AMD Carrizo Pre-release thread

[AtenRa]

First of all even the GT3 Broadwell is only 133mm2. Secondly...not really a match CPU wise

And we still have to see Carrizo on the benchmark front. Those 15W chips may throttle more than not.

It is 133mm2 + PCH dont forget that and not really a match iGPU wise

Also dont forget how much more expensive Intel 14nm FF is and with lower yields than 28nm.

 

[ShintaiDK]

It is 133mm2 + PCH dont forget that and not really a match iGPU wise

Also dont forget how much more expensive Intel 14nm FF is and with lower yields than 28nm.

The PCH is what, 45mm2 on a 32nm process without the use of HDL?

Lower yields? I assume you got some papers ready on that?

And I think we already estabished countless times that graphics performance isnt really a seller. You should know that quarter after quarter with one disaster after the other for AMD.

AMD had a chance of bringing the diesize and cost down, instead they blew it away and kept increasing it. And unless they raise the price on Carrizo contra kaveri. They also have to eat the FCH revenue.

 

[el etro]

Are my eyes seeing this correctly? The freaking NB has comparable die-area as the x86 processor space? Seriously?

When your SKU's ASP is predominately determined by the performance of something that occupies <20% of your silicon outlay, you should feel reasonably confident in proclaiming that somewhere in your journey to today the wheels very much came off your wagon :\

CMT effect.

 

[coercitiv]

Yes, and prices havent been lower for desktops where the PCH isnt integrated that you took the quote out of context with.

But desktops have also integrated the GPU and your claim covered Core -> Haswell transition. How's that for context?

How did the price/die evolve there over time if you go back to the GPU less? You may notice that Intel gets payed for it.

No comment.

Platform wise, the U prices havent been lower either. Its just a matter of where you pay. Higher mobo cost or higher CPU.

And now you spin it towards platform costs, as if this does not apply to AMDs offerings

 

[ShintaiDK]

But desktops have also integrated the GPU and your claim covered Core -> Haswell transition. How's that for context?

Chipset prices stayed the same after the GPU was moved from chipset to the CPU.

No comment.

I assume you acknowledge you are wrong then.

And now you spin it towards platform costs, as if this does not apply to AMDs offerings

I have multiple times said that AMD needs to increase the price of Carrizo over Kaveri to offset the lost of the FCH revenue. Unless its AMD that ends up with the bill.

 

[AtenRa]

The PCH is what, 45mm2 on a 32nm process without the use of HDL?

133 + 45 = 178mm2 with DDR-3L 1600MHz only memory and inferior iGPU performance

Lower yields? I assume you got some papers ready on that?

I really dont have to say anything here, everyone knows how good Intels 14nm yields are.

And I think we already estabished countless times that graphics performance isnt really a seller. You should know that quarter after quarter with one disaster after the other for AMD.

A few post earlier I have quoted Lisa Su saying that ASP, Revenue and sales of Mobile APUs increased YoY and Q to Q. I know you dont like it but thats a fact. AMD APU sales will only get better with a product like Carrizo.

AMD had a chance of bringing the diesize and cost down, instead they blew it away and kept increasing it. And unless they raise the price on Carrizo contra kaveri. They also have to eat the FCH revenue.

The only ones tooting ONLY die sizes and forgetting everything else that add cost to a product is you and mrmt.
I know you dont like it but die size is not the only factor affecting the cost of your IC.

 

[USER8000]

AMD had a chance of bringing the diesize and cost down, instead they blew it away and kept increasing it. And unless they raise the price on Carrizo contra kaveri. They also have to eat the FCH revenue.

Kaveri was 245mm2 with a 49.2mm2 FCH. Carrizo is 250mm2 without the need for a FCH or the associated packaging or cooling required for the NB.

You are just argueing for the sake of argueing.

To everyone else,just let him argue with himself.

 

[ShintaiDK]

133 + 45 = 178mm2 with DDR-3L 1600MHz only memory and inferior iGPU performance

So 178 vs 250 and an advantage nobody wants to pay for with AMD? Super!

I really dont have to say anything here, everyone knows how good Intels 14nm yields are.

Why dont you share it then? You obviously know the yield on 133mm2 Broadwell dies and 250mm2 Carrizo dies. Or is it because you dont?

A few post earlier I have quoted Lisa Su saying that ASP, Revenue and sales of Mobile APUs increased YoY and Q to Q. I know you dont like it but thats a fact. AMD APU sales will only get better with a product like Carrizo.

Will it? I think the future is to decide that. And what happens with revenue and ASP if products are just moved from cat cores to big cores. Even if there is a decline in overall shipping numbers? Also how much does the mobile part really account for with AMD? Is the AM3 socket still accounting for 30% of the 350M$ revenue?

The only ones tooting ONLY die sizes and forgetting everything else that add cost to a product is you and mrmt.
I know you dont like it but die size is not the only factor affecting the cost of your IC.

So how does AMDs cost structure look like when it sells its products?

 

[ShintaiDK]

Kaveri was 245mm2 with a 49.28mm2 FCH. Carrizo is 250mm2 without the need for a FCH or the associated packaging.

With Kaveri AMD sold a CPU and they sold a FCH.

With Carrizo AMD only sells a CPU.

And the million dollar question is. Where did the FCH revenue and profit go?

There are 2 options as I have described several times.
Either AMD takes on the extra cost and loss of revenue.
Or AMD increases the price of Carrizo to offset the FCH loss.

Had AMD gone for a sub 200mm2 design, they could have gone out victorious nomatter the case. But they didnt.

 

[NTMBK]

With Kaveri AMD sold a CPU and they sold a FCH.

With Carrizo AMD only sells a CPU.

And the million dollar question is. Where did the FCH revenue and profit go?

There are 2 options as I have described several times.
Either AMD takes on the extra cost and loss of revenue.
Or AMD increases the price of Carrizo to offset the FCH loss.

Had AMD gone for a sub 200mm2 design, they could have gone out victorious nomatter the case. But they didnt.

My assumption is that they will charge the same for a Carrizo platform as they did for a Kaveri platform. Manufacturers don't buy just the CPU/APU, they buy the chipset to go with it.

 

[USER8000]

With Kaveri AMD sold a CPU and they sold a FCH.

With Carrizo AMD only sells a CPU.

And the million dollar question is. Where did the FCH revenue and profit go?

There are 2 options as I have described several times.
Either AMD takes on the extra cost and loss of revenue.
Or AMD increases the price of Carrizo to offset the FCH loss.

Had AMD gone for a sub 200mm2 design, they could have gone out victorious nomatter the case. But they didnt.

Emm,or they sell it for the same price they were targeting with Kaveri platform price and actually have to only produce one chip which is 2% larger in surface area,plus the mainboards and cooling will be less complex since there is no FCH to take into account. Kaveri as a platform will cost more than Carrizo for AMD to make and sell,and cost more for OEMs due to the more complex packaging and cooling required. Since AMD is stuck on an old node,price is important.

You do realise this is one of the reasons we have seen increasingly more of the northbridge and southbridge functionality moved onto the main CPU die - Intel,AMD and virtually every CPU maker out there is moving towards this or has already done so.

But,honestly going to let you and the others here argue it out as I can see me just going in circles here,and I already have got a headache just reading this thread now.

 

[Abwx]

Had AMD gone for a sub 200mm2 design, they could have gone out victorious nomatter the case. But they didnt.

They preferer to target the i3/i5/i7, the very segment where there s no possible contra revenues, and FYI Carrizo is not only better on GPU but it will surely have substancialy better perf/Watt than Broadwell on CPU loading, currently even a Beema is better in this register.

 

[Erenhardt]

With Kaveri AMD sold a CPU and they sold a FCH.

With Carrizo AMD only sells a CPU.

And the million dollar question is. Where did the FCH revenue and profit go?

There are 2 options as I have described several times.
Either AMD takes on the extra cost and loss of revenue.
Or AMD increases the price of Carrizo to offset the FCH loss.

Had AMD gone for a sub 200mm2 design, they could have gone out victorious nomatter the case. But they didnt.

I think amd will happily swallow a per unit revenue loss for the more units shipment and it the end, an increase in total revenue.

Tighten the screws on your armrests. LOL

 

[mrmt]

Will it? I think the future is to decide that. And what happens with revenue and ASP if products are just moved from cat cores to big cores. Even if there is a decline in overall shipping numbers? Also how much does the mobile part really account for with AMD? Is the AM3 socket still accounting for 30% of the 350M$ revenue?

The reason of the record commercial sales is that the channel sales imploded because of the inventory choke, so they have to sell chips on the commercial segment for even lower margins than they traditionally get, and the reason behind record ASP is because cat core sales imploded. I thought that it was very unprofessional from AMD to quote this in the Q&A because it is spinning a very bad fact like a good thing, when there's nothing good about it. More important, this is not something they can hide grouping and disagrouping numbers like they usually do. It's really basic stuff, really, any incompetent financial analyst is able to realize what I'm saying here.

But given the amount of times that this quote was spun as something positive on the web, it seems that there are people idiot enough to buy AMD spin at face value, and if these people are really AMD investors, AMD will be rewarded again for misleading its shareholders.

 

[raghu78]

With Kaveri AMD sold a CPU and they sold a FCH.

With Carrizo AMD only sells a CPU.

And the million dollar question is. Where did the FCH revenue and profit go?

There are 2 options as I have described several times.
Either AMD takes on the extra cost and loss of revenue.
Or AMD increases the price of Carrizo to offset the FCH loss.

Had AMD gone for a sub 200mm2 design, they could have gone out victorious nomatter the case. But they didnt.

We know how you like to turn everything into a negative. If AMD was selling Kaveri + FCH for a given price now they will sell a Carrizo for the same price or higher price assuming Carrizo increases ASP as per AMD expectations. So even at same price gross margins increase as you have to supply one chip instead of two. If ASP increases its even better

As far as the notebook OEM is concerned they are not bothered whether AMD ships them 1 or 2 chips. If AMD can ship a single chip with the functionality of APU+southbridge then thats all upside for AMD.

 

[ShintaiDK]

We know how you like to turn everything into a negative. If AMD was selling Kaveri + FCH for a given price now they will sell a Carrizo for the same price or higher price assuming Carrizo increases ASP as per AMD expectations. So even at same price gross margins increase as you have to supply one chip instead of two. If ASP increases its even better

As far as the notebook OEM is concerned they are not bothered whether AMD ships them 1 or 2 chips. If AMD can ship a single chip with the functionality of APU+southbridge then thats all upside for AMD.

Sorry I may have missed it. But where did AMD state that?

 

[DrMrLordX]

The problem with Kaveri is that they didn't sell nearly enough of them along with that FCH. So if sales increase by a non-trivial amount thanks to the reduced BOM attached to Carrizo implementations, the reduced price per unit will not be such a problem.

 

[ET]

People, please ignore ShintaiDK. This discussion adds nothing to this thread. It doesn't matter if you think he's wrong, he's trolling and you're feeding him. I composed a reply to him yesterday and ended up not posting it, and so can you. Let's keep the thread to facts and speculations about Carrizo.

 

[JDG1980]

DDR-4 confirmed and die picture

This looks like it could be a good product in the semi-premium notebook segment. They might also be able to sell NUC-style nettops (preferably fanless). What AMD needs to avoid is having all their chips dumped into cheap craptops that cut corners everywhere and provide a terrible experience. That has happened far too often until now.

The holy grail for AMD would be a design win on the next MacBook Pro. They wouldn't be able to sweep the board (single-threaded performance almost certainly won't be up to snuff with Intel's competing products), but maybe they could get their chips in the 13-inch version. Graphics performance is actually important on these because of the ultra-high-resolution screen, and Intel's IGPs have barely been able to keep up.

 

[jpiniero]

It's probably too late for Apple. The new Macbook Retina is the future of the laptop line there, and the Pro will be discontinued eventually. I imagine they will just refresh with Intel until they consolidate all the laptop lines into just the Macbook Retina.

 

[AtenRa]

AMD 14nm ZEN based APUs next year could be a viable solution for some Apple products, Carrizo is not.

 

[xthetenth]

It's probably too late for Apple. The new Macbook Retina is the future of the laptop line there, and the Pro will be discontinued eventually. I imagine they will just refresh with Intel until they consolidate all the laptop lines into just the Macbook Retina.

The pro and new macbook are very different products. One's a serious work capable machine, the other's an ultraportable analogue to the Surface coming from the opposite direction.

Refreshing with intel seems likely though.

 

[Arachnotronic]

From AMD Q1 2015 earnings call.

That's nonsense and I feel really bad for people who use this as a "positive" to make their investment decisions. You do realize that AMD's ASPs have gone up because pretty much all of their low-end business has dried up, right?

If I used to sell 50 items at $100 a piece and 50 items at $200 a piece, my ASP was $150 across 100 items.

If I suddenly stop selling those 50 items at $100 a piece and only sell 30 items at $200 a piece, my ASPs have magically gone up to $200. Yet my business is far worse off since I'm only selling 30 items.

 

[raghu78]

AMD 14nm ZEN based APUs next year could be a viable solution for some Apple products, Carrizo is not.

AMD needs to deliver on two things to get into a Macbook Pro -

1. competitive single thread performance in power constrained 35w notebook SKUs. I would consider a deficit of 10-15% in peak single thread performance against Intel Skylake core i5 /core i7 as competitive.
2. a HBM solution to provide discrete GPU performance without any bandwidth bottlenecks.

A quad core Zen APU with HBM will compete with a Nvidia discrete GPU of similar size (by which I mean the size of GPU portion of the APU which is around 100 - 120 sq mm). If the CPU portion is competitive in single thread against Intel core i5 then Apple would be seriously interested as they will save power when compared against a discrete CPU and GPU combo which would mean better battery life. HDL makes a lot of sense in these power constrained scenarios and if AMD Zen can hit a max frequency of 3.5 Ghz (for single thread) then they would perform well in all scenarios -ST, MT and CPU+GPU (games).

AMD needs to bring HBM onto its APUs and pitch the Zen based Bristol Ridge APUs to Apple for Macbook Pro. The APU is the perfect choice for power constrained devices like notebooks. Not having to drive power hungry communication through external PCI-E buses between CPU and GPU is a huge advantage in power savings which translates to better battery life. As far as Apple is concerned they can also pay less than they would pay to Intel and Nvidia together for two discrete components for comparable performance. This is the most attractive part to Apple as they would love to improve margins. AMD would not have any problem selling the APU at a lesser price than a Intel + Nvidia chip combination. The margins would still be outstanding if even they undercut the discrete solution by 20 - 25% percent.

 

[AtenRa]

That's nonsense and I feel really bad for people who use this as a "positive" to make their investment decisions. You do realize that AMD's ASPs have gone up because pretty much all of their low-end business has dried up, right?

If I used to sell 50 items at $100 a piece and 50 items at $200 a piece, my ASP was $150 across 100 items.

If I suddenly stop selling those 50 items at $100 a piece and only sell 30 items at $200 a piece, my ASPs have magically gone up to $200. Yet my business is far worse off since I'm only selling 30 items.

Lisa Su
Mobile APU ASPs and revenue increased from the year-ago period, highlighted by increases in commercial client APU shipments and revenue from the fourth quarter, setting a record for commercial client processor sales.

The context of the discussion was that ASP was up and also Mobile APU sales volumes increased which your paradigm lacks