Bulldozers Weak/Strong points?

[Phynaz]

Lol that comes off as highly ironic. IBM pursuing a licensing model that is causing you to migrate away from their hardware platform.

Lol, yes, that hasn't been lost on us.

But in reality, just like every other large corporation, IBM is a holding company. The software and hardware sides operate independently of each other.

 

[PreferLinux]

Do we get to include software prices in the system cost?

How many more Opteron cores do I need to get equivalent performance to the Xeon system? Let's say I have to license a few more cores of say Oracle, or SAP, or WebSphere or maybe even PeopleSoft on the Opteron. What does that do to the price equation? Over the lifespan of the system?

Or, let's say I'm building out a VDI infrastructure. How many more servers and software licenses do I need to service an equivalent number of users with the Opteron system compared to the Xeon?

That's the dirty little secret that doesn't seem to get much conversation.

How about we try with a LAMP setup?

 

[PreferLinux]

..

 

[Cerb]

How about we try with a LAMP setup?

Or Postgres, or Java, or...

Still, it would require remaking much of the software, which can be a bigger cost than swapping HW. As costs rise, over time, for non-COTS HW and SW, there will be a point where one may only sacrifice one or two upgrade cycle's worth of cost to swap over to not being tied to such a proprietary system.

 

[JFAMD]

IBM software is mainly sold on IBM systems, I'm not invited to many of those parties.

Oracle wants to sell oracle software on SPARC, but about a third of their licenses are site licenses.

We get this FUD all the time so I actually had my team go back and compare the total x86 server TAM against the software licensing models.

There is about 10% of the total x86 server TAM that is inaccessible based on thier licensing model.

That means when I get to 90% market share I will have a real issue. I will worry about it then.

 

[Phynaz]

John, you are backpedaling now. You went from "the majority of those are site licensed" to stating that you don't know about IBM and 30% of Oracle is site licensed.

And you accuse me of spreading FUD?

How about you answer my original questions:

How many more Opteron cores do I need to get equivalent performance to the Xeon system? Let's say I have to license a few more cores of say Oracle, or SAP, or WebSphere or maybe even PeopleSoft on the Opteron. What does that do to the price equation? Over the lifespan of the system?

Or, let's say I'm building out a VDI infrastructure. How many more servers and software licenses do I need to service an equivalent number of users with the Opteron system compared to the Xeon?

You keep bringing up the AMD cost advantage. So prove you have a significantly lower TCO over the life of the system. Include things like software licenses, hotswaps and parts spares, etc..

 

[JFAMD]

Database represents about 10% of the total server market TAM. MSFT owns ~55% of the total x86 database market. Oracle owns about 20-30% of the x86 TAM. IBM is ~10%.

So just looking at database alone, if you pull the IBM and oracle non-site license out the number you end up with ~20-25% of the database market not really accessible to AMD. so, if ~25% of database is not available and database is about 10% of the total market, then you are looking at Oracle and IBM database accounting for ~2.5% of the total x86 TAM.

I don't have my work computer open so I don't have the exact numbers in front of me. I will get you the numbers, but it is ~10% of the total TAM where there is licensing issues. While apps like SAP represent a large revenue opportunity in terms of total system price, the relevance to the actual CPU TAM is very small because they represent a very small number of CPUs relative to the total market opportunity.

 

[Phynaz]

You're still avoiding the questions John.

 

[Arachnotronic]

He's marketing...

 

[Lonbjerg]

He's marketing...

But, but....his signature says he isn't speaking for AMD...

 

[Idontcare]

You keep bringing up the AMD cost advantage. So prove you have a significantly lower TCO over the life of the system. Include things like software licenses, hotswaps and parts spares, etc..

I don't even remotely know what the numbers themselves shake out to be but it would seem by virtue of deduction that there is a good reason why AMD's server market-share is as low as it is.

While Intel gained 3% market share in 2010 (reaching 93%), AMD dropped 3% in market share to 7%.

http://seekingalpha.com/article/267774-amd-continues-to-challenge-intel-in-server-processor-market

Presumably there is something about AMD server gear that the market is finding undesirable in comparison to Intel's...be it performance, power-consumption, TCO, etc.

Otherwise the gear would be selling in a ratio that favored AMD a lot more and Intel a lot less.

 

[jimbo75]

Can you find anything? Or is it just a simple case of people being thick? You tell me why instead of asking why, lets see what you think the reason is based on all of the reviews you've seen. Then we'll be able to figure out if it really is people just being thick and buying the wrong option, because all of the numbers sure point towards that being what is happening. Unless of course you have your own numbers and own reasons?

Start with this article http://www.anandtech.com/show/3894/server-clash-dellr815/1

Now tell me why AMD is losing market share based on that.

 

[Idontcare]

Can you find anything? Or is it just a simple case of people being thick? You tell me why instead of asking why, lets see what you think the reason is based on all of the reviews you've seen. Then we'll be able to figure out if it really is people just being thick and buying the wrong option, because all of the numbers sure point towards that being what is happening. Unless of course you have your own numbers and own reasons?

Start with this article http://www.anandtech.com/show/3894/server-clash-dellr815/1

Now tell me why AMD is losing market share based on that.

Is your post directed towards me?

If it is then my answer is simple -> performance/TCO is what matters and the performance/TCO for AMD is apparently lagging Intel.

(TCO includes all the things that come with the compute environment, from budgets for salaries and personnel to expenses for electricity to expenses for software and rent for floorspace and so on...taken with respect to the revenue/profit that is enabled by that compute environment at the business)

Basically we have two positions we can take here - one position is that 93% of the market is full of idiots, idiots that AMD's server sales team still can't figure out how to manipulate as well as Intel's server sales team, OR Intel actually sells the stuff that business savvy people want to buy and AMD server sales team has done well enough to convince 7% of them to think otherwise.

Personally I like to give humanity a fair amount of credit when it comes to the people who are professionals in this industry because they know what they are doing in my experience, and I believe the numbers reflect the reality of TCO "down in the trenches".

The irony that is not lost on me is that when the performance/TCO tables were turned in AMD's favor that was all you heard about. Now that it appears to favor the Intel gear things get a little wonky in marketing land but the server numbers exist for some good reason.

If Intel's server sales team are simply that good and they've convinced 93% of the market to buy inferior server gear then I'd posit the notion that maybe AMD needs to hire themselves a few more of those Intel sales guys and get with the program...

 

[podspi]

You're still avoiding the questions John.

Phynaz, this isn't a sales call... I think he has made his point, which is that the majority of the market (which you clearly are not part of) isn't licensed on a core-basis. Since AMD's market share is relatively small, they still have a lot of room to grow before their relatively high core-count becomes a disadvantage (at least due to licensing concerns).


While I am not certain that the "many-cores, high-throughput" strategy is the best way to go, AMD is not the only company going this route. I've seen a few articles on super-dense Atom servers, and if we are to believe that ARM is going to be a force in the server market they are going to have core-counts much higher than similar throughput Bulldozer servers, and possibly higher core-counts than similar throughput Atom servers too. If this isn't a problem for Atom and ARM-based servers, it certainly isn't going to be a problem for Bulldozer-based servers.

 

[jimbo75]

Is your post directed towards me?

Not specifically, but it would be nice if somebody addressed it rather than asking the question.

If it is then my answer is simple -> performance/TCO is what matters and the performance/TCO for AMD is apparently lagging Intel.

(TCO includes all the things that come with the compute environment, from budgets for salaries and personnel to expenses for electricity to expenses for software and rent for floorspace and so on...taken with respect to the revenue/profit that is enabled by that compute environment at the business)

So based on this, you must believe MC is worse in performance/TCO than all previous AMD server cpu's? I don't recall AMD losing 1/3rd market share in a quarter before at least, so that has to mean that MC is AMD's worst ever server product. There can't be any other reason?

Basically we have two positions we can take here - one position is that 93% of the market is full of idiots, idiots that AMD's server sales team still can't figure out how to manipulate as well as Intel's server sales team, OR Intel actually sells the stuff that business savvy people want to buy and AMD server sales team has done well enough to convince 7% of them to think otherwise.

I say a lot of the market is full of idiots. Maybe not 93%, but it's too many. There is also another reason.

Personally I like to give humanity a fair amount of credit when it comes to the people who are professionals in this industry because they know what they are doing in my experience, and I believe the numbers reflect the reality of TCO "down in the trenches".

Did they just suddenly come into this whole "TCO" thing? I mean just around the moment MC was released? They must have, how else could AMD lose 33% of their market almost overnight?

If Intel's server sales team are simply that good and they've convinced 93% of the market to buy inferior server gear then I'd posit the notion that maybe AMD needs to hire themselves a few more of those Intel sales guys and get with the program...

Yeah I think AMD can't afford to buy the kind of "marketing" intel does so well. This is what annoys the hell out of me when I see the comments you and others make. JF might not be daft enough to come out and say it, but we all know what the real reason is.

At least you are semi-honest in realising it's nothing to do with the actual hardware, just realise what you mean when you are celebrating AMD's loss of market share with a superior product. That's right, you are celebrating ignorance and dodgy backroom deals over actual technology.

 

[krumme]

You're still avoiding the questions John.

No - you are avoiding his answer. Your question was very valid and important, and John answered so detailed he could and still keep his job. I was surprised he gave away so much information. I am pretty sure you understand the principle in the answer.

 

[RobertR1]

For virtualization, you get licensed off Cores but it depends on the workload on the VM's whether RAM or CPU is your limitation for needing a new chassis/blade. The truth is, if you're already on an Intel platform, then you stick with it. It just makes administration easier as you can migrate VM's on the same platform without any headahces. Frankly, I haven't came across a mixed VM environment where they are running AMD and Intel. I haven't come across any that are running AMD or even considering it.

AMD will have a major uphill battle coming into exiting VM enviroments ruled by Intel. They'll need to make their processors play nice with the various hypervisors making administration of a mixed enviornment seamless.

 

[JFAMD]

You're still avoiding the questions John.

I was not avoiding your question, I was riding my bike and having a life.

Here is your answer.

Total x86 processor TAM is ~6.5B

Cloud represents ~$560M, there are no licensing issues with cloud software

Network infrastructure and collaboration represents ~$3.351B, there is no licensing issues with network infrastructure and collaboration.

HPC represents ~$563M. There are some codes that are licensed by the core, however, nobody has an exact count. IDC estimates that 70% of HPC is custom code and only 30% of the HPC market is commercial code. Just to keep you from whining let's assign all 30% to the licensed by core category, even though it is probably 10% at most. That means ~$442M is addressable by AMD.

Then there is database/business apps which is ~$2.067B. IBM and SAP licese by core and they are ~9%. Oracle Enterprise is ~27%, but 1/3 of it is site license (VERY conservative figure), so we net out ~18% not being addressable, for a total of ~27% of the workload. That means ~$1.508B is addressable.

So, adding them all up you get about $5.8B out of $6.5B, or, as I said earlier, ~90%.

Which means that AMD is in a great position up to the point where we own ~90% of the market. So, when we get to 90% share your argument will be perfectly valid.

You can continue to throw up all of the examples that you want that AMD is disadvantaged on licensing, but it just isn't true. There will always be examples of individual deals where we are disadvantaged, but those are 1 in 10 at the most, making it the rare exception, not the rule.

 

[JFAMD]

For virtualization, you get licensed off Cores but it depends on the workload on the VM's whether RAM or CPU is your limitation for needing a new chassis/blade. The truth is, if you're already on an Intel platform, then you stick with it. It just makes administration easier as you can migrate VM's on the same platform without any headahces. Frankly, I haven't came across a mixed VM environment where they are running AMD and Intel. I haven't come across any that are running AMD or even considering it.

AMD will have a major uphill battle coming into exiting VM enviroments ruled by Intel. They'll need to make their processors play nice with the various hypervisors making administration of a mixed enviornment seamless.

1. Virtualization is not "licensed off cores", do you have some proof of this?

2. Administration happens at the platform level and HP to HP or Dell to Dell is the major consideration. You *generally* cannot do live migration between Intel and AMD because the processor architectures are different, however:

A. Most customers don't migrate VMs. They treat VMs like servers, they want to touch them when they start them up for the first time and then when they take them out of service.

B. When they do actually migrate VMs they tend to schedule that work for off hours when nobody is there. You can suspend a VM and move it from one server to another and turn it on relatively quickly. When you do a live migration it can take minutes or hours depending on network traffic. To do a live migration you need shared, pooled storage as well.

C. Most people have "virtualization clusters", one for accounting, one for email, one for engineering, etc. If they are going to migrate VMs they can do it amongst the clusters, but you typically don't see the engineering guys or the accounting guys wanting to share. In this world, having a pool of Intel of engineering and a pool of AMD for accounting is perfectly reasonable.

 

[jones377]

John, thanks for the numbers. So why does AMD have such a small marketshare in servers then? Are Intel servers simply considered good enough that no alternative is even considered? You dumped the prices on Magny-Cours compared to previous gen with the only result being falling server CPU ASP and barely maintaining the low market-share.

It seems to me that sometimes it's not enough for AMD to have a competitive product out, Intel needs to stumble as well, as seen in the Netburst era. Not only was netburst a pretty horrible server product from the perf/w standpoint, it was also hobbled with a crappy system architecture not suited for multi socket servers. But since then they have adressed both of these issues and pretty much taken back all lost marketshare in addition to stealing marketshare from the RISC vendors.

Will a superior microarchitecture from AMD really be able to turn this around? I don't see Intel doing another netburst anytime soon....

 

[Arachnotronic]

You know, I feel like AMD can get away with "more cores" or whatever per TDP because their caches are TINY compared to what Intel puts on their chips.

 

[classy]

For virtualization, you get licensed off Cores but it depends on the workload on the VM's whether RAM or CPU is your limitation for needing a new chassis/blade. The truth is, if you're already on an Intel platform, then you stick with it. It just makes administration easier as you can migrate VM's on the same platform without any headahces. Frankly, I haven't came across a mixed VM environment where they are running AMD and Intel. I haven't come across any that are running AMD or even considering it.

AMD will have a major uphill battle coming into exiting VM enviroments ruled by Intel. They'll need to make their processors play nice with the various hypervisors making administration of a mixed enviornment seamless.

I am going to assume you meant per processor. VMWare Enterprise Vsphere licenses per processor up to 6 cores.

Sticking with the same platform is correct advice.

But the AMD statements are not true. In my enviroment right now I run a 3 host cluster using Dell 905s with optys. I have managed both Intel and AMD virtualization platforms and both work with zero issues. If someone says AMD sucks, then its the admin who sucks, certainly not the platform. I attend VMUGs and virtual conferences when I can and I have spoken with many who use AMD hosts.

Intel is used significantly more, but the notion AMD is rarely used or not considered is just not true.

 

[Dresdenboy]

So based on this, you must believe MC is worse in performance/TCO than all previous AMD server cpu's? I don't recall AMD losing 1/3rd market share in a quarter before at least, so that has to mean that MC is AMD's worst ever server product. There can't be any other reason?

Yes, it can. (sth like that)

Now to your point:
Istanbul and later Lisbon were still available besides MC. So there was no need to buy a MCM 8- or 12-core processor. Imagine there wouldn't have been MC, only 6-cores as before.

Would no drop have happened then? If not, why did it happen with additional options?

And if a drop would have happened then (not more than 6 cores CPUs offered), could there have been external reasons causing that drop?

 

[krumme]

1. Virtualization is not "licensed off cores", do you have some proof of this?

2. Administration happens at the platform level and HP to HP or Dell to Dell is the major consideration. You *generally* cannot do live migration between Intel and AMD because the processor architectures are different, however:

A. Most customers don't migrate VMs. They treat VMs like servers, they want to touch them when they start them up for the first time and then when they take them out of service.

B. When they do actually migrate VMs they tend to schedule that work for off hours when nobody is there. You can suspend a VM and move it from one server to another and turn it on relatively quickly. When you do a live migration it can take minutes or hours depending on network traffic. To do a live migration you need shared, pooled storage as well.

C. Most people have "virtualization clusters", one for accounting, one for email, one for engineering, etc. If they are going to migrate VMs they can do it amongst the clusters, but you typically don't see the engineering guys or the accounting guys wanting to share. In this world, having a pool of Intel of engineering and a pool of AMD for accounting is perfectly reasonable.

Now we know that MC is not held back by license per core. Well then why is the AMD market share like it is?

How is BD going to be a better proposition?

 

[podspi]

Now we know that MC is not held back by license per core. Well then why is the AMD market share like it is?

How is BD going to be a better proposition?

I can't speak with any authority (since I haven't seen one of course) but I believe BD brings to the table:

1) More performance/watt (always a good thing)

2) Higher singlethread performance (throughput isn't everything)


Also, I think the whole idea of super-dense servers have just started to take off, and imho AMD is better positioned to take advantage of this than ARM is -- though ARM certainly looks like it is going to make a serious effort.