Kaveri performance

[SiliconWars]

There is no chance of Kaveri being fabbed at TSMC for so many obvious reasons.

1) If AMD had that kind of capacity at TSMC they'd put it to much better use than Kaveri.
2) TSMC has no expertise in this kind of chip.
3) Kaveri is a continuation of Llano/Trinity/Piledriver which is the massive part of what AMD is paying the WSA with.

Now anyone could invent a lot of convoluted reasons why things can change (Jaguar to GF allowing Kaveri to TSMC) etc, but none of it will happen.

The market is settled. Big cores do not matter to either AMD or Intel. Both will stagnate. AMD needs more than what Kaveri is capable of providing in order to break Intel's stranglehold on the desktop and laptop market. In the end, they've both settled for the way it is.

Kaveri will pay the WSA in the same way Richland currently is while keeping AMD's big core chips at the market share they currently have. There is nothing to gain from AMD moving it to TSMC. It's all risk and the rewards are not there.

Jaguar will remain at TSMC. There may be a chance of GF fabbing the console chips down the line, but even that is doubtful to me when I think about it. The reason? AMD will be on 20nm at TSMC soon enough, moving Jaguar and graphics there long before consoles or any desktop APU would be, so why not just keep TSMC fabbing the consoles at 28nm until 20nm is at the same mature stage as 28nm is now?

GF will be used as the fab for the lagging edge markets until they prove themselves at 20nm and 14nm.

You are right about the AMD/GF relationship in some ways, but there is no great big-core comeback coming out of TSMC.

 

[Idontcare]

2) TSMC has no expertise in this kind of chip.

Why do you say this? What is it about Kaveri that qualifies it as being a kind of chip for which TSMC has no expertise? Is it because of the clockspeed?

(TSMC has big-chip experience, Nvidia's massive 550mm^2 GPUs, as well as fabbing big MPUs like SUN's T3 chips, etc)

 

[ShintaiDK]

If AMD manufactors Kaveri at TSMC. Where does that leave the WSA and GloFo.

 

[NostaSeronx]

If AMD manufactors Kaveri at TSMC. Where does that leave the WSA and GloFo.

Waived. Except for SOI nodes.

 

[NTMBK]

Waived. Except for SOI nodes.

If you could point us at where that was announced, that would be great ta.

 

[SiliconWars]

Why do you say this? What is it about Kaveri that qualifies it as being a kind of chip for which TSMC has no expertise? Is it because of the clockspeed?

(TSMC has big-chip experience, Nvidia's massive 550mm^2 GPUs, as well as fabbing big MPUs like SUN's T3 chips, etc)

Nothing really close to this clockspeed (granted we don't know what the clocks will be on Kaveri but assume high) and no APU of this complexity, if we assume Jaguar is the most complex APU they've done.

I don't know how much of a drawback that would be, but there is probably some kind of risk there that would contribute to AMD not considering TSMC for this. There are other reasons that are more valid I'm sure.

 

[SiliconWars]

Waived. Except for SOI nodes.

Even if this happened AMD would need some extra 35000 WPM for Kaveri alone at TSMC.

 

[mrmt]

Waived. Except for SOI nodes.

Do you mind telling your sources? Because what you are stating here, if true, would require an amendment of the WSA, which then would require notifying investors of a relevant fact through a SEC filling. This also contradicts something that AMD CFO stated in a Q&A with investors.

So you are either BS or the most important part of the information isn't the WSA waiver, but that AMD is purposely deceiving its shareholders and breaking the law.

 

[desura]

The only thing that matters is the IGP performance. CPU speed stopped mattering...years ago.

 

[el etro]

AMD Processors IPC has stopped into time. Kaveri needs to push it a lot to make their future apus competitive against the new haswell counterparts(saying about the i3 and Pentiuns, this ones will come more clocked this time).
BTW, the rumors about the future TOP EXCVTR apu being only 65W TDP is awesome. This shows AMD is very conscious too about the shrinking of the Personal Computing market.

 

[frozentundra123456]

The only thing that matters is the IGP performance. CPU speed stopped mattering...years ago.

Then why does amd have such a small portion of the market?

 

[SiliconWars]

Then why does amd have such a small portion of the market?

Intel rebates/bribery/etc. Do we really need to go through this one AGAIN frozentundra or do you own any other arguments?

 

[NostaSeronx]

Then why does amd have such a small portion of the market?

Here is AMD's reasoning for it:

Intel Corporation’s dominance of the microprocessor market and its aggressive business practices may limit our ability to compete effectively.

Intel Corporation has dominated the market for microprocessors for many years. Intel’s market share, margins and significant financial resources
enable it to market its products aggressively, to target our customers and our channel partners with special incentives, and to discipline customers who do
business with us. These aggressive activities have in the past and are likely in the future to result in lower unit sales and a lower average selling price for our
products and adversely affect our margins and profitability.

Intel exerts substantial influence over computer manufacturers and their channels of distribution through various brand and other marketing programs.
As a result of Intel’s dominant position in the microprocessor market, Intel has been able to control x86 microprocessor and computer system standards and
benchmarks and to dictate the type of products the microprocessor market requires of us. Intel also dominates the computer system platform, which includes
core logic chipsets, graphics chips, motherboards and other components necessary to assemble a computer system. Original Equipment Manufacturers
(OEMs), that purchase microprocessors for computer systems are highly dependent on Intel, less innovative on their own and, to a large extent, are
distributors of Intel technology. Additionally, Intel is able to drive de facto standards for x86 microprocessors that could cause us and other companies to have
delayed access to such standards.

Intel has substantially greater financial resources than we do and accordingly spends substantially greater amounts on marketing and research and
development than we do. We expect Intel to maintain its dominant position and to continue to invest heavily in marketing, research and development, new
manufacturing facilities and other technology companies. To the extent Intel manufactures a significantly larger portion of its microprocessor products using
more advanced process technologies, or introduces competitive new products into the market before we do, we may be more vulnerable to Intel’s aggressive
marketing and pricing strategies for microprocessor products.

Intel also leverages its dominance in the microprocessor market to sell its integrated graphics chipsets. Intel manufactures and sells integrated graphics
chipsets bundled with their microprocessors and is a dominant competitor with respect to this portion of our business. Intel could also take actions that place
our discrete GPUs at a competitive disadvantage, including giving one or more of our competitors in the graphics market, such as Nvidia Corporation,
preferential access to its proprietary graphics interface or other useful information.

As long as Intel remains in this dominant position, we may be materially adversely affected by Intel’s:
• business practices, including rebating and allocation strategies and pricing actions, designed to limit our market share and margins;
• product mix and introduction schedules;
• product bundling, marketing and merchandising strategies;
• exclusivity payments to its current and potential customers and channel partners;
• control over industry standards, PC manufacturers and other PC industry participants, including motherboard, memory, chipset and basic
input/output system, or BIOS, suppliers and software companies as well as the graphics interface for Intel platforms; and
• marketing and advertising expenditures in support of positioning the Intel brand over the brand of its OEM customers.

Intel’s dominant position in the microprocessor market and integrated graphics chipset market, its existing relationships with top-tier OEMs and its
aggressive marketing and pricing strategies could result in lower unit sales and a lower average selling price for our products, which could have a material
adverse effect on us.

Do you mind telling your sources?

It is super vague, but 28-nm for AMD is only TSMC. With the consoles being the only exception for this rule. They should be getting fabbed at TSMC, GlobalFoundries/IBM/Samsung, and UMC or SMIC(Uses IBM aka Common Platform).

Second Amendment to Wafer Supply Agreement. On March 4, 2012, the Company entered into a second amendment to the WSA with GF. The
primary effect of this second amendment was to modify certain pricing and other terms of the WSA applicable to wafers for the Company’s microprocessor
and APU products to be delivered by GF to the Company during 2012. The second amendment also granted the Company certain rights to contract with
another wafer foundry supplier with respect to specified 28nm products for a specified period of time. In consideration for these rights, the Company agreed to
pay GF $425 million and transfer to GF all of the capital stock of GF that it owned. As a result of the Company receiving these rights in the first quarter of
2012, the Company recorded a charge related to this limited waiver of exclusivity from GF of $703 million consisting of the $425 million cash payment and a
$278 million non-cash charge representing the carrying and fair value of the capital stock that the Company transferred to GF. Pursuant to the second
amendment, the Company paid the full amount of $425 million by December 31, 2012. Of this amount, the final portion of $175 million was paid during
the Company’s first fiscal quarter of 2013.

Specified products: Dublin, Macau, Kaveri, Witchita, Krishna, as far as I can tell.

With the delay and change we get a 55-nm(R600) to 45*cough*40-nm(R700) performance increase.

Other than that Kaveri is the only 28-nm Steamroller, officially on the map. While the 20-core product with Steamroller cores that comes after Warsaw hasn't appeared yet. My assumption is that we will see 32-nm PD-SOI with 12/16-core products with Warsaw transition to 2x-nm ETSOI with 12/16/20-core products. 2x-nm ETSOI should get started on its ramp exactly after 28-nm FDSOI from STM is finished.

 

[CHADBOGA]

The only thing that matters is the IGP performance. CPU speed stopped mattering...years ago.

Don't be ridiculous.

Then why does amd have such a small portion of the market?

Because CPU performance still matters.

Intel rebates/bribery/etc. Do we really need to go through this one AGAIN frozentundra or do you own any other arguments?

AMD's marketshare has constantly been falling ever since Intel released the clearly better Core 2 architecture.

AMD's marketshare increased when they had the architecture advantage in the Athlon64 over the P4.

The above is by far the most significant factors, everything else is trifling in importance.

 

[SiliconWars]

AMD's marketshare has constantly been falling ever since Intel released the clearly better Core 2 architecture.

AMD's marketshare increased when they had the architecture advantage in the Athlon64 over the P4.

The above is by far the most significant factors, everything else is trifling in importance.

This is also easily explained by Intel not having their rebate/bribery/marketing program in full swing until after Athlon64/P4.

And here is the problem with using market share as the sole value of superiority. Throughout history there have been numerous cases of the better technology losing out in the market (betamax being a well-known example), for various reasons.

Hell we shouldn't even be using "PC's", so inferior were they to machines like the Amiga, ST and Archimedes (which is one reason why ARM beating Intel would be so damned hilarious). But them with the most money get to dictate who wins, it really is that simple and it's true in every single aspect of your life.

 

[ShintaiDK]

This is also easily explained by Intel not having their rebate/bribery/marketing program in full swing until after Athlon64/P4.

And here is the problem with using market share as the sole value of superiority. Throughout history there have been numerous cases of the better technology losing out in the market (betamax being a well-known example), for various reasons.

It cant be that AMD products are simply undesireable in almost all metrics. Noooo... :awe:

 

[CHADBOGA]

This is also easily explained by Intel not having their rebate/bribery/marketing program in full swing until after Athlon64/P4.

Accept that AMD alleges it was in full place before Core 2.

 

[SiliconWars]

It cant be that AMD products are simply undesireable in almost all metrics. Noooo... :awe:

Yet AMD gained how much market share last quarter? (see how easy it is)

 

[mrmt]

Here is AMD's reasoning for it:It is super vague, but 28-nm for AMD is only TSMC. With the consoles being the only exception for this rule. They should be getting fabbed at TSMC, GlobalFoundries/IBM/Samsung, and UMC or SMIC(Uses IBM aka Common Platform).Specified products: Dublin, Macau, Kaveri, Witchita, Krishna, as far as I can tell.

The third amendment to the WSA killed the waiver from the second amendment. This was exactly the issue brought up by analysts in the Q&A and this is what lead Kumar to say that all CPU production and part of the GPU production would move to Globalfoundries. Again, state your souces, because you cannot reasonably expect a defunct amendment to back you up on this one.

 

[NostaSeronx]

*sarcasm* So, you are saying my confidential info from TSMC & AMD backstabbing GlobalFoundries is wrong? */sarcasm*

I don't know if you are blind but seriously you need to get your understanding of the english language up. AMD doesn't have any processors on 28-nm Gate-First. They shuffled everything towards 28-nm Gate-last because it was producing more units for less. Basically, anything said about GlobalFoundries and 28nm from AMD is damage control. It is that hostile between GlobalFoundries and AMD with 28-nm.

 

[NTMBK]

Basically, anything said about GlobalFoundries and 28nm from AMD is damage control. It is that hostile between GlobalFoundries and AMD with 28-nm.

So the bit where they explicitly said that the 28nm successor to Trinity (i.e. Kaveri) was being fabbed at GloFo was just "damage control"? :whiste:

"If you look at the roadmaps that we have presented at Financial Analyst Day, there is a 28-nanometer successor product to Trinity on the roadmap that we will ramp next year, and that is also manufactured at GLOBALFOUNDRIES."

http://www.brightsideofnews.com/new...lk-at-gf2c-more-details-from-the-new-wsa.aspx

EDIT: And your track record for secret inside information on GloFo 28nm is pretty shabby. How did that 28nm Richland work out for you? http://forums.anandtech.com/showpost.php?p=34275041&postcount=10

 

[NostaSeronx]

So the bit where they explicitly said that the 28nm successor to Trinity (i.e. Kaveri) was being fabbed at GloFo was just "damage control"? :whiste:

Temash, Kabini, Kaveri, Berlin, Seattle, Beema, and Mullins are all being fabbed at TSMC.

The only exception is the console APUs because there is a middle man.

EDIT: And your track record for secret inside information on GloFo 28nm is pretty shabby. How did that 28nm Richland work out for you? http://forums.anandtech.com/showpost.php?p=34275041&postcount=10

Pre-2013, nine month gap of information. For the Richland thing it was an explanation for the Steamroller module w/o L2 pictured dieshot.

 

[NTMBK]

Temash, Kabini, Kaveri, Berlin, Seattle, Beema, and Mullins are all being fabbed at TSMC.

The only exception is the console APUs because there is a middle man.

You keep saying that with no evidence. AMD have explicitly stated that Kaveri is at GlobalFoundries. Just repeating it with no evidence is not an argument.

 

[NostaSeronx]

You keep saying that with no evidence. AMD have explicitly stated that Kaveri is at GlobalFoundries. Just repeating it with no evidence is not an argument.

Berlin has been explicitly stated to be fabbed at TSMC. Not on a different process, not on a different node offering, but on the same process with the same node offering as Jaguar.

 

[NTMBK]

Pre-2013, nine month gap of information. For the Richland thing it was an explanation for the Steamroller module w/o L2 pictured dieshot.

Richland would have had to have been in production at the point that you posted in order for it to come out in June 2013 (and some OEM parts launched even earlier, if I remember correctly). Arguing "gap of information" is nonsense. If you had inside information about what was being made where, then you would have known that Richland was coming out on 32nm, and that it was just a respin of Trinity as opposed to a completely new part (which a move to 28nm would imply).