progect 47 system demonstrated at SIGGRAPH 2018 a system with One Petaflop of compute(FP32) in a single cabinet with 20, Epyc 32 core CPUs and 80 Vega 10 Die based Radeon Instinct MI25s.

This article(1) lists:

"HPE's new EPYC 7000-series powered dual-socket ProLiant DL385 Gen10 and the upcoming single-socket ProLiant DL325 Gen10 servers.
A customer that's already making use of ProLiant DL385 Gen10 is the Center for Research Computing (CRC) at the University of Notre Dame. The CRC has several projects on-going that need heavy-duty high-performance computing (HPC), such as the Department of Energy's Center for Shock Wave-processing of Advanced Reactive Materials (C-SWARM), the Computational Hydraulics Laboratory (CHL), and Terrestrial Ecosystems and Climate Modeling projects." (1)

" "The AMD and HPE systems gave us performance that allows larger analytics," said Paul Brenner, Associate Director of the Notre Dame Center for Research Computing. "A lot of these are memory-bound applications, fluid dynamics specifically, so [the systems] gave us a nice architecture upon which we could perform larger memory, high-performance calculations with good performance between the cores and the memory itself." "(1)


"EPYC: One year on and AMD has transformed the server landscape

AMD did something that was long overdue -- it upended a stagnant and complacent server market, and all in a year."


August 7, 2018 | 08:29 PM - Posted by Anonymouse (not verified)

Blimey that is a utterly execrable graph. Whoever drew it deserved to be beaten with their own lopsided axes.

August 8, 2018 | 03:23 AM - Posted by James

The server market just moves slowly. In a lot of cases, configuration decisions have to be made well in advance of when the product is available. Epyc just hasn't had time to succeed yet. Intel is still charging ridiculous prices in the server space though. They are going to keep on going along as if nothing is wrong to try to maintain their massive profit margins. I don't know if that is going to work if 7 nm works out for AMD and they can offer 128 core / 256 thread dual socket systems for reasonable prices. The cost for a similarly powerful Intel system will be astronomical.

Epyc is just a compelling product though. The masive IO capabilities and massive core counts just can't be easily beat. Intel is still going to win some benchmarks due to lack of optimization and the different cache architectures. Intel's cache architecture will do better when there is a lot of fine grained shared memory / messaging between threads. It will burn a lot of power to do that though. Also, a lot of applications can be optimized to not do so much fine grained thread to thread communication. Some algorithms require it, but many do not. Optimizing such things will probably get better performance and power consumption on both types of systems. Such communication overhead limits scalability to many cores anyway.

A lot of companies will be looking at the features and realizing that they just can't do what they want to do on an Intel system due to limited IO or just cost. After meltdown, which is a massive hit in servers and virtualization environments, a lot of people are thinking that it would be good to diversify their systems such that a vulnerability from one vendor doesn't take out their entire service. Also, the people I know who confirure machines are looking at Intel's product line up and asking are the new cpus even worth it. The cost for a certain level of compute seems to have gone up, not down. Perhaps due to the larger number of memory channels, which aren't always needed. Some of them seem to just be sticking with broadwell based Xeons for the moment. Hopefully they will give AMD a try. I would like to see the end of the Intel monopoly.

August 8, 2018 | 11:57 AM - Posted by FabricResearchIsOngoingForRedsTeam (not verified)

Intel's higher core count CPUs using Intel's new fabric interconnect took a hit in latency also in order to be scalable to higher core counts. So the ring Bus has a little better latency but not much in that way of scaling for really high core counts.

I wish that AMD would try a Ring Bus topology on the Zeppelin Die and then go over the IF to other dies but maybe that's not going to be as good of a solution for scalability either, though it would help gaming workloads that where limited to running on a single die.

The best connection topology that AMD has expermented with is in that research paper(1) with die chiplets on active and passive interposers where the misaligned butter-donut Interconnect etched into a Active silicon interposer showed promice in the supercomputer cluster run Interconnect simulations the AMD researchers tested. In that paper the entire inconnect fabric traces and circuitry was etched on the Interposer's silicon for a NoC(Network-On-Chip)and that included routers. Then they simulated processor chiplets that where placed in a misalined fashon across the interposer/NoC and wired to the routers in a misalined fashon to discourage inter-chiplet communication traffc jams on the network.

Silicon interposers enable the integration of multiple
stacks of in-package memory to provide higher bandwidth
or lower energy for memory accesses. Once the
interposer has been paid for, there are new opportunities
to exploit the interposer. Recent work considers
using the routing resources of the interposer to
improve the network-on-chip’s (NoC) capabilities. In
this work, we exploit the interposer to “disintegrate” a
multi-core CPU chip into smaller chips that individually
and collectively cost less to manufacture than a single
large chip. However, this fragments the overall NoC,
which decreases performance as core-to-core messages
between chips must now route through the interposer.
We study the performance-cost trade-offs of interposer based,
multi-chip, multi-core systems and propose new
interposer NoC organizations to mitigate the performance
challenges while preserving the cost benefits." (1)

If AMD does increase is CCX Unit count from 4 CPU cores to say 6 or 8 on Zen2 then that's going to have some better latency affect on gaming workloads that use more than 4 cores/8 threads also as opposed to having more hops over the IF. Could AMD also desgn a consumer gaming oriented Zeppelin Die configuration that's ring Bus based with maybe IF for cross-die communication, that may be possible also.
There are so many ways that AMD could tweak the CCX unit's core counts and cache sizes to make for better overall gaming latency performance.

The optimization of gaming software for first generation Zen/Zen+ is still ongoing based around CCX Untis and Zeppelin Dies where Die/CCX affinity can be targeted in gaming software to keep latency to a minimum. For Zen2 having a larger L3 cache will be helpful also in addition to maybe larger L1/L2 cache sizes. Larger caches are the best way to avoid the high latency memory accesses or even cross CCX unit/cross die IF latency cache coherency traffic.


"Enabling Interposer-based Disintegration of Multi-core


August 8, 2018 | 04:29 AM - Posted by Pit Nicker (not verified)

Haha, oh wow!
Was impressed by the graph until I realized AMD's share varies between 0-3% and Intel's between 97-100%.

August 8, 2018 | 04:34 PM - Posted by WhenTheTAMisBigThemLittleOnePercentsAddUp (not verified)

AMD had around 23% server market share tops with its Opteron server CPUs over a decade ago and first generation Zen/Epyc is a whole closer to Intel's Xeon in IPC than Opteron and AMD will probably beat that 23% market share number within 2-3 years time.

So Intel's server market share direction is heading Down in the wrong direction and AMD with such a small share currently up from almost 0% to now 3%, has only one way to go and that's UP! So AMD is such a lean and low overhead operation(No Expensive Fab Upkeep costs or excessive middle manegement fat like Intel has) that they can squeeze profits out of Gross Margins in the 33-37% range while Intel will have to lower its markups to be able to attempt to keep AMD/Epyc from quickly taking back that 23% and probably higher server market share.

Intel's quarterly Gross Margins are above 60% and if they drop to below 55% then Intel is in deep share price didgeridoodoo and the Wallstreet Quants will break out the pooper scooper on ratings for Intel's Margin Basis Points.

AMD's CEO has stated that she expects around 5% Epyc/Server market share by the last reporting quarter of this year and low double digits by the end of 2019. And the server market has been in 2018 having some Quarters of high 20% to lower 30% year on year growth in 2018 and in 2017 server sales totaled around 65.15 billion dollars. So 3% of 2017's total server revenues is around 1.95 billion dollars with each 1% representing about 651.5 million dollars using 2017's numbers. So that can add up fast for AMD with each percentage point and 2018 looks to be about 25%-30% larger revenue wise for total yearly server systems sales.

So that's just standard server sales revenues and does not include other types of specilized HPC sales. Then there is the GPU compute and AI/Infrencing market where AMD has GPUs to offer like Nvidia offers and Intel is years off from competing with AMD and Nvidia in that GPU Compute/AI market. So using 2017's yearly server revenues if AMD got back to around 23% that would be around 15 billion dollars yearly of Epyc server sales, and 2018 is going to be much larger than 2017 in server market revenues.

And AMD's CEO statment, 5% by the last businees quarter of this year and low double digits by end of 2019, was made on that most recent quarterly earnings confrence call where SEC rules apply. So AMD's figures can not be too far off with the information that AMD has available internally for its management team. If the CEO states it then barring unforseen problems there must be purchase contracts waiting to be finilized that can not be reported until they fully close. Some Server customers have longer Certification/Vetting periods than others but loads of customers are interested in Epyc/Naples and also Epyc/Rome on 7nm on the Zen2 CPU micro-arch.

August 8, 2018 | 09:16 AM - Posted by Alfredo (not verified)

The graph is super misleading, dual scales, super narrow ranges...

August 8, 2018 | 10:33 PM - Posted by Dr_b_

The graph looks like a sketch of a fish, and if you look closely you can see that it came from "Wells Fargo Securities LLC" which we all know has been engaged in all manner of tom-foolery.

August 9, 2018 | 10:56 AM - Posted by Lance L (not verified)

Ollie Williams: "I SEE A FISH!"

Post new comment

The content of this field is kept private and will not be shown publicly.
  • Lines and paragraphs break automatically.
  • Allowed HTML tags: <a> <em> <strong> <cite> <code> <ul> <ol> <li> <dl> <dt> <dd> <blockquote><p><br>
  • Web page addresses and e-mail addresses turn into links automatically.

More information about formatting options

This question is for testing whether you are a human visitor and to prevent automated spam submissions.