The Kaveri Pre-Launch Thread (A10-7800 and A10-6800k @3,5 Ghz)

[AtenRa]

Apples to Oranges. If AMD had prioritized retail chips for trinity or Richland could they have had product on shelves a few months earlier? Probably.

I predict Kaveri will be ramped and readily available in OEM machines around the middle of the year, similar to Richland. If you remember, Richland was already shipping during CES 2013. So kaveri is in fact shipping a full year after Richland.

You confusing mobile Richland that was shipping in CES 2013 and released in March with Desktop Richland that was released in June and was immediately available in retail, much like Kaveri now.

 

[Abwx]

Because that's the way the business works.

Perhaps it works like this in your imagination but in real world
big manufacturers are top on the list, they just dont want
to have a products portfolio that is inferior to local builders.

Edit ; like this one

http://www.ldlc.com/fiche/PB00159637.html

Dont think that the HPs and other Acer will stick to previous APUs while shops will upgrade to the new offerings.

 

[erunion]

You confusing mobile Richland that was shipping in CES 2013 and released in March with Desktop Richland that was released in June and was immediately available in retail, much like Kaveri now.

I was comparing first shipping Richland versus first shipping Kaveri.

But also fully ramped versus fully ramped.

 

[Enigmoid]

You confusing mobile Richland that was shipping in CES 2013 and released in March with Desktop Richland that was released in June and was immediately available in retail, much like Kaveri now.

Richland was the same physical die as trinity. Kaveri is completely different. I would not be surprised if there was difference in availability and OEM shipments.

 

[jj109]

No MPU company worth its money would accept the kind of margins AMD is getting on the console contracts.

Considering that AMD is getting $100 for something larger than the Tahiti die, what's the margin per die? $50? $20? It's a mystery to me.

Say that they ship another 10 million units through 2014, it's still enough to pay off the WSA penalty barring ridiculously low margins.

 

[norseamd]

how fast would a 7850k go with water cooling taking heat out of the equation. that is why it is clocked lower than richland right?

what is the future of fm2+? how fast do you think carrizo will be?

 

[AtenRa]

Richland was the same physical die as trinity. Kaveri is completely different. I would not be surprised if there was difference in availability and OEM shipments.

Kaveri is shipping to OEMs since Q4 2013, it will be immediate available after the official release on the 14th of January.

 

[mrmt]

Considering that AMD is getting $100 for something larger than the Tahiti die, what's the margin per die? $50? $20? It's a mystery to me.

Assuming the $100 number correct, $15-$17 per die, which puts the costs (and GLF price tag) in $83-$85 in the case of the PS4 SoC, and $91-$93 in the case of the XBO

Say that they ship another 10 million units through 2014, it's still enough to pay off the WSA penalty barring ridiculously low margins.

Let's say that console sales will stay way above the past levels and that AMD will be able to fulfill the WSA quota with consoles only. At 17% operating margins they will be getting ~200MM in operating profits, per year, or a paltry 50MM per quarter. In terms of operating profits that's less what Nvidia has been getting with Tegra, which everyone + dog considers a failure.

When I say that the deal would only make sense for AMD, it's because everyone else is aiming at much larger cash streams. Only AMD with its WSA and it's poor financial situation would grab a deal as cheap as this one.

 

[Abwx]

Assuming the $100 number correct, $15-$17 per die, which puts the costs (and GLF price tag) in $83-$85 in the case of the PS4 SoC, and $91-$93 in the case of the XBO

A waffer cost at most 4000$ and is 706 cm2 from wich about
140 chips can be extracted wich equate to barely 30$/chip,
adding tests and BGA packaging would hardly increase
the bill over 35$ , so for a 100$ price gross margin is 65$/chip ,
i dont see from where you re extracting your flawed number
wich are completely wrong but hey , it s not about being accurate...

 

[AtenRa]

It will be less than $4000 for a 2,5 year old and depreciated 28nm TSMC process

 

[Abwx]

It will be less than $4000 for a 2,5 year old and depreciated 28nm TSMC process

Could be as "low" as 3500$ but it wouldnt change the result
by more than a 2-3$ out of the 65$ gross margin per chip,
what mmrt actualy "estimated" was the net profit per chip
wich is believed to be about 20% of the unknown selling
price , 100$ in this exemple.

 

[ams23]

A waffer cost at most 4000$ and is 706 cm2 from wich about
140 chips can be extracted wich equate to barely 30$/chip,
adding tests and BGA packaging would hardly increase
the bill over 35$ , so for a 100$ price gross margin is 65$/chip ,
i dont see from where you re extracting your flawed number
wich are completely wrong but hey , it s not about being accurate...

I don’t think you have any clue what you are talking about. The Xbox One APU alone has > 5 billion transistors! And the actual data needed to compute margins on these custom console APU's is unknown to the general public. In all likelihood, the gross margins on these custom console APU's will be very poor relatively speaking. We can gain more insight into this by looking at quarterly trends in corporate gross margins to see if there is any noticeable new business drag or lift.

 

[mrmt]

I don’t think you have any clue what you are talking about. The Xbox One APU alone has > 5 billion transistors! And the actual data needed to compute margins on these custom console APU's is unknown to the general public. In all likelihood, the gross margins on these custom console APU's will be very poor relatively speaking.

AMD gave a range on the operating margins in two of their Q&A, and given that there's basically no other SG&A, operating margins will be really close to gross margins.

 

[inf64]

Looks like there will be actual systems built around Kaveri since pre-order started from amd's select partners:
http://community.amd.com/community/.../07/pre-order-amd-s-next-generation-apu-today

 

[Spawne32]

Looks like there will be actual systems built around Kaveri since pre-order started from amd's select partners:
http://community.amd.com/community/.../07/pre-order-amd-s-next-generation-apu-today

So it begs the question once again, why would puget systems be selling kaveri, but release a review about it basically saying it has piss poor performance right before release and hurt their own sales of the product? That makes no sense.

 

[mrmt]

So it begs the question once again, why would puget systems be selling kaveri, but release a review about it basically saying it has piss poor performance right before release and hurt their own sales of the product? That makes no sense.

Because they don't want to mislead their customers.

 

[Homeles]

So it begs the question once again, why would puget systems be selling kaveri, but release a review about it basically saying it has piss poor performance right before release and hurt their own sales of the product? That makes no sense.

http://begthequestion.info/

how fast would a 7850k go with water cooling taking heat out of the equation. that is why it is clocked lower than richland right?

what is the future of fm2+? how fast do you think carrizo will be?

It's clocked lower than Richland because it's on an inferior process, in regards to switching speeds. It's not a heat thing; it's power. At the same power levels, their 32nm process is faster than the 28nm process they're using for Kaveri. This is because they moved from PDSOI to bulk transistors.

As far as the future of FM2+ goes, there's a fairly good chance that Carrizo will require a new socket. AMD may choose to further integrate components on die or move them on package. There's also the possibility of DDR4. Both of these would break socket compatibility.

As far as how fast it'll go, who knows. It'll be considerably faster in regards to floating point workloads (~2x) as AMD is expanding the floating point unit width to 256bits (or 128x2). As far as integer workloads go, it will have double the ALUs and AGUs, and double the L1 instruction cache. Integer performance should increase a fair bit as a result.

There's a chance that clock speeds further regress from Kaveri, though. AMD is moving to high density cell libraries with Excavator, and increasingly implementing automation, which comes at the cost of switching speed.

At this point, the main thing holding AMD back is GloFo's 28nm process.

 

[inf64]

Unfortunately I think Carrizo is stuck with this process node .

 

[NostaSeronx]

Unfortunately I think Carrizo is stuck with this process node .

Carrizo/Toronto can always be KV-B0.

At the same power levels, their 32nm process is faster than the 28nm process they're using for Kaveri. This is because they moved from PDSOI to bulk transistors.

28nm SHP isn't bulk by GlobalFoundries Standards.

ASTA Process;
90nm SHP = SOI
65nm SHP = SOI
45nm SHP = SOI
32nm SHP = SOI
ISDA Process;
28nm SHP = SOI
20nm SHP = SOI
14nm SHP = SOI
10nm SHP = SOI

But, in the same slides they give Kaveri 64 ROPs with 2 RBEs. So, they could possible meant LPS or LPH or HPP.

http://i.imgur.com/346erhz.png
^-- AMD and Chartered is GlobalFoundries.

 

[ShintaiDK]

Unfortunately I think Carrizo is stuck with this process node .

Its gonna get even worse at 20nm. I have a feeling it will be a 4 year cycle there. Specially since they will do a staged release, first without finfets, then with finfets that they call 14/16nm for marketing reasons.

Its bad for AMDs CPUs. But its even worse for the discrete GPUs. Not to mention it might be 20nm (with finfets) vs 7nm or 10nm. Utterly hopeless.

 

[norseamd]

can amd use tsmc for cpus after kaveri?

 

[NostaSeronx]

can amd use tsmc for cpus after kaveri?

No, they are locked with GlobalFoundries till 2024. (Or was it 2022?)

 

[ShintaiDK]

2024.

 

[norseamd]

that is rediculus. why? seems like a bad spin off deal to me.

 

[Centauri]

It was. AMD's books would look a lot better without the enormous WSA payments they've had to make.