Last month, we published a story on how AMD may have shifted its focus from Bulldo er to Bobcat in the wake of the former’s disappointing performance. Since it ran, we’ve spoken to multiple independent sources that told us AMD’s 28nm Bra os-based APUs were in serious trouble and would likely never see the light of day. AMD reportedly finali ed the decision to cancel both products right around the time it fired most of its PR and marketing divisions. Sunnyvale will likely announce a new set of 28nm APUs at its Financial Analyst Day in February. Unlike the failed Krishna/Wichita designs, the new chips will be manufactured by TSMC.
The implications and financial repercussions could be enormous. Moving 28nm APUs from GlobalFoundries to TSMC means scrapping the existing designs and laying out new parts using gate-last rather than gate-first manufacturing. AMD may try to mitigate the damage by doing a straightforward 28nm die shrink of existing Ontario/ acate products. While the resulting chips wouldn’t incorporate any of the architectural improvements originally intended for Krishna/Wichita, they’d lower Bra os’ power consumption and buy time for a new processor.
There’s no easy way to yank Krishna and Wichita out of AMD’s product lineup
If AMD goes this route, it might be able to launch a 28nm APU in 2012, but a true 28nm replacement for Krishna/Wichita is likely at least 18 months away. Without the GF-designed APUs, AMD will have no choice but to flog 40nm Bra os parts while Intel and ARM break into new markets and debut chips built on more advanced technologies.
The momentum AMD built for itself through 2011 is on the verge of stalling out. Trinity, the Bulldo er-based successor to Llano, will reportedly deliver strong GPU performance and a better CPU-GPU interconnect but AMD will be doing well to match Llano’s performance-per-watt ratio, much less exceed it. Furthermore, Llano itself benefited from a significant halo effect; it was viewed as the first iteration of a design philosophy whose performance Bulldo er would significantly improve.
AMD and GlobalFoundries: The bloom is off the rose
AMD’s decision to shift production from GloFo over to TSMC, while dramatic, is the latest in a long chain of events that point towards increasingly frosty relations between the once unified companies. In its Q3 call, AMD noted that “we no longer have the right to designate a representative to the GlobalFoundries’ Board of Directors” as its share of GlobalFoundries had decreased to 9.6 percent. In June, GlobalFoundries fired CEO Doug Grose (formerly AMD’s VP of manufacturing) and replaced him with Ajit Manocha, whose most recent executive experience was at Spansion, AMD’s red-headed, unmourned, and unprofitable memory spinoff.
Multiple GF executives have left the company in the past year and one of the casualties of AMD’s bloodletting last month was John Bruno, the chief architect behind Trinity. In 2009, GF officers confidently talked up their intent to win AMD’s discrete GPU business at 28nm. Today, AMD’s 28nm GPUs are all being built at TSMC.
AMD reportedly cancelled Krishna/Wichita when it became clear that GlobalFoundries wouldn’t be able to ramp the parts at volume and was unwilling to negotiate a new wafer agreement. Right now, AMD only pays GF for good 32nm dies, which means GlobalFoundries has been stuck building Llano at a loss for most of this past year. That agreement expires on January 1, at which point Sunnyvale goes back to paying a flat fee per wafer.
AMD doesn’t want to pay a per-wafer fee for a slow, low-yield product ramp, while GlobalFoundries is stuck with product lines dedicated to 32nm SOI production for a company with decidedly uncompetitive products. Llano performed well but was difficult to ship in volume, while Bulldo er scaled much more easily but was re-spun multiple times in an attempt to improve performance. From GF’s perspective, it took significant pains to ensure that AMD’s Bulldo er was ready for prime time — and AMD returned the favor by designing an architectural atrocity.
The two companies will continue to collaborate on AMD’s main product lines — neither has a choice — but the warm fu ies are at an all time low.
The widening gyre
Rory Read’s decision to gut AMD’s marketing and PR departments makes a lot more sense given what we know now, even if the dismissal of individuals like Bruno and Killebrew still leaves us scratching our heads. Slashing these departments frees up funds for immediate investment into ramping a 28nm APU at TSMC. It’s a desperate Hail Mary — and a watershed moment for the company. Instead of focusing on AMD’s historic core markets, Read is doubling down on Bobcat.
Will it work? We’re dubious. Trinity should be strong enough to anchor AMD’s mainstream notebook parts through 2012, but this cancellation leaves AMD without a new ultra low-power product family at a time when its competitors are heavily focused on providing such solutions. The company’s ability to launch tablet parts or hold the ground it gained in netbooks is very much in question. At the upper end of its product stack, AMD needs a Bulldo er core that delivers the power efficiency and performance the company originally promised, but there’s no such chip on the hori on.
Read is betting that the incremental 10-15% improvements the company plans to deliver to Bulldo er will be enough to keep the chip attractive to OEMs and end users. It’s a huge gamble — but it’s probably AMD’s last chance to establish itself as anything other than a cheap alternative for x86 processors. Intel’s lead time on process technology deployments has grown every year — if AMD doesn’t field something competitive in the next 24 months, the gulf between the two may grow too wide to leap.
Manufacturing bombshell: AMD cancels 28nm APUs, starts from scratch at TSMC
0 comments:
Post a Comment