News & Rumors: Xbox One (codename Durango)

[liolio]

AMD has cancelled the products using 20nm which gets me even more iffy about the odds of a shrink for either the XB1 or the PS4. AMD has no plan to port any of the IPs used in Durango to 14nm process.
With all the complicated virtualized wizardry MSFT has injected in the XB1 I really hope it allows them a serious redesign that is the only sensible way they seem to have toward serious price reduction.

 

[Rikimaru]

AMD has no plan to port any of the IPs used in Durango to 14nm process..

This does not matter.
Both consoles will have 14nm shrinks almost certainly. It benefits consoles more than PC. Simplifies cooling, PSU, power circutry, size and of course size of the APU.

Three Xbox 360 chips even got redesigned to single one.

 

[liolio]

This does not matter.
Both consoles will have 14nm shrinks almost certainly. It benefits consoles more than PC. Simplifies cooling, PSU, power circutry, size and of course size of the APU.

Three Xbox 360 chips even got redesigned to single one.

You don't know, it is a statement based on history when shrink happened every 18 month, it was not that costly compared to the gains. We are past that curve, actually I expect lots of phones/tablets SOC to remain at 28nm because late version of those process are offering good enough performances for the low/mid range segment.

 

[Rikimaru]

We are past that curve, actually I expect lots of phones/tablets SOC to remain at 28nm because late version of those process are offering good enough performances for the low/mid range segment.

Because these chips are small even at 28nm. Console APUs are rather big.

 

[liolio]

Because these chips are small even at 28nm. Console APUs are rather big.

I suspect that transition to 14nm will be slow one for anything but high end and it will not be made on the basis of chips/wafers (and related cost saving) but power efficiency (/sustained performance). Yields are better for tiny chips sure.
If Jaguar are ported to 14nm why not use them? It is not like AMD could not use ~50mm2 (supposedly) cheap SOC in its line up.
I don't know what the deal is between MSFT, SOny and AMD but as AMD is killing CPU lines, is backing off from using newer lithography I wonder if it could have legal implication down the road with AMD no longer in a situation to deliver on engagements it could have take with consoles manufacturers...

EDIT
That attitude vs taken for "granted" shrink is remanent to me to discussion about next years ago where people dismissed again and again any consideration for power usage and were hoping for incredibly high throughput (in GFLOPS). It is also something that is easily noticed in financial markets (though here we are past dementia level and I'm quite anxious thinking I will have to live through the unwinding of that madness...).
Be it "this time is different" or "business as usual" the opposition in those mantra is a false one, relate to our strong tendency to denial.

 

[Rikimaru]

If Jaguar are ported to 14nm why not use them? It is not like AMD could not use ~50mm2 (supposedly) cheap SOC in its line up.

AMD already has better Puma+ cores. They do not need Jaguars for consumer SoCs anymore.

This year Carrizo-L have Puma+. Usually consumer CPUs are not just die shrinks.

 

[fehu]

I hope to use the right terms
being jaguar and gcn fully synthetized designs, doesn't amd "just" need to pass them on a 14nm library?

 

[Deleted member 11852]

I suspect that transition to 14nm will be slow one for anything but high end and it will not be made on the basis of chips/wafers (and related cost saving) but power efficiency (/sustained performance). Yields are better for tiny chips sure.

High end or high volume. If you're cranking out millions of chips a week then a shrink will save so much money (assuming yields don't nosedive) because of the savings on wafers. This would be Apple volume though, not console volume.

 

[upnorthsox]

It's a complete redesign to go to finfet. Unless it was already part of your contract with AMD I don't think you can assume that it's more economical to go to 14nm.

 

[Rikimaru]

I remember AMD already talked about console shrinks at investors QA.
http://www.guru3d.com/news-story/amd-will-give-consoles-a-die-shrink-first.html

I think 20nm one is cancelled. But will do 14nm.

 

[upnorthsox]

To be clear, that article is saying that AMD will benefit from any chip saving on the shrink. MS and Sony would only save from associated items like cooling and power supply savings.

That's also talking about 20nm which doesn't require a finfet design so I wouldn't assume it also applies to 14nm.

 

[liolio]

AMD already has better Puma+ cores. They do not need Jaguars for consumer SoCs anymore.

This year Carrizo-L have Puma+. Usually consumer CPUs are not just die shrinks.

Excuse my inaccuracy as I do not differentiate jaguar and Puma or kabini/Temash and Mullins/Beema, mostly the same CPU with functional turbo on what seems to be a slightly better process (GF). They are more or less the same and commercial success goes on showing that so I use jaguar in a loose manner.
With regard to shrink you mean like Intel TIC-TOC, one step was a shrink. AMD was also doing lots of shrink for its CPU with minimal change to its architecture. Now even Intel can't keep up with that pace, you guys think that it would be trivial from AMD to port complex IPs (they won't use on top of it) to 14nm. GPU were shrink but usually with minimal change (like memory bus width), anyway only consoles are completely static designs. (Half nodes shrinks are long gone too.)

Anyway we will see soon enough but my personal bet is that a lot has happened since both MSFT and Sony contracted AMD and consoles designs were decided upon: bad 20nm process (delay and then cancelled products), end of the Bobcat/ Jaguar/puma line, bad results, few advances architectural advances to GCN (compare to the pace of competition), massive layoffs, those buyout rumors. the pace of progress wrt lithography has slown down to crawl, price reduction are no longer a given, then throwing AMD situation into the mix... I've a tough time being bullish.

 

[3dilettante]

It's a complete redesign to go to finfet. Unless it was already part of your contract with AMD I don't think you can assume that it's more economical to go to 14nm.

It should be a very heavy or possibly near-total redesign of the circuit implementation, but the architectural features could be maintained. There's still an opportunity for an eventual shrink. At least power-wise these chips are not where the consoles would be happiest.
If they are willing to pay, AMD would be willing to look at the money.

To be clear, that article is saying that AMD will benefit from any chip saving on the shrink. MS and Sony would only save from associated items like cooling and power supply savings.

It says there are manufacturing cost improvements, and that the console makes would see additional benefits in terms of material costs surrounding the chip. I am skeptical that the console makers left a loophole like the savings from manufacturing transitions solely to AMD.

 

[liolio]

I remember AMD already talked about console shrinks at investors QA.
http://www.guru3d.com/news-story/amd-will-give-consoles-a-die-shrink-first.html

I think 20nm one is cancelled. But will do 14nm.

Hear of that the sources are not linked properly in the news. Overall it sounds extremely weird and unprofessional, pointing to competing products from Nvidia, giving details about competing architectures, etc. things I expect no real, serious AMs representative to do in such a direct manner.
The text is in italic as in direct quotation of AMD rep but it is not. You track back the article at techpowerup to realize that it is a completely unsubstantiated claim as far as I can tell (or read...).

 

[upnorthsox]

It should be a very heavy or possibly near-total redesign of the circuit implementation, but the architectural features could be maintained. There's still an opportunity for an eventual shrink. At least power-wise these chips are not where the consoles would be happiest.
If they are willing to pay, AMD would be willing to look at the money.


It says there are manufacturing cost improvements, and that the console makes would see additional benefits in terms of material costs surrounding the chip. I am skeptical that the console makers left a loophole like the savings from manufacturing transitions solely to AMD.

My guess would be that the contracts are discounted based on quantity delivered/sold. Thus why advantageous for AMD to reduce costs over time and where MS/Sony can gain savings.

 

[liolio]

Back to my redesign idea, clearly if MSFT can get some 360 games to work on the One, having the One working on leaner design built mostly of the same IPs should be almost trivial to them.
No matter which process they use the the XB1 is not in the sweet spot of efficiency for AMD products (perfs per watts and mm2) comparing the R7 250 to the HD 7750 shows that really well. AMD reaches higher performances per watts when clocks are pushed up, ceteris paribus it burns more power. The thing is it is not a ceteris paribus comparison as it is a redesign and AMD made progress since the release of the PS4 and the Xb1, especially in the turbo department.

 

[3dilettante]

My guess would be that the contracts are discounted based on quantity delivered/sold. Thus why advantageous for AMD to reduce costs over time and where MS/Sony can gain savings.

There are cost schedules attached to the existing APUs. AMD hinted at this earlier when they realized slightly better margins for the semi-custom division.
The semi-custom division's reason for existing is to get customers to pay for the NRE, so a node transition when the partner is not paying for it or has not contracted for it is a departure.
A FinFET device would be an imposition on the console maker as well, as it would most likely not be electrically compatible and possibly mechanically different as a result of that. AMD didn't run much of the validation, as news articles on Microsoft's testing lab showed, and AMD doesn't control the manufacturing of anything else. At least with SHAPE, there may also be IP a console maker has that it might not grant free reign for AMD to re-implement.
It seems like a lot can go wrong here, and what stops the console makers from insisting on the 28nm chips they can use?

 

[upnorthsox]

There are cost schedules attached to the existing APUs. AMD hinted at this earlier when they realized slightly better margins for the semi-custom division.
The semi-custom division's reason for existing is to get customers to pay for the NRE, so a node transition when the partner is not paying for it or has not contracted for it is a departure.
A FinFET device would be an imposition on the console maker as well, as it would most likely not be electrically compatible and possibly mechanically different as a result of that. AMD didn't run much of the validation, as news articles on Microsoft's testing lab showed, and AMD doesn't control the manufacturing of anything else. At least with SHAPE, there may also be IP a console maker has that it might not grant free reign for AMD to re-implement.
It seems like a lot can go wrong here, and what stops the console makers from insisting on the 28nm chips they can use?

Indeed, all good points. I would have to say the prospects of a slim model even if they are not seeing much if any savings from a reduced chip itself.

 

[liolio]

AMD already has better Puma+ cores. They do not need Jaguars for consumer SoCs anymore.

This year Carrizo-L have Puma+.

Oops I'm an idiot who post too much indeed, I missed Carrizo-L unveiling, my bad.
I does not change my pov on shrink, it may happen but it doesn't match AMD announcements.

 

[Cyan]

Rare seem to be slowly but steadily making a comeback, as this video shows.