Would you buy an Opteron SMP board for enthuthiasts?

[JF-AMD]

I am no expert in the client space.

So, I actually went to do a little research for everyone in order to become more of an expert in the client space. The discussion would be better served with facts over feelings. It was actually worse than I was portraying it.

Xeon, as a desktop brand, represented ~1.3% of the overall client market in 2007. Today, Xeon as a brand, represents ~.3%, less than 1%, (and probably far less than people have been giving it credit.) That means the market for 2P desktops is merely a quarter of what it once was just 2.5 years ago. A rapidly shrinking market is not one to invest in, you want to invest in a growing market.

Now, what is so different about 2007 and today? In 2007, dual cores ruled the market. There were a lot of people that wanted 4 threads, and now they get that in a single chip. As you hit thread saturation in a single socket, there is less demand for that second chip. And, as we are at 6 today, and 8 next year, the need for that second socket, as a market aggregate, is continuing to shrink.

I also did the quick math on what it would take to bring a dual socket part to market, assuming that you take the exact same server design, call it anything else and sell it. You would need to sell about 2X what the current xeon sales desktop sales are (according to IDC) in order to make that a profitable product. That assumes $0 in advertising by the way, adding advertising on the top only makes the situation worse.

Heap on the fact that to bring the product to market you are probably assuming ~6-12 months to hit the market window from a productization standpoint, adding a new 2P platform would mean that your launch slot would be in 2011, which is right on top of Bulldozer. If the market continues to shrink (because of thread saturation), then the 8 cores of Bulldozer will accelerate that trend.

So I would need to be at 2X the market in sales, but the market is shrinking fast, and will probably accelerate that trend when we get more cores.

I am not trying to say that people aren't justified in wanting this product, just that there is no way to make the numbers work. I agree that for some it would be a very cool thing to have; I just wish people wouldn't say that it is so easy to do or that there is a huge market waiting to be tapped. Neither of those statements are true. If you had access to cost models and market share data all of this would be crystal clear. Unfortunately I can't share that, but I believe I have already shared a lot to hopefully show this in a different light.

[Deluded]

I am no expert in the client space.

So, I actually went to do a little research for everyone in order to become more of an expert in the client space. The discussion would be better served with facts over feelings. It was actually worse than I was portraying it.

Xeon, as a desktop brand, represented ~1.3% of the overall client market in 2007. Today, Xeon as a brand, represents ~.3%, less than 1%, (and probably far less than people have been giving it credit.) That means the market for 2P desktops is merely a quarter of what it once was just 2.5 years ago. A rapidly shrinking market is not one to invest in, you want to invest in a growing market.

Now, what is so different about 2007 and today? In 2007, dual cores ruled the market. There were a lot of people that wanted 4 threads, and now they get that in a single chip. As you hit thread saturation in a single socket, there is less demand for that second chip. And, as we are at 6 today, and 8 next year, the need for that second socket, as a market aggregate, is continuing to shrink.

I also did the quick math on what it would take to bring a dual socket part to market, assuming that you take the exact same server design, call it anything else and sell it. You would need to sell about 2X what the current xeon sales desktop sales are (according to IDC) in order to make that a profitable product. That assumes $0 in advertising by the way, adding advertising on the top only makes the situation worse.

Heap on the fact that to bring the product to market you are probably assuming ~6-12 months to hit the market window from a productization standpoint, adding a new 2P platform would mean that your launch slot would be in 2011, which is right on top of Bulldozer. If the market continues to shrink (because of thread saturation), then the 8 cores of Bulldozer will accelerate that trend.

So I would need to be at 2X the market in sales, but the market is shrinking fast, and will probably accelerate that trend when we get more cores.

I am not trying to say that people aren't justified in wanting this product, just that there is no way to make the numbers work. I agree that for some it would be a very cool thing to have; I just wish people wouldn't say that it is so easy to do or that there is a huge market waiting to be tapped. Neither of those statements are true. If you had access to cost models and market share data all of this would be crystal clear. Unfortunately I can't share that, but I believe I have already shared a lot to hopefully show this in a different light.

Hmm... thanks for sharing this with us. I apologize for my previous statement. I can see from your standpoint and how it would be a unwise choice given the circumstance...

I can see that your (AMD)'s mind is quite made up on this topic, but I'd like to ask this one last request to see if it's even possible: call Asus or tyan or supermicro or whoever else, and ask them to release a beta BIOS on their existing motherboards that allow adjustments that we seek. Perhaps you'd need to release that beta BIOS to a very small group of people backed by whatever contract that legally binds them if that particular BIOS becomes public, therefore reducing the chance for a beta-BIOS getting into the hands of some idiot server IT admin somewhere and enabling AMD to lock horns with Intel once more.

I mean, it's a thought. If you or AMD can't do it because of financial cost, then no biggie. You guys released TWKR chips to support the overclocking community, and I hope you haven't forgotten about us.

[zanzabar]

or since we have some amd rep in thread action, how about phenoms that are clocked on the NB to start with and maybe a higher HTT like 222 that would keep the HTT buss at 2ghz but u would have an advantage on ram and with a higher NB like x12 that would give u 2662 and it would be much faster on the IO and for gaming/windows start. i know that i would not sacrifice the multi on my own chips unless it was 14x or higher as of now but for people who cannot overclock on OEM systems it would be a huge boost

[BeepBeep2]

If by an incredible stroke of luck something should happen because of this innocent thread then it should be called the Phenom XS edition

XS Logo engraved to the IHS


...

Anyway, what about a 12 core TWKR chip with no funds behind it? An "AS IS" processor? ...with board manufacturers doing whatever they want to adapt it?

Looking at all the people that bought Gulftown...

[blindbox]

I am no expert in the client space.

So, I actually went to do a little research for everyone in order to become more of an expert in the client space. The discussion would be better served with facts over feelings. It was actually worse than I was portraying it.

Xeon, as a desktop brand, represented ~1.3% of the overall client market in 2007. Today, Xeon as a brand, represents ~.3%, less than 1%, (and probably far less than people have been giving it credit.) That means the market for 2P desktops is merely a quarter of what it once was just 2.5 years ago. A rapidly shrinking market is not one to invest in, you want to invest in a growing market.

Now, what is so different about 2007 and today? In 2007, dual cores ruled the market. There were a lot of people that wanted 4 threads, and now they get that in a single chip. As you hit thread saturation in a single socket, there is less demand for that second chip. And, as we are at 6 today, and 8 next year, the need for that second socket, as a market aggregate, is continuing to shrink.

I also did the quick math on what it would take to bring a dual socket part to market, assuming that you take the exact same server design, call it anything else and sell it. You would need to sell about 2X what the current xeon sales desktop sales are (according to IDC) in order to make that a profitable product. That assumes $0 in advertising by the way, adding advertising on the top only makes the situation worse.

Heap on the fact that to bring the product to market you are probably assuming ~6-12 months to hit the market window from a productization standpoint, adding a new 2P platform would mean that your launch slot would be in 2011, which is right on top of Bulldozer. If the market continues to shrink (because of thread saturation), then the 8 cores of Bulldozer will accelerate that trend.

So I would need to be at 2X the market in sales, but the market is shrinking fast, and will probably accelerate that trend when we get more cores.

I am not trying to say that people aren't justified in wanting this product, just that there is no way to make the numbers work. I agree that for some it would be a very cool thing to have; I just wish people wouldn't say that it is so easy to do or that there is a huge market waiting to be tapped. Neither of those statements are true. If you had access to cost models and market share data all of this would be crystal clear. Unfortunately I can't share that, but I believe I have already shared a lot to hopefully show this in a different light.

I actually understand you from the start, but me, like everyone, can't resist seeing these babies overclocked

[JF-AMD]

I can see that your (AMD)'s mind is quite made up on this topic, but I'd like to ask this one last request to see if it's even possible: call Asus or tyan or supermicro or whoever else, and ask them to release a beta

Now I see where you got your name from

They would not do that work unless we guaranteed some degree of overclockability. Which means we incur all the cost anyway.

And only Opteron's mind is made up. I am the server guy, I am only concerned about Opteron.

[snoro]

I see your point JF-amd but there is a little group over here called the wcg team and some of us want all the core and speed we can have and we know ther is a little bit more left in the opteron chip for wcg. The only thing us are asking to say to tyan or supermicro ( i would never trust a asus server motherboard) that they gan hand out some special bios but with a drawback, it can be a sort of nda contract or simply a break of the warranty of the motherboard we have.Heck we would even be ready to pay a little bit more(maybe 50$ just for that special bios). We dont need any proof or validation that it will work we just want the option to do it. It can be hidden from the public better than a cia secret but we dont mind, We just want to have the option to be able to buy also from amd and not only from intel because they are at the moment the one with most ppd producing chip with the sr-2 and the xeon 56xx series. We dont mind about high initial cost because that was already planned but high electricity usage is something we dont want so a cluster of alot of desktop chip ca sometime be easily beaten a one mp rig because of the lower power usage.Thats why we are interested in mp and dp motherboard.So please, JF-Amd, dont leave us being obliged to buy only from intel just because you think we are not worth it. Remember a happy cruncher with a amd mp rig will also be more tempted to go for amd after when he need a gaming rig or his friends are looking for a new computer and ask the cruncher what they should get. Just look at the donation we had from gigabyte. it was a very nice move from them(especially dino22) helping us and by doing so, every cruncher of our team will now remember gigabyte when they have to buy a new motherboard for themself or for a friend.

[Deluded]

Now I see where you got your name from

They would not do that work unless we guaranteed some degree of overclockability. Which means we incur all the cost anyway.

And only Opteron's mind is made up. I am the server guy, I am only concerned about Opteron.

Easily done. Just point them to s7 post -- he overclocked two mangy cours 1.3Ghz above stock on stock vcore. I don't know about you, but to us, that is DAMN impressive! 24 cores @ 3Ghz, up from 1.7Ghz stock....

I can tell you, that taking an Opteron and rebranding it as phenom would be costly. You are basically doubling the back end validation, qualification and dev resources. Not design resources, but pretty much everything else (there is not nearly as much leverage as you would guess.)

I understand if you can't share anymore data/info, but I'm not getting it (somewhat) from this post... You're telling us that it's not so easy as just slapping on a IHS with a different name and some die marking that would make it identify it to CPU-Z as a xxx processor rather than a Opty? For something as simple as rebranding you have to go through a whole cycle of testing and development?

[zir_blazer]

Now I see where you got your name from

They would not do that work unless we guaranteed some degree of overclockability. Which means we incur all the cost anyway.

And only Opteron's mind is made up. I am the server guy, I am only concerned about Opteron.

The first thing that I think about is that you are the wrong guy to ask these topics about. As overclocking the Processor is dependant on the Motherboard and BIOS more than the Processor itself (Locked Multiplier? Who cares. Push the Base Clock), it must be an independent Motherboard vendor that decides to provide tweaking tools for enthusiast in some models without AMD support. If enthusiast buy Opterons to overclock them is not something that you should be worried about, as you can't actually control what your parts will be used for, but attempt to sell them at a determined market. I doubt that the AMD Desktop guys likes that some dirt cheap Sempron gets tons of overclocking punishment from an enthusiast instead of them buying a much more expensive enthusiast class Athlon 64 FX, in the same way that there could be many small server guys that buy Desktop class Hardware for use as Desktop-as-Server (Didn't Google assembled clusters of Desktop machines?) and you can't prohibite them to do so. Though is something that I know that AMD is aware of and that is why Opterons 1xx migrated to S939, as you attemped to take on that market by using Server brandname parts at Desktops price points.
If anything, providing BIOS options for even the most basic tweaking should be idea and responsability of an independent Motherboard vendor without your support, as it isn't the intended use of the platform that you build.

[zanzabar]

i think that they can lock the clocks in the firmware that goes into the bios

[zir_blazer]

Someone posted this in the News section. I doubt that eVGA got Intel blessings for doing an enthusiast class Dual Xeon Motherboard, yet they still did it. We need someone similar for doing that with AMD Opterons or whatever can be used for Dual Processor.

BT: But can you still overclock a single CPU Extreme Edition in the board though?

Shamino: No. You can't even run a single QPI [consumer Core-i7] CPU in it. The reason being that Intel twist the CPU in its socket for Xeon CPUs so it's in such a way that using a normal Core-i7 points the only available QPI to the other socket, and not the X58 chipset. Xeons with dual QPI point to both each other and the X58. Intel did it on purpose to make sure people bought their Xeon CPUs.

So unless some engineering company makes a link chip that diverts the QPI traffic to the X58, it won't happen, but I can't see why any company would bother.

Interesing...

[JF-AMD]

I understand if you can't share anymore data/info, but I'm not getting it (somewhat) from this post... You're telling us that it's not so easy as just slapping on a IHS with a different name and some die marking that would make it identify it to CPU-Z as a xxx processor rather than a Opty? For something as simple as rebranding you have to go through a whole cycle of testing and development?

So, let's take an Opteron and make it a Phenom. Pretty simple, just flip the bit in the CPUID, it's done, right?

Well, here is all of the other stuff that has to happen (even with the exact same design - think Shanghai and Phenom):

1. Forecasting/biz opps - someone has to understand the market and decided how many to make and where to sell them.
2. Program management - You have to manage the development even though it is the same silicon.
3. Software development - believe it or not, you need people to make sure the products do what you want it to do under your OS.
4. Testing - Yes, you have different test procedures for server and desktop (I don't care if you can't run world of warcraft on my processors)
5. Inventory - I have to put them in a different location in the warehouse and handle them differently because they have a different name on them. I have to mark and fuse them, so once I do that there is some amount of excess/obsolete that I have to expect (and after 4x4....)
6. Marketing - if we build it they won't come, we need to tell them about it.
7. Partner support - you will have to go out and support this with all of your partners, both design in and sales out.
8. Advertising - Remember when someone complained that we don't advertise? Well, this crowd seems to think it is important, so toss that on there.
9. Sales - If we are going to build it we will have to sell it. Selling this is additional effort that takes away from sales productivity (nothing magically appears at Newegg on its own).
10. Support - once people buy it they will have questions. I am guessing that you will want them answered.


Then, of course, you will need new testers. Those are hundreds of thousands of dollars. You can't really run them through the server testers because ours are for different platforms.

1-5 are the really expensive things. All of that would need to happen if you have a different CPUID (i.e. new brand).

6-10 are the things we have to do if you expect us to actually sell the parts. But don't assume that they are cheap.

Testers are hardware, that is a major capital expense.

On average, the cost to do a new product (not counting design) typically has a $5M or so price tag. That is why you need volume. Remember that everything that is different requires a new process and requires us to handle things differently. And that adds cost.

[BeepBeep2]

So, let's take an Opteron and make it a Phenom. Pretty simple, just flip the bit in the CPUID, it's done, right?

Well, here is all of the other stuff that has to happen (even with the exact same design - think Shanghai and Phenom):

1. Forecasting/biz opps - someone has to understand the market and decided how many to make and where to sell them.
2. Program management - You have to manage the development even though it is the same silicon.
3. Software development - believe it or not, you need people to make sure the products do what you want it to do under your OS.
4. Testing - Yes, you have different test procedures for server and desktop (I don't care if you can't run world of warcraft on my processors)
5. Inventory - I have to put them in a different location in the warehouse and handle them differently because they have a different name on them. I have to mark and fuse them, so once I do that there is some amount of excess/obsolete that I have to expect (and after 4x4....)
6. Marketing - if we build it they won't come, we need to tell them about it.
7. Partner support - you will have to go out and support this with all of your partners, both design in and sales out.
8. Advertising - Remember when someone complained that we don't advertise? Well, this crowd seems to think it is important, so toss that on there.
9. Sales - If we are going to build it we will have to sell it. Selling this is additional effort that takes away from sales productivity (nothing magically appears at Newegg on its own).
10. Support - once people buy it they will have questions. I am guessing that you will want them answered.


Then, of course, you will need new testers. Those are hundreds of thousands of dollars. You can't really run them through the server testers because ours are for different platforms.

1-5 are the really expensive things. All of that would need to happen if you have a different CPUID (i.e. new brand).

6-10 are the things we have to do if you expect us to actually sell the parts. But don't assume that they are cheap.

Testers are hardware, that is a major capital expense.

On average, the cost to do a new product (not counting design) typically has a $5M or so price tag. That is why you need volume. Remember that everything that is different requires a new process and requires us to handle things differently. And that adds cost.

Thank you.

[stangracin3]

6. Marketing - if we build it they won't come, we need to tell them about it.

this makes you sound lazy.
I gaurantee you would have little to no marketing because this would be buzzing around the forums more than when the 48xx's came out and scared the out of nvidia

[chew*]

One thing i have learned in this industry is "life finds a way"

For instance.....3 way sli on an amd chipset......or opterons with indentity crysis doing 3d at speeds/voltages only other c2 revision chips can only dream about......in fact at speed/voltages even some c3's dream about.

No that isn't a TWKR guys, a real TWKR would ID as AMD phenom 42 and also required more volts.......Cpu-z is just confused as it has no freaking clue what the cpu is

[BeepBeep2]

One thing i have learned in this industry is "life finds a way"

For instance.....3 way sli on an amd chipset......or opterons with indentity crysis doing 3d at speeds/voltages only other c2 revision chips can only dream about......in fact at speed/voltages even some c3's dream about.

No that isn't a TWKR guys, a real TWKR would ID as AMD phenom 42 and also required more volts.......Cpu-z is just confused as it has no freaking clue what the cpu is

To put chew* screen into perspective I can't even CPU-Z validate my C3 at that speed/voltage.

Makes me want to pick up an Opteron 1385.

[Deluded]

So, let's take an Opteron and make it a Phenom. Pretty simple, just flip the bit in the CPUID, it's done, right?

Well, here is all of the other stuff that has to happen (even with the exact same design - think Shanghai and Phenom):

1. Forecasting/biz opps - someone has to understand the market and decided how many to make and where to sell them.
2. Program management - You have to manage the development even though it is the same silicon.
3. Software development - believe it or not, you need people to make sure the products do what you want it to do under your OS.
4. Testing - Yes, you have different test procedures for server and desktop (I don't care if you can't run world of warcraft on my processors)
5. Inventory - I have to put them in a different location in the warehouse and handle them differently because they have a different name on them. I have to mark and fuse them, so once I do that there is some amount of excess/obsolete that I have to expect (and after 4x4....)
6. Marketing - if we build it they won't come, we need to tell them about it.
7. Partner support - you will have to go out and support this with all of your partners, both design in and sales out.
8. Advertising - Remember when someone complained that we don't advertise? Well, this crowd seems to think it is important, so toss that on there.
9. Sales - If we are going to build it we will have to sell it. Selling this is additional effort that takes away from sales productivity (nothing magically appears at Newegg on its own).
10. Support - once people buy it they will have questions. I am guessing that you will want them answered.


Then, of course, you will need new testers. Those are hundreds of thousands of dollars. You can't really run them through the server testers because ours are for different platforms.

1-5 are the really expensive things. All of that would need to happen if you have a different CPUID (i.e. new brand).

6-10 are the things we have to do if you expect us to actually sell the parts. But don't assume that they are cheap.

Testers are hardware, that is a major capital expense.

On average, the cost to do a new product (not counting design) typically has a $5M or so price tag. That is why you need volume. Remember that everything that is different requires a new process and requires us to handle things differently. And that adds cost.

Wow. Thanks JF! I really appreciate it. Now I understand your side of the story. It pains me to say it, it seems you are indeed correct and the course of action you have to take regarding your optys is correct and what we ask is indeed a little bit too much to ask of AMD... but damn, you sure know how to kill dreams!

No matter, I'm sure someone on XS will be insane enough to mod a BIOS for their own use.

Although that position of tester/software testing is sounding mightily interesting... any chance you might have any info with regards to open positions in LA regions?

This quote of yours piqued my interest...

I have already explained that allowing overclocking on opteron would result in lower net shipments of Opteron products. We would end up selling less because the net decrease in server sales would never be offset by the small number of enthusiast sales.

If I want to sell more I need to keep Opteron focused where it is and leave the Phenom to enthusiasts.

You sound like you speak from experience. Are you perhaps speaking of AMD's decision to allow server grade processors to be sold as desktop processors back in s939 days, and the backlash from that moment onward (which apparently is still felt today, given your quote)?

[MrMojoZ]

this makes you sound lazy.
I gaurantee you would have little to no marketing because this would be buzzing around the forums more than when the 48xx's came out and scared the out of nvidia

Hmm are they lazy or do some forum dwellers not know when to give it up. Hmmmmm.

[Sn0wm@n]

Easily done. Just point them to s7 post -- he overclocked two mangy cours 1.3Ghz above stock on stock vcore. I don't know about you, but to us, that is DAMN impressive! 24 cores @ 3Ghz, up from 1.7Ghz stock....



I understand if you can't share anymore data/info, but I'm not getting it (somewhat) from this post... You're telling us that it's not so easy as just slapping on a IHS with a different name and some die marking that would make it identify it to CPU-Z as a xxx processor rather than a Opty? For something as simple as rebranding you have to go through a whole cycle of testing and development?

opteron's are meant for 24/7 100% use for a couple of years ... so they need to be able to take a beating in a small enclosed space ..... so the extra validation steps are to ensure that each cpu branded as an opteron gets to live up to its name


but to us its only a different name on the ihs ....

[JF-AMD]

This quote of yours piqued my interest...

You sound like you speak from experience. Are you perhaps speaking of AMD's decision to allow server grade processors to be sold as desktop processors back in s939 days, and the backlash from that moment onward (which apparently is still felt today, given your quote)?

When the pricing for Opteron fell below the pricing for Athlon at the time, it was a disconnect inside the company (long story, too hard to explain.)

We saw an uplift in server parts as people started buying Opteron instead.

The problem was that resellers who were buying Athlon and not dealing with Opteron suddenly felt the shift. And Opteron resellers, who don't deal with consumers now had to contend with that.

Then, when prices came back in line, everyone jumped back to Athlon (who says price doesn't matter.) That disrupts the supply chain and creates all kinds of problems with forecasting, inventory, etc.

We never wanted to sell server parts as desktops and never meant for that to happen, it was a disconnect internally where the 2 teams were not in sync. So when people say we allowed it to happen in the past and why can't we do that again, they don't realize it wasn't intentional.

[BeepBeep2]

When the pricing for Opteron fell below the pricing for Athlon at the time, it was a disconnect inside the company (long story, too hard to explain.)

We saw an uplift in server parts as people started buying Opteron instead.

The problem was that resellers who were buying Athlon and not dealing with Opteron suddenly felt the shift. And Opteron resellers, who don't deal with consumers now had to contend with that.

Then, when prices came back in line, everyone jumped back to Athlon (who says price doesn't matter.) That disrupts the supply chain and creates all kinds of problems with forecasting, inventory, etc.

We never wanted to sell server parts as desktops and never meant for that to happen, it was a disconnect internally where the 2 teams were not in sync. So when people say we allowed it to happen in the past and why can't we do that again, they don't realize it wasn't intentional.

Well what about Opteron 12xx series and 13xx series and Phenom/Phenom II?

[demonkevy666]

http://www.bit-tech.net/hardware/mot...-motherboard/1

I have no idea what he's talking about most of the time, but it's still interesting.

SR2 keeps getting mentioned in here, yet no one linked this from new section ?
this why is it's out...

[JF-AMD]

Well what about Opteron 12xx series and 13xx series and Phenom/Phenom II?

Can you elaborate? I am not following your question.

[zanzabar]

Can you elaborate? I am not following your question.

they are the same thing (or have corresponding electrically identical parts parts), they just have a different laser etch on the IHS and name on the cpu ID (and the testing and binning for opterons but thats not needed for a phenom). so for a mangy since all of the testing is done on the opterons and that is were u would pull from, then u would not have to do more testing just send out a small batch with a different etching and an unlocked multi and maybe a high VID as u can change the multi on some boards but not the VID. send it out to some benchers and get a huge jump on CB and Wprime

[zir_blazer]

so for a mangy since all of the testing is done on the opterons and that is were u would pull from, then u would not have to do more testing just send out a small batch with a different etching and an unlocked multi and maybe a high VID as u can change the multi on some boards but not the VID. send it out to some benchers and get a huge jump on CB and Wprime

I repeat: We shouldn't look forward for AMD support on this. What we need is an independent Motherboard vendor or BIOS developer that wants to take that risk all by themselves.