Why is AMD losing the next gen race to Nvidia?

yep at the end we won't see the server market as a upswing for AMD for probably a year and that is if they are competitive price and performance. This was probably the same reason we didn't see them make much market penetration back with the original Opterons, that and Intel's snide tactics. And it was about a full year before their marketshare started going up.

Intel knows this, they know if AMD can get a foot hold in the server market its much harder to get any lost marketshare back, its going to be a hard fought battle and they are not going to just give it up if their products are just as capable.
OTOH as itsmydamnation said AMD has a good plan in place.
 
best made plans of mice of men.....

Plans means very little when you have Goliath on one side. AMD is still tenuous on the financial side of things which Intel can exploit if need be.

simple analogy, when a company like Intel is entrenched, AMD needs a plan like blitzkrieg, to get them out. They had that with Opteron and it still took one year for them to start showing the results. If they are going in with the same firepower as what Intel has (which is still to be determined, most likely not), its going to be a long drawn out battle where its going to be very tough for AMD to capitalize and gain marketshare.
 
Why? Becuase you say it is? Commodity hardware is commodity hardware. While im sure you can find orgs that still have the wont get fired for buying IBM/Cisco/etc mantra, its disappearing fast. Its all Fast devops/rapid system development lifecycles, infrastructure mangers care less and less about the actual hardware every day ( see white box switches/routers/gluster/etc) and more and more about API's and scalable management ( its why im such a big fanboy of Cisco UCS and look how quick they grew that market share).

The drive to the cloud also takes spending away from what i will call lower quality people ( you know the technical person who got promoted to manager and still stuck in two decades ago) and into large high performance infrastructure operators. You think they will buy intel cuz intel?

Meanwhile you ignore time and time again each time i point it out, a server is a platform, not just single core max perf and even from day dot AMD will have advantages vs intel in the platform, advantages they didn't have with agenna/istandbul/mangy-cours/bulldozer. If they have on SOC hardware crypto/compress engines like some of the rumors say ( like Seattle) then they have a major advantage against Intel with the big web infrastructure shops ( would take time to see support appear in "enterprise" software).

I.T. Infrastructure is littered with the caucuses of hardware companies who thought like you do while their base turned to commodity, many of those caucuses are thax to intel themselves. You really think the people responsible for road map and paid on performance at intel think the way you do?


So when are you going to go back through all the post i took the time to write in response to you and actually answer and of the questions i asked of you?
 
Why? Becuase you say it is? Commodity hardware is commodity hardware. While im sure you can find orgs that still have the wont get fired for buying IBM/Cisco/etc mantra, its disappearing fast. Its all Fast devops/rapid system development lifecycles, infrastructure mangers care less and less about the actual hardware every day ( see white box switches/routers/gluster/etc) and more and more about API's and scalable management ( its why im such a big fanboy of Cisco UCS and look how quick they grew that market share).

The drive to the cloud also takes spending away from what i will call lower quality people ( you know the technical person who got promoted to manager and still stuck in two decades ago) and into large high performance infrastructure operators. You think they will buy intel cuz intel?

Meanwhile you ignore time and time again each time i point it out, a server is a platform, not just single core max perf and even from day dot AMD will have advantages vs intel in the platform, advantages they didn't have with agenna/istandbul/mangy-cours/bulldozer. If they have on SOC hardware crypto/compress engines like some of the rumors say ( like Seattle) then they have a major advantage against Intel with the big web infrastructure shops ( would take time to see support appear in "enterprise" software).

I.T. Infrastructure is littered with the caucuses of hardware companies who thought like you do while their base turned to commodity, many of those caucuses are thax to intel themselves. You really think the people responsible for road map and paid on performance at intel think the way you do?


So when are you going to go back through all the post i took the time to write in response to you and actually answer and of the questions i asked of you?


Look you got a direct answer for a guy that does this for a living. And he isn't in middle management he is one step away from the CTO of the biggest company in its industry.

You are thinking of things at a micro level, where a corporation of this size, infrastructure is a just a portion of cost of total operations. Also they change slowly, they don't move fast just because a new technology comes out and it is a must have (and in this case it might not even be a must have, its ok, they have something that competes lets see what we can do with it). They have their IT protocols that must be followed. And most companies are sticklers about their protocols.

I'll give you an example, more than half of Cargrill's stand alone computers for scan guns, still run 32 bit Windows XP! They still need serial ports!, they still need parallel ports! Its not they don't want to change, its because the other companies they do work with haven't changed yet!

And we don't know about those rumors, great it might have some cool encryption things, but that's all up in the air. Things like that can help operations yes if they have them, but things like that companies might not even be able to use yet because of IT protocols.

Plus I even stated desktop, (which should include laptop) is the best place for Zen, it will make ground there, Just not in servers.

If I go back through your post and answer your questions, which if you go back through mine most of them are answered already in this very thread what will you say?

Look I work at NBC, I'm a Sr. Producer, two steps away from upper management, my computer still runs Windows 7 where some of my projects I need Windows 10, i can't get it natively installed on my office computer, because Windows 10 is not allowed on the network, I have to do it through VMware on my office computer or which I did recently was ask for a second computer that is not attached to the network.

Stupid things like this, is what company protocols stop you from doing, if this is just a satellite system, imagine what the protocols for servers are at NBC Comcast! And NBC has one of the most progressive IT departments I have ever worked with, they try to push new technology as quickly as possible to get the upper hand in everything they do. If you have ever seen a large scale company's IT protocol manuals, damn those things are large, might be the size of an entire set of Encyclopedia Britannica's. *might not get this joke if you are younger then 25 years of age ;)

The amount of company facing VS's we have, are limited, most of them are just staging servers. All of our consumer facing servers, are the real deal, we don't want to split resources up for them, and you know what the reason for that is?

2012 Superbowl we migrated many of our consumer facing servers to VS's about 1 year or so before this. It brought down the entire network in the first 5 minutes of the super bowl.

For every minute our servers were down we were losing 10 million bucks! Then add in the fact they had to call in pretty much their entire IT team in and pay them 1.5 times to 3 times the amount they normally get paid *world wide mind you. They got it up and going in hour or so, but to fix problem took 9 months to a year after that.

Just because it looks good on paper doesn't mean in the real world it works out as planned. In theory it was a great move for the IT department to go to VS's simplified their work a great deal, and cut down costs, but in reality it create a situation that could have been catastrophic if we couldn't get things up and going before half time of the SB. You think our ad partners were happy? We had to make them happy by cutting our costs down and taking a hit.The repercussions weren't just limited to just us, not only that the repercussion were long felt through out another 2 years.

My previous job I worked at Credit Suisse, have you seen any type of financial models and what their needs are from a server point of view?

You think they want to go to the cloud with things like that?

1 second delay in processing a financial model for their traders can cost them hundreds of millions of dollars. Even a .5 second delay would do it.

Do you know where most financial companies house their servers? Just look for the Atlantic trunk lines, most of the biggest trading company's servers will be right there. They will pay anything for performance in every aspect. All they want to get an few microseconds advantage over their competition.

And again their software most of them are built on Fortran and Cobol. They can't switch over to new things as quickly as you would like.

This is why I say the cost for these types of corporations, is inconsequential in the bigger scheme of things.

We are talking about multi billion dollar companies, not your mom and pop needs a website stuff.
 
Last edited:
no he didn't pretty much stated the opposite of what itsmydamnation, even itsmydamnation stated my friend who apparently to him is just a manager would be a carcass.

( you know the technical person who got promoted to manager and still stuck in two decades ago)

WTF is that, he calls people trolls just because they don't agree with him. The world doesn't run on ideology of new tech is better. New tech is better if it can be used more productively than old tech without any headaches,and that is it.

Then he brings in rumors that we don't know if they have those things in their chips.....

Performance doesn't matter for him right, for him it doesn't matter, not for everyone else. Anything that is time sensitive vs money, performance in every metric matters. Why do you think the 1 peta flop race is even there? What performance doesn't matter? Of course those needs are different than most company needs. But its an out liner.

How many company's do you know would even want to put their corporate documents and files on the cloud? I can't think of one company I have worked for that would want to house their sensitive materials outside of their company.

You think all those government servers are going to be just updated to the next tech? Tell ya what most of their databases are also based on Fortran and Cobol as well. Front end is updated though. Government is very slow to change.


AMD_server_share_Sterne.GIF


Now see that Opetron server Launch, 2004, this is when they had such an advantage that could not be denied. How do they expect to do the same thing when their tech and Intel's are pretty much the same now, if it is the same. Even getting 1% will be difficult. Opeteron's gained 2.5% once launched and stayed that way for a full year before they started making traction (most likely once Intel contracts started running out), this time that isn't likely going to happen, company's will just resign with Intel if everything is close.

You also have to look at the OEM and 3rd party sellers that got a backlash after barcelona and its problems, are they willing to take that chance and go with AMD again, when Intel is still there and just as performant?
 
Last edited:
It seems you insist Zen is doomed entirely. IMD gives you an example of a market where that likely is not the case. He never said anything about other markets. He asked you do solidify your aguments (that Zen is still-born for the whole planet). I think you have to except that there are markets which suit Zen well, and vice versa, and they are not small/niche.
Nobody claimed Zen is the super-cure and changes the future, or traps intel or whatever power-fantasies can be had. It seems to have good chance so far, to be something which is bought.
 
I see Zen picking up in the desktop and notebook sales, I don't see it in servers, at least not in the short term. I have stated this right from the start (desktops) of his first post on the subject.

I think they will gain marketshare fast on the consumer side of things, around 2.5% per quarter if performance is relatively the same, till they reach around 35% which things will start slowing down after that.

Just have to look at what most businesses run with, which are HP, Dell, and Lenovo, their contracts extend to a specific OEM, those OEM can start introducing Zen CPU's which business can pick up within the same contract year.

This isn't about the end user or enthusiast, its about mass volume.

I remember AMD stating getting server marketshare is hard, when Opteron first came out, getting 1% is a lot of money but its very hard to do that. Its highly competitive, much more so then desktop markets because of long term contracts, loyalty, trust. Which the last two AMD screwed up with Barcelona. These things don't change just because AMD has a new product.
 
Last edited:
I don't have time for this. @itsmydamnation said that AMD could bring Zen cheaper and possibly either with more cores or more memory, @Razor1 said that performance would still-born Zen, but @Razor1 "friend" directly contradicted him saying that performance wasn't relevant, that they can just add more racks (if cheaper then more racks, if +25% cores then more density) and that more memory would be welcome (if +25% memory then one happy friend).

Contractual obligations are expected and AMD has an uphill battle, but don't expect to win a nobel prize for "discovering" that.
 
I don't have time for this. @itsmydamnation said that AMD could bring Zen cheaper and possibly either with more cores or more memory, @Razor1 said that performance would still-born Zen, but @Razor1 "friend" directly contradicted him saying that performance wasn't relevant, that they can just add more racks (if cheaper then more racks, if +25% cores then more density) and that more memory would be welcome (if +25% memory then one happy friend).

Contractual obligations are expected and AMD has an uphill battle, but don't expect to win a nobel prize for "discovering" that.


I don't expect any award out of this, because I have to state the obvious after being called a troll. Which I conveniently ignored, till his last rant which looked down and me and my friend. WTF seriously have to pull up google images to show how Opteron's marketshare was like to a person that is in the server industry? It should be burned into their memory, its their job to know those types of things.

Hyperscale tech in not for everyone yet, there will come a point when most companies start migrating over to them yes, but not yet.
 
Last edited:
WTF seriously have to pull up google images to show how Opteron's marketshare was like to a person that is in the server industry? It should be burned into their memory, its their job to know those types of things.

Though to be fair it is also their job to present the facade that it will be different this time, at least in public.
 
..till his last rant which looked down and me and my friend.

I couldn't see any way in which he passed judgement on your friend, and I'd suggest maybe not steering the conversation in that direction.

There have been some really interesting points raised about competitiveness relating to per-socket costs, and memory bandwidth and capacity costs. These seem to be really good points that I hadn't considered. Like most people, I'd been focusing on per-core outright performance, and with a secondary consideration of performance per watt.

Do you have any counterpoints regarding these issues? I now think it's an area AMD can be competitive in, and am almost daring to feel hopeful that AMD might have an advantage in an important area of the server market.
 
That's not the way I read this

( you know the technical person who got promoted to manager and still stuck in two decades ago)

Sorry to say it but he has his head stuck up his ass. If his company was to trying to get a deal from any major corporation and talked like this to the next possible CTO of that company they would go tell him to F off, they are a dime a dozen.

Well we just don't know yet from what AMD will have with Zen, where will it be at in per core outright performance, everything points to something around Haswell level, although what AMD has shown so far its at Broadwell with their blender presenation.

We don't know what the stability of AMD platforms are like yet either, taking about ram, we don't know the access times of their platforms.

Intel has been pushing more into networking, storage and other parts of server infrastructure, which AMD hasn't been, Intel has more advantages than just the CPU. Intel is very well aware they are going to be seeing competition on the server side, and they started diversifying years ago because they saw this battle coming, not just from AMD, but also ARM. Just have to look at how many companies have tried and failed to go up against Intel in the server business. Qualcomm is there but I think its going to be years before they will be able to do anything productive. I think the only company with ARM based tech that has a chance right now is IBM with their open cloud services, which are joint with other large corporations. They started this 2 years back, and still no inroads. If its taking IBM this long and they still haven't made any market penetration, how can we expect AMD, a company that has burned bridges before to start gaining rapid marketshare (rapid is like 2.5% per quarter), like they did in the past.

Power usage I expect to be comparative.
 
Last edited:
But the bright side for AMD and everybody else is that Intel is not a fast moving target anymore since process shrinks are coming to an end, and they were Intel's main advantage.
 
Intel still has a sizable lead in process technology 2 years maybe more (if we look at the time Intel as able to make complex CPU's on 14nm vs when Samsung, GF were ready with their version of 14nm, its still a large disparity). But what Zen will really provide in the short term is, the increase of margins, and a small uptake of marketshare gains in the desktop/notebook side, their bottom line will increase dramatically, 10-15% increase in margins and 5% increase in that specific marketshare, its going to be around ~250 mill per quarter. If Zen is capable of doing this and it should if its performance is around Haswell, the base troubles AMD has been having for the past decade will be over.

R&D expenditure will increase again, marketing efforts will too. AMD will be able to start planning on how to diversify their portfolio to better take on Intel and nV. Of course Intel and nV aren't sitting idle so things will change on their side too
 
Last edited:
I couldn't see any way in which he passed judgement on your friend, and I'd suggest maybe not steering the conversation in that direction.

There have been some really interesting points raised about competitiveness relating to per-socket costs, and memory bandwidth and capacity costs. These seem to be really good points that I hadn't considered. Like most people, I'd been focusing on per-core outright performance, and with a secondary consideration of performance per watt.

Do you have any counterpoints regarding these issues? I now think it's an area AMD can be competitive in, and am almost daring to feel hopeful that AMD might have an advantage in an important area of the server market.
Well one on memory and that is Skylake Xeon is a change to earlier Intel models in terms of memory architecture, such as 6-channel DDR4 that then reduces cost in memory from a context raised earlier, along with the benefits of improved memory bandwidth of Skylake Xeon over Broadwell and earlier, and also now having options such as omni-path fabric.

Xeon Skylake is scheduled for 1H 2017 and engineering samples are now out, so Zen from a datacenter perspective will be more compared to Skylake Xeons rather than Broadwell, and the challenge for AMD is how quickly they can release high core count Zen processors as the longer they take the more it will be weighed against Skylake Xeon.
The desktop Zen models now look to be available Feb 2017 according to Benchlife, it would be a surprise if they also manage to launch the datacenter high core processors (say 20+) at the same time, just look at how it is staggered with Intel.
So I do think Zen server processors will be competing against the revised Skylake Xeons rather than Broadwell Xeon when businesses are considering their projects.

Cheers
 
I'd imagine there's another technology or capability AMD is leveraging if they expected to pick up marketshare. That China JV would seem to indicate there is some interest there. It may simply be AMD is ahead of Intel on stacked memory technologies or are toying with a socketed GPU for a coprocessor. Along the lines of Intel's Knights Landing. All the chip guys seem to think there is a desirable market there. While a GPU might not be ideal for VMs in most cases, data processing and some databases might be able to heavily leverage that advantage. A basic Zen might just be a bunch of processing cores, but so far most seem to be avoiding possibilities of an interconnect. 32 cores plus a large Vega in one or more other sockets may outstrip Intel's offerings. It stands to reason AMD has a leg up on Intel in regards to graphics or more parallel processors. If that weren't the case they'd be making larger GPUs. The SSG might be another interesting example of what AMD is targeting. So IPC of just the processor might be a bit limited considering the bigger picture.

As for those VMs. All that cloud space is a lot more than just web servers. Managed apps for small businesses really do seem to be a better solution for most. No local servers, complex networks, security, etc that needs maintained that smaller entities likely don't have the personnel to manage appropriately. That approach also changes what kind of servers are being sold.
 
I'd imagine there's another technology or capability AMD is leveraging if they expected to pick up marketshare. That China JV would seem to indicate there is some interest there. It may simply be AMD is ahead of Intel on stacked memory technologies or are toying with a socketed GPU for a coprocessor. Along the lines of Intel's Knights Landing. All the chip guys seem to think there is a desirable market there. While a GPU might not be ideal for VMs in most cases, data processing and some databases might be able to heavily leverage that advantage. A basic Zen might just be a bunch of processing cores, but so far most seem to be avoiding possibilities of an interconnect. 32 cores plus a large Vega in one or more other sockets may outstrip Intel's offerings. It stands to reason AMD has a leg up on Intel in regards to graphics or more parallel processors. If that weren't the case they'd be making larger GPUs. The SSG might be another interesting example of what AMD is targeting. So IPC of just the processor might be a bit limited considering the bigger picture.

As for those VMs. All that cloud space is a lot more than just web servers. Managed apps for small businesses really do seem to be a better solution for most. No local servers, complex networks, security, etc that needs maintained that smaller entities likely don't have the personnel to manage appropriately. That approach also changes what kind of servers are being sold.
Then that is competing more against Knights Landing rather than traditional Xeons, and KNL has had a big improvement over previous generations.
I was wondering the same myself, but I am not sure how well AMD will do in that space against Intel or IBM-Nvidia/Cray-Nvidia, all with their own memory/interconnect/fabric solutions.

Cheers
 
Intel is having a tough time against nV in that market and AMD is 5 years late, I don't see them having any traction there specially without any software backing them up.
 
@Razor1 , i dont have time to reply, but i will just say that comment wasn't directed at you or your friend, its an actual trait that is rampant in the Infrastructure side of companies. The point was the mantra of all ways buy X is getting washed away in the software defined world. Things are being designed and expected to be able to scale horizontally and vertically almost instantaneously and that is forcing change in the infrastructure space.
 
Back
Top