Full Skylake reveal result? Waiting for Zen.

[looncraz]

Probably not, but still the "high volume" APU parts will need a south bridge, which have to come from ASMedia, and given that we should be talking about a monolithic die, ASMedia has to license IP for AMD, and certainly ASMedia isn't going to give AMD a free lunch.

And that will paid for by the motherboard companies, who already have strong business ties with AsMedia, not AMD. Of course, either way, WE end up paying for it... but we'd pay for it either way, so it's a total non-issue.

And, in the event AMD does use AsMedia to supply an on-die south bridge (doubtful, but possible), you can bet they already have a deal worked out. AMD did just hand AsMedia a considerable amount of business.

You talk as if ASMedia is some charity. ASmedia will either get paid directly by AMD or they will get a chunk of AMD revenue share in the ecosystem. AMD might end up better than they are today, but doesn't change the fact that:

- AMD will have to pay for ASMedia one way or another.

AsMedia was just handed a hundreds of millions of dollars worth of business. If you don't think that came with strings attached, you're wrong. You can rest assured that AMD will be benefiting greatly from this arrangement.

However, AMD will certainly *NOT* be paying royalties or anything to AsMedia for off-die components. They have no reason to do so. Usually, in fact, third party chipset manufacturers usually pay licensing fees to the CPU company for access and production rights. In this case, I imagine AMD has given AsMedia exclusive rights to produce their chipsets in exchange for a guarantee that AsMedia will create appropriate chipsets for AMD's products that contains the latest technologies.

That is all that would need to happen for everyone to win and no one to lose. It is also the only deal that would make sense for AMD or for AsMedia. AMD doesn't get direct access to AsMedia's IP, so they have no reason to pay anything. They're simply giving someone else exclusive rights to develop their chipsets.

- AMD might end up in a worse financial position in the long run compared to a situation where they could muster enough resources to develop these technology by themselves.

This is possible, yes. Not because of any royalties, but because they are now not collecting on any SB / GPP chips. They also probably lose some control over choosing what is included, which might make the market choices a little more confusing (probably not at first, though).

However, it allows AMD to reduce their R&D burden and their time-to-market. They now just create a system protocol and share that with AsMedia, likely with some general requirements included for minimum platform capabilities. AsMedia makes the products because they know no one else can do so, and they include their existing IP just to save money. For them, it will be like dragging & dropping their designs onto a basic chip layout, testing it, and saying: here ya go, I want $4.20 for this south bridge!

- Given the minuscule volume of AMD processors as of lately, whatever developments made specifically for them will be very hard to pay off because of the small number of units to amortize R&D costs.

This is also why AMD had to drop as much of their cruft as possible. For some companies, vertical integration makes sense... for companies which can't keep up in terms of competitiveness and profitability, it is usually better to find your profitable little niche. AMD's, hopefully, will be x86 CPUs, gaming graphics, consoles, etc...

They don't need their own fabs anymore (settlement with Intel allowed this) and they don't need their own chipsets anymore (AsMedia will handle this). At the same time, they have to get rid of all of their money-pits... which is exactly what they have been doing. Their chipsets were just there to enable their CPUs to sell. But, they couldn't keep up.

I expect the highest end AM4 motherboards to be quite a bit more capable after the AsMedia deal than I ever thought before. In fact, my first concern when I heard about Zen was AMD's outdated platform... that no longer concerns me.

Which Intel product has AMD >>technology<< on it?

Almost every one of their CPUs. AMD-64 is used on Intel CPUs to provide 64-bit and 32-bit concurrent compatibility. The fact that you can run 32-bit software on 64-bit operating systems is entirely thanks to AMD. Intel was going full-on 64-bit, with no backwards compatibility. Their attempt (Itanium) failed horribly, and AMD hit the market with awesome 64-bit CPUs.

There are other bits as well, I'm sure, as more was included in the cross-licensing agreement, but I have no idea what Intel actually implements that we can't see. And, of course, AMD uses Intel technologies.

 

[ShintaiDK]

Agreed.


Yes, it did. There were games with lesser CPU requirements if Glide was used, versus DirectX.


The early games ran like crap w/ 32-bit color anyway. Marketing feature it was really. And later, with Voodoo5 32-bit color as well as larger textures were added (you could run any Glide game in 32-bit color, if desired). Sure, DirectX has massively evolved since, but back in the day, Glide was the preferable API to choose from, if one was looking for speed.

Apologize for a bit of the off topic content here.

No it didnt. And Glide never supported 32bit colours. Performance wise Glide quickly lost its edge as well

 

[ShintaiDK]

And that will paid for by the motherboard companies, who already have strong business ties with AsMedia, not AMD. Of course, either way, WE end up paying for it... but we'd pay for it either way, so it's a total non-issue.

And, in the event AMD does use AsMedia to supply an on-die south bridge (doubtful, but possible), you can bet they already have a deal worked out. AMD did just hand AsMedia a considerable amount of business.

AMD stopped chipset development, including IP blocks on their CPUs. So AMD has to pay AsMedia.

AsMedia is also a huge liability for AMD if it wish to reenter the server segment. Not that its likely to happen anyway, but still.

AMD didnt outsource this because its good business, they had to do it because they was forced to. Just like the fabs, HQ etc.

You are dead wrong if you think AMD does these thing because they wish to. They will try and spin the PR as its a great thing. But its not. And you see a company with an increasingly more difficult cost structure.

 

[looncraz]

AMD stopped chipset development, including IP blocks on their CPUs. So AMD has to pay AsMedia.

AsMedia is also a huge liability for AMD if it wish to reenter the server segment. Not that its likely to happen anyway, but still.

AMD didnt outsource this because its good business, they had to do it because they was forced to. Just like the fabs, HQ etc.

You are dead wrong if you think AMD does these thing because they wish to. They will try and spin the PR as its a great thing. But its not. And you see a company with an increasingly more difficult cost structure.

Let me explain to you how this works, since you don't seem to know, or are laboring under a false impression:

AMD has an interface from the CPU to the system board. Everything on the CPU belongs to AMD, generally speaking. They make their own memory controller and transport logic. This has not been a part of chipsets for many many years. Good thing, as well, since AsMedia has no expertise in this area - and using a 3rd party's technology here would be a logistical nightmare. AMD's CPU team has the expertise in this area as the core logic, memory controller, and data fabric all integrate tightly.

This integration eliminates the need for a north bridge chip entirely. One is not designed, one is not integrated on the CPU, it simply doesn't exist. The role it served has been usurped by the CPU itself. There is no room for IP blocks on the die in this equation (except for SoCs/AM1) beyond PCI-e, for which AMD already owns up-to-date designs and licenses.

Since there is no indication, what-so-ever, that AMD will integrate everything on the die, there is no reason to assume that AMD will be paying royalties or licensing fees for AsMedia's IP. However, if they were, paying AsMedia would probably still be cheaper than losing hundreds of millions trying to keep their sinking ship afloat.

And, in the end, motherboards would then be cheaper to build, and some of that savings would fall down to the consumer (though not all of it, naturally).

But, most likely, AMD will build a normal CPU/APU, with a high-speed system interface and PCI-e and interface with a south bridge chip. Just like Intel. And wouldn't they? It would be stupid to include anything extra on the CPU, on an untested process, which will just make the CPU larger, reduce yields, and increase costs with absolutely no benefit.

HERE:

AsMedia will now be handling the "AMD Fusion Controller Hub." That's it.

 

[ShintaiDK]

Not sure who you try to educate because it doesnt seem like you kept yourself up to date in any way. Carrizo for example got it all on the CPU. And AMDs last north bridge chip is older than your account.

Its not going to be any cheaper because they was forced to outsource it. Its simply another pee in the pants. And the outside PCH/FCH is disappearing fast.

 

[looncraz]

Not sure who you try to educate because it doesnt seem like you kept yourself up to date in any way. Carrizo for example got it all on the CPU. And AMDs last north bridge chip is older than your account.

Its not going to be any cheaper because they was forced to outsource it. Its simply another pee in the pants. And the outside PCH/FCH is disappearing fast.

You know, that big banner on it that says "AMD \"CARRIZO\" NOTEBOOK AND AIO PLATFORM" should have completely given away everything here.

As I said, other than SoC (such as Carrizo), AMD will not need to pay any royalties. In addition, and try to stay with me here, AMD already owns or licenses all of that IP on Carrizo. That situation does not change. AMD will still be licensing designs, IP, etc. just like it does now... for SoCs only.

That will not impact a single Zen-based CPU until 2017 or later, unless AMD went so crazy as to include everything on the die... which would be really stupid.

Notice, though, the only things from the FCH that need to evolve:

USB and SATA.

Seriously, that's it. Everything else can remain exactly the same for years to come and you would not hear a single complaint from anyone. Those things are just standard buses that have been around since forever. GPIO, UART, I2C, and I2S, for example, haven't changed since the 90s. They are usually used for connecting simple sensors and low-speed devices.

AsMedia has great USB 3.0 and SATA3 controllers already. AMD had already licensed their SATAe tech, in fact, and I imagine that prior agreement may have been reworked as part of the new partnership.

I imagine AMD could pay about $0.25 per CPU to AsMedia for USB and SATA IP, if not less, since AMD will be paying for the tech to be produced, and will also be the ones integrating it onto their dies if it comes to that. Not much of a burden.

 

[Exophase]

And Glide never supported 32bit colours.

Glide 3.0 did. Read the user manual here: http://alasir.com/software/glide/glide3_pg.pdf.gz See for example table 11.3.

 

[mrmt]

However, AMD will certainly *NOT* be paying royalties or anything to AsMedia for off-die components. They have no reason to do so. Usually, in fact, third party chipset manufacturers usually pay licensing fees to the CPU company for access and production rights. In this case, I imagine AMD has given AsMedia exclusive rights to produce their chipsets in exchange for a guarantee that AsMedia will create appropriate chipsets for AMD's products that contains the latest technologies.

Since there is no indication, what-so-ever, that AMD will integrate everything on the die, there is no reason to assume that AMD will be paying royalties or licensing fees for AsMedia's IP. However, if they were, paying AsMedia would probably still be cheaper than losing hundreds of millions trying to keep their sinking ship afloat.

The SoC share is reaching half of Intel product mix and should go even further in the next years. AMD should have even more than this, because everything they sell now is SoCs, except for the decrepit piledriver FX line. In the end the desktop Zen will matter very little for this discussion, because the SoCs, which will have a much higher unit counts, will have ASMedia IP integrated on the die and consequently AMD will be the one forking money for it.

And that will paid for by the motherboard companies, who already have strong business ties with AsMedia, not AMD. Of course, either way, WE end up paying for it... but we'd pay for it either way, so it's a total non-issue.

Non-issue to who? Because last time I checked AMD has been in sore need of extra revenue. The last thing they need is to hand over whatever is left of their market to a partner. And that's exactly what happened with ASMedia.

AsMedia was just handed a hundreds of millions of dollars worth of business. If you don't think that came with strings attached, you're wrong. You can rest assured that AMD will be benefiting greatly from this arrangement.

Hundreds of millions? How so? AMD entire CPU business is a hundred of millions affair, their chipset business shouldn't break the tens of millions per year. This is *exactly* the reason AMD had to outsource chipset manufacturing, not enough volume to sustain R&D.

 

[looncraz]

The SoC share is reaching half of Intel product mix and should go even further in the next years. AMD should have even more than this, because everything they sell now is SoCs, except for the decrepit piledriver FX line. In the end the desktop Zen will matter very little for this discussion, because the SoCs, which will have a much higher unit counts, will have ASMedia IP integrated on the die and consequently AMD will be the one forking money for it.

You didn't seriously just use Intel's SoC statistics to try and say something about AMD's totally incomparable market positioning, did you? :thumbsdown:

AMD has stated that they will be focusing on more profitable markets. SoC markets are NOT the most profitable. The only SoC products they have are Carrizo and the AM1 platform. These are dirt-cheap products - the top of the line Carrizo (FX-8800P) goes into sub $700 laptops... and AM1 APUs are all between $25 and $50.

If you think AMD will be paying for top of the line AsMedia IP for that level of product, you are sadly mistaken. And these products do not currently require anything they don't already have.

Non-issue to who? Because last time I checked AMD has been in sore need of extra revenue. The last thing they need is to hand over whatever is left of their market to a partner. And that's exactly what happened with ASMedia.

No, they don't need more revenue, they need more PROFIT. Their peripheral and chipset operations were expenses they could not afford. AsMedia makes very good products (hence why most/all of the motherboard manufacturers use them) - many of their products outperform Intel's... that's a darn good thing to see coming to an AMD platform. Better still, AsMedia products are mature and have good drivers and support. AMD is buying what they could only deliver after a year or so on the market and tens of millions spent in development.

Hundreds of millions? How so? AMD entire CPU business is a hundred of millions affair, their chipset business shouldn't break the tens of millions per year. This is *exactly* the reason AMD had to outsource chipset manufacturing, not enough volume to sustain R&D.

Time is a factor in my statement () Also, AMD has $4b in revenue a year, mostly from CPUs. It is not hard to imagine that reaching $6B with Zen or Zen+ if they bring back a level of competitiveness similar to, or better than, the phenom II vs Core 2 Quad days.

And, again, it's not product volume that makes or breaks AMD. It's net income, otherwise known as profit. Though, to be more precise, it's a combination of margin, profit, operating capital, and revenue.

If AMD had the world's fastest CPUs and GPUs, 50% market share, 50% gross margin, they could still be showing massive losses and be in immediate danger of bankruptcy if they spent more money than they had to spend. Their operating expenses are quite large, as well as their R&D budget. Something has to give, otherwise there will be no more AMD.

There are no downsides with using AsMedia. In fact, even if AMD is paying AsMedia for each CPU they make, it just effects the Gross Margin. They could still end up being more profitable since they didn't have the platform issues. They can now take those employees and focus them onto something more profitable - such as the lucrative server market.

 

[AtenRa]

AMD stopped chipset development, including IP blocks on their CPUs. So AMD has to pay AsMedia.

Documentation and/or links for this ???

 

[Dresdenboy]

Creating their own logic to recreate existing functionality causes redundant R&D costs. Just see Idontcare's quote in Shintai's sig.

Freeing some teams this way makes them available for other projects.

 

[AtenRa]

Creating their own logic to recreate existing functionality causes redundant R&D costs. Just see Idontcare's quote in Shintai's sig.

Freeing some teams this way makes them available for other projects.

Im asking about the integrated FCH being designed by ASmedia or not.

 

[Dresdenboy]

Im asking about the integrated FCH being designed by ASmedia or not.

That makes no difference. AMD integrated TrueAudio, TrustZone and other IP blocks already and they don't appear as separate chips on the package or PCB.

BTW, Anton Shilov said nearly the same:

Advanced Micro Devices and Asmedia have reportedly signed a pact under which the latter will develop core-logic sets for the former. The move will help AMD to reduce its expenses and use internal engineering resources to develop leading-edge microprocessors and graphics processing units.

Working in a similar world, I know how much things do look in reality. Just imagine AMD having to support all those I/O standards. An ebook about SATA, partially provided by Mindshare, already lists 464 pages of human prosa covering the topic. I didn't find the specs themselves, but suppose, that they're coming with at least twice the pages.

Developing all the support for standardized I/O incl. the validation, verification and test necessary to put that into reliable products is a burden on top of the already complicated CPU/GPU/IMC stuff. ASMedia and similar IP providers are able to make use of redundancies by developing something once and licensing it to many. BTW by trying to keep business, such companies might be better motivated (even if caused by fear sitting in the limbic system) than one's own employees.

 

[AtenRa]

Well all that is good and fine but it is another thing to design a in-house separated chip and another to integrate a FCH in to a 14nm FF SOC.
ASMedia will have to design for the GloFo 14nm FF LPP and be in contact with AMD for the entire SOC design. This will create a nightmare because AMD will not want to unveil parts of its design and IP to ASMedia and ASMedia will need access to certain IP and designs of the SOC to integrate the FCH in to the final design.
And since ASMedia doesnt have any past experience with x86 CPU designs and because of the x86 licenses, i strongly believe the integrated FCH in AMD SOCs are AMDs in-house designs.

edit: not only that, but ASMedia will have to use a very expensive 14nm FF process for a low volume GPP chipset. Or, they will have to design a separated GPP Chipset at 28nm and another design for the 14nm FF integrated FCH. Not going to happen, the design alone for the 14nm FF FCH will be too high for it to be financial viable for the entire project. Dont forget they will not manufacture and sell the integrated FCH at high margins but they will only get royalties.

 

[ShintaiDK]

Well all that is good and fine but it is another thing to design a in-house separated chip and another to integrate a FCH in to a 14nm FF SOC.
ASMedia will have to design for the GloFo 14nm FF LPP and be in contact with AMD for the entire SOC design. This will create a nightmare because AMD will not want to unveil parts of its design and IP to ASMedia and ASMedia will need access to certain IP and designs of the SOC to integrate the FCH in to the final design.
And since ASMedia doesnt have any past experience with x86 CPU designs and because of the x86 licenses, i strongly believe the integrated FCH in AMD SOCs are AMDs in-house designs.

This makes no sense. All ASMedia has to do is simply delivering an IP block. Its no different than what we see with ARM for example.

 

[Dresdenboy]

Well all that is good and fine but it is another thing to design a in-house separated chip and another to integrate a FCH in to a 14nm FF SOC.
ASMedia will have to design for the GloFo 14nm FF LPP and be in contact with AMD for the entire SOC design. This will create a nightmare because AMD will not want to unveil parts of its design and IP to ASMedia and ASMedia will need access to certain IP and designs of the SOC to integrate the FCH in to the final design.
And since ASMedia doesnt have any past experience with x86 CPU designs and because of the x86 licenses, i strongly believe the integrated FCH in AMD SOCs are AMDs in-house designs.

edit: not only that, but ASMedia will have to use a very expensive 14nm FF process for a low volume GPP chipset. Or, they will have to design a separated GPP Chipset at 28nm and another design for the 14nm FF integrated FCH. Not going to happen, the design alone for the 14nm FF FCH will be too high for it to be financial viable for the entire project. Dont forget they will not manufacture and sell the integrated FCH at high margins but they will only get royalties.

There are NDAs to cover possible leaks. But as I know a long time GF'er (working there since first Fab 30 production), I know that at that stage there isn't much to know about the microarchitecture, but more about power and frequency. So it likely happens at many design levels. And ASMedia just needs to know the interfaces, nothing more.

14nm production should be covered by AMD as they integrate the ASMedia stuff. The latter might get the design rules and tooling, but don't need to create their own chips. As you've seen the last years, there are final product designs already with the second stepping. Mask set prices drove pre-production testing efforts.

 

[tential]

How fast is Zen expected to be with 8 cores with hyperthreading? Sandybridge fast means it's quite interesting. Slower than that, not so much....

 

[ShintaiDK]

How fast is Zen expected to be with 8 cores with hyperthreading? Sandybridge fast means it's quite interesting. Slower than that, not so much....

AMD cant even tell you that yet.

 

[looncraz]

How fast is Zen expected to be with 8 cores with hyperthreading? Sandybridge fast means it's quite interesting. Slower than that, not so much....

For single threaded IPC, 40% over Excavator is about 0.5% over Haswell.

It should beat an i5 easily (AMD usually doesn't do much differentiation based on enabled features, so I expect all of their CPUs will have SMT enabled - except, maybe, Athlons and Semprons when/if those arrive with Zen cores).

The real chore for Zen is to match or beat the Haswell i7s in single threaded, and maybe catch up with Skylake in multithreaded loads (AMD's usually scale better than Intel, and if Zen's data fabric and cache layout are as claimed, this should be no exception).

That comes down to its SMT, though, or willingness to throw more cores into the fray. I'd be very surprised if AMD's SMT was equal to Intel's SMT, but they have probably learned a great deal from the CMT design and may well have kept many elements of Bulldozer's front-end to leverage for SMT. With the instructions being fed and retired single-file they would lose most of the module-induced pipeline stages, so I'd expect a 14~17 cycle pipeline on Zen, much like Sandy Bridge.

 

[svenge]

How fast is Zen expected to be with 8 cores with hyperthreading? Sandybridge fast means it's quite interesting. Slower than that, not so much....

AMD cant even tell you that yet.

And even if AMD would tell you, their prior history indicates that one ought to discount the answer by about 30% or so.

 

[tential]

For single threaded IPC, 40% over Excavator is about 0.5% over Haswell.

It should beat an i5 easily (AMD usually doesn't do much differentiation based on enabled features, so I expect all of their CPUs will have SMT enabled - except, maybe, Athlons and Semprons when/if those arrive with Zen cores).

The real chore for Zen is to match or beat the Haswell i7s in single threaded, and maybe catch up with Skylake in multithreaded loads (AMD's usually scale better than Intel, and if Zen's data fabric and cache layout are as claimed, this should be no exception).

That comes down to its SMT, though, or willingness to throw more cores into the fray. I'd be very surprised if AMD's SMT was equal to Intel's SMT, but they have probably learned a great deal from the CMT design and may well have kept many elements of Bulldozer's front-end to leverage for SMT. With the instructions being fed and retired single-file they would lose most of the module-induced pipeline stages, so I'd expect a 14~17 cycle pipeline on Zen, much like Sandy Bridge.

Depends on pricing. With an 8 core 16 thread Zen chip, I might use that over a 6 core intel chip if that 8 core zen chip is far cheaper and gets the job done for a multi person gaming rig. But I'm doubting it, but I'll take a look at it when it comes out obviously.

 

[Dresdenboy]

BTW, what is the base of the 40% increase? Average, lowest, or highest XV IPC in many or some benchmarks? Or SPEC CPU as usual?

Doing two 256b AVX ops per cycle vs. 128b doesn't increase IPC (technically).

 

[jhu]

BTW, what is the base of the 40% increase? Average, lowest, or highest XV IPC in many or some benchmarks? Or SPEC CPU as usual?

Doing two 256b AVX ops per cycle vs. 128b doesn't increase IPC (technically).

Same IPC but double FLOPs.

 

[Dresdenboy]

Same IPC but double FLOPs.

Of course. Which leads to the funny situation, that higher CB scores are touted as higher IPC, while with going from 128b to 256b SIMD the IPC actually went down.

 

[ShintaiDK]

BTW, what is the base of the 40% increase? Average, lowest, or highest XV IPC in many or some benchmarks? Or SPEC CPU as usual?

Nobody knows. It could for the sake of argument just be SMT. The sole point its so unspecified raises the red flag.