• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

Richland & Kabini rumours

Page 33 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Status
Not open for further replies.
Hutchinman said:
It seems that 28nm production for Kabini and Temash will be at TSMC. Global Foundries couldn't ramp up production/fix their process problems fast enough.
Click to expand...

Kabini was stated to launch on TSMC 28nm process and that was going as planned. This was due to a waiver on the WSA where GLF allowed AMD to manufacture Brazos/Kabini on TSMC. This waiver was killed in December, when the WSA received its last amendment.

I think we can agree that regardless of whatever problems GLF is having with their 28nm process, there wouldn't be enough time to transpose Kabini to GLF and launch the chip in Q1 even if GLF had a working process by now.
 
mrmt said:
Kabini was stated to launch on TSMC 28nm process and that was going as planned. This was due to a waiver on the WSA where GLF allowed AMD to manufacture Brazos/Kabini on TSMC. This waiver was killed in December, when the WSA received its last amendment.

I think we can agree that regardless of whatever problems GLF is having with their 28nm process, there wouldn't be enough time to transpose Kabini to GLF and launch the chip in Q1 even if GLF had a working process by now.
Click to expand...
And these 28nm ramping issues GloFo has are probably the reason for Kaveri's launch delay of 1 quarter.
 
mrmt said:
Kabini was stated to launch on TSMC 28nm process and that was going as planned. This was due to a waiver on the WSA where GLF allowed AMD to manufacture Brazos/Kabini on TSMC. This waiver was killed in December, when the WSA received its last amendment.

I think we can agree that regardless of whatever problems GLF is having with their 28nm process, there wouldn't be enough time to transpose Kabini to GLF and launch the chip in Q1 even if GLF had a working process by now.
Click to expand...

Yes but what about Temash? Even if Kabini got the go ahead, IIRC the waiver was only for Brazos, not Temash.

Does that mean GLF is giving AMD a break, or is a storm based off the WSA coming?
 
Hutchinman said:
Yes but what about Temash? Even if Kabini got the go ahead, IIRC the waiver was only for Brazos, not Temash.

Does that mean GLF is giving AMD a break, or is a storm based off the WSA coming?
Click to expand...

Isn't Temash the same chip as Kabini?

They just downclock and and use the most choice dies for Temash.
 
nehalem256 said:
Isn't Temash the same chip as Kabini?

They just downclock and and use the most choice dies for Temash.
Click to expand...

I believe they use the same Jaguar cores, lower TDP, lower clock, lower core count.

Therefore they are not the same chip and I don't believe that AMD asked for an out for Temash.

So either GF gave AMD a pass or something else is going on.
 
Hutchinman said:
I believe they use the same Jaguar cores, lower TDP, lower clock, lower core count.

Therefore they are not the same chip and I don't believe that AMD asked for an out for Temash.

So either GF gave AMD a pass or something else is going on.
Click to expand...


there are quad core temash chips. people are expecting a different on SOC south bridge for all the I/O just like how it is now with hondo. unless they have come up with a better method to completely fuse the extra I/O so it doesn't cost any power.
 
Hutchinman said:
Yes but what about Temash? Even if Kabini got the go ahead, IIRC the waiver was only for Brazos, not Temash.

Does that mean GLF is giving AMD a break, or is a storm based off the WSA coming?
Click to expand...

Both Kabini and Temash have taped out on TSMC 28nm, not on GLF 28nm, and the tape out occurred well before the WSA amendment, which means that the original plan was for both chips to launch on TSMC process and the waiver did include them.

GLF is NOT giving AMD a break, as they simply killed the waiver in the last WSA revision, kept the take-or-pay clauses that hit AMD this quarter *and* collected the breach fine in full.

As for a storm based off WSA, it is clear by now that AMD management wanted to cut their commitments to GLF and increase the TSMC relationship, meaning that, best case scenario, TSMC 28nm process is better than GLF, and worse case scenario, GLF 28nm sucks big time. As AMD is very shady on its disclosures to investors, we'll have to wait and see. My bet? GLF 28nm will be very late and will have worse performance than TSMC 28nm, but not an outright disaster.
 
mrmt said:
Both Kabini and Temash have taped out on TSMC 28nm, not on GLF 28nm, and the tape out occurred well before the WSA amendment, which means that the original plan was for both chips to launch on TSMC process and the waiver did include them.

GLF is NOT giving AMD a break, as they simply killed the waiver in the last WSA revision, kept the take-or-pay clauses that hit AMD this quarter *and* collected the breach fine in full.

As for a storm based off WSA, it is clear by now that AMD management wanted to cut their commitments to GLF and increase the TSMC relationship, meaning that, best case scenario, TSMC 28nm process is better than GLF, and worse case scenario, GLF 28nm sucks big time. As AMD is very shady on its disclosures to investors, we'll have to wait and see. My bet? GLF 28nm will be very late and will have worse performance than TSMC 28nm, but not an outright disaster.
Click to expand...

I am glad that the chips are going to TSMC, the way Global Foundries has been treating AMD the last couple of years is downright tragic.

Besides, as you've said, Global Foundries has had a terrible track record with its processes recently. I do believe that there is indeed something wrong with their 28nm process. I don't know for sure if it is a disaster, but my gut agrees with you that something is seriously wrong. One doesn't delay if the problem is minor, you don't tell anyone and take orders as usual, then by the time you actually start production, the problems have been fixed. However, to be sure we will have to wait and see.
 
Last edited:
mrmt said:
GLF 28nm will be very late and will have worse performance than TSMC 28nm, but not an outright disaster.
Click to expand...

Just for kicks I went through the GLF web site. They still say they are going to deliver customer 20nm wafers in 2012.

Maybe they should update that 😀
 
mrmt said:
Both Kabini and Temash have taped out on TSMC 28nm, not on GLF 28nm, and the tape out occurred well before the WSA amendment, which means that the original plan was for both chips to launch on TSMC process and the waiver did include them.

GLF is NOT giving AMD a break, as they simply killed the waiver in the last WSA revision, kept the take-or-pay clauses that hit AMD this quarter *and* collected the breach fine in full.
Click to expand...

I seem to recall GloFo boasting of increased revenues recently - if it was due to payments from AMD, then ho hum. ATIC, which owns GloFo is socking it to AMD through GloFo whilst propping up AMD by infusing some cash and getting some preferred shares. What a bizarre game of whack-a-mole!

I hope, for AMD's sake, GF 28nm is good in time for Kaveri and yielding high clocks for SR in 2014 (which probably should have been on a 20nm process but apparently that isn't going to happen either.).
 
Hutchinman said:
Yes but what about Temash? Even if Kabini got the go ahead, IIRC the waiver was only for Brazos, not Temash.

Does that mean GLF is giving AMD a break, or is a storm based off the WSA coming?
Click to expand...

Historically, AMD never goes more than 2 quarters without amending the WSA or the accounting of the Gloflo transaction...🙄
 
Ajay said:
I seem to recall GloFo boasting of increased revenues recently - if it was due to payments from AMD, then ho hum. ATIC, which owns GloFo is socking it to AMD through GloFo whilst propping up AMD by infusing some cash and getting some preferred shares. What a bizarre game of whack-a-mole!

I hope, for AMD's sake, GF 28nm is good in time for Kaveri and yielding high clocks for SR in 2014 (which probably should have been on a 20nm process but apparently that isn't going to happen either.).
Click to expand...

The fact that AMD didn't have a stipulation that GF have functioning processes to use as part of the WSA is basically criminal.
 
Ajay said:
I seem to recall GloFo boasting of increased revenues recently - if it was due to payments from AMD, then ho hum.
Click to expand...

GLF is a closed company, which means no quarterly balance published, but this increase is related to a 34% increase that was floated around by some sites last year. GLF's 2011 revenue is estimated in 4 billion, with the increase they are almost going for 5.5 billion.

An increase of such a magnitude cannot come form the WSA amendment. In fact, while the penalty imposed to AMD is good for GLF results, one of the result of the penalty is to reduce forecast revenue.
 
Hutchinman said:
I am glad that the chips are going to TSMC, the way Global Foundries has been treating AMD the last couple of years is downright tragic.
Click to expand...

Yes, but according to Kumar all APU's are going to GLF. As for Kabini, I don't know how many time AMD plans to keep it on the market. If it's two years like Brazos, we're probably going to see Kabini at 28nm GLF rather sooner than later, if AMD is going for a shorter life cycle, then it's Kabini successor that will go on 28nm GLF. I'm more inclined to the former alternative, as there is no way AMD can fulfill their purchase commitments to GLF without bringing Kabini real soon to GLF.

IDC made some nice posts on how GLF is mistreating AMD and I tend to agree with him. Keep in mind that GLF did what they did when AMD answered for 35-45% of their entire revenues. If we take that 5.5 billion number and add a small increase, AMD share on GLF revenues will be less than 20% and that with a more stringent WSA. It will be a rocky ride.
 
itsmydamnation said:
remember there were 28nm shrinks of bobcat that were cancelled so it would have been 1 year or so per core.
Click to expand...

AMD never tick-tock'ed in their entire history, and when they decided to do so they failed badly, either fielding poor designs or not sorting out foundry issues. In practice, AMD isn't delivering a yearly cadence and I don't think they have the means and the money to go that route.
 
mrmt said:
AMD never tick-tock'ed in their entire history, and when they decided to do so they failed badly, either fielding poor designs or not sorting out foundry issues. In practice, AMD isn't delivering a yearly cadence and I don't think they have the means and the money to go that route.
Click to expand...


So you make up stuff vs actual fact...... the issue was the late delivery of process from global foundries would have given Krishna a very shot lifetime . In "practice" lets look at what AMD is/ has delivering :

Ontario jan 2011
llano june 2011
bulldozer oct 2011
trinity may 2012
vishera oct 2012
hondo oct 2012
richland Dec 2012
Kaveri H2 2013
Kabini/Temash Q1-2 2013

xbox/ps Q4-2013+ most rumors are amd CPU/GPU , especially with lines like
"Our semi-custom APUs already have a number of confidential high-volume design wins in place" from rory.

Now it might just be me but the only thing not on a yearly refresh is in fact Ontario/Kabini because Krishna was canceled. AMD has to move to a fast iteration model, it removes risk for AMD. They are going to/are increase/ing the use of automation in chip design to allow them to sustain the amount of required design throughput.

Bulldozer was a disaster but to me it looks like AMD is evolving we just have to wait for SR in H2 2013 to see what kind of ground can be made in the "big core" area.
 
Also there is a rumor now about 2014 successor to Kabini/Temash(from FUD who somehow apparently had seen latest mobile roadmap). It's called Beema or Bhima and is based on Jaguar+ core on 20nm(?) node .
 
itsmydamnation said:
So you make up stuff vs actual fact...... the issue was the late delivery of process from global foundries would have given Krishna a very shot lifetime . In "practice" lets look at what AMD is/ has delivering :

Ontario jan 2011
llano june 2011
bulldozer oct 2011
trinity may 2012
vishera oct 2012
hondo oct 2012
richland Dec 2012
Kaveri H2 2013
Kabini/Temash Q1-2 2013
Click to expand...

Trinity isn't Llano evolution, it only existed because the 45nm Bulldoze was canned and it took a lot of tweaks after the 32nm Bulldozer was launched to bring it to mobile power levels. Hondo is a variant of Ontario, not an all new design, and Richland is a rehashed trinity, not an all new design. Once you exclude that you are left with Bulldozer => Piledriver, a descent upgrade but only because the former chip was an outright disaster. So no tick-toc here.

Same with process. AMD started with 32nm in 2011, and wouldn't switch to 28nm if the costs weren't so compelling. If they move to 20nm in 2014 that's three years in the same node, not tick-tock here too, and I'm not even talking about the half-baked 14nmXM, which is NOT a full-fledged 14nm process, so no tick-tock here too.

You could just look at AMD chips on the price ladder. The reason on why AMD pricing is going south (and their revenues/market share/gross margin) is a simple one: Intel is delivering more than AMD in less time.
 
inf64 said:
Also there is a rumor now about 2014 successor to Kabini/Temash(from FUD who somehow apparently had seen latest mobile roadmap). It's called Beema or Bhima and is based on Jaguar+ core on 20nm(?) node .
Click to expand...

I mentioned that last page. I was surprised no one commented on that. I was thinking no one could resist speculation.
 
One question, If Samsung has its own fab, seems they do? Why don't AMD work with them instead? Cause Samsung's Exynos Roadmap shows the SoC getting smaller and smaller. I'm guessing they are like the Intel of ARM SoC's.

*Completely clueless*
 
Status
Not open for further replies.

TRENDING THREADS

Back
Top