The first thing is that Rick Bergman did leave AMD for a CEO position at a Synaptics. This is great news for the Bergman family, they no longer have to fret about losing one of the houses after 6 long days on the dole. Times are tough, but Synaptics just got a winner, hope you had stock.
For AMD, this is the best of the very bad scenarios. As we pointed out earlier, if the company was canning it’s best people, it is probably game over. As it stands, Rick leaving for a CEO-ship is the best case scenario, but it is still analogous to being told you have a very curable form of cancer. This isn’t a good thing, however the doctor just told us it is now officially less bad.
Speaking of bad for AMD, we come back to today’s announcement of an impending revenue miss. The short story is that for Q3, AMD will have a revenue increase of 4-6% instead of a projected 10%, and gross margins of 44-45% instead of 47%. This is not something Wall Street wants to hear, and AMD stock is down about 8.5% in after hours trading as this is written.
This confirms the rumors that have been going around for a while, they can’t get enough Llanos to supply the demand. The fingers are pointed at Global Foundries, and tensions between the two are said to be high. For AMD, it is not a case of high costs, they are only paying for good chips, but lost opportunity. Global Foundries is eating the bad chips, but that won’t go on forever. That said, they have done some house cleaning lately too.
Getting back to AMD, we are hearing that the inability to supply is not just costing them lost sales, it is also costing them in intangibles. OEMs are not happy that they can’t get chips, and neither are their customers. Is is a chain of unhappiness, and things are not looking to be solved any time soon. The two core Llano dies are set to come out in Q4, but that should only help going forward, the current problems do not have an imminent fix.
How bad is it? We previously mentioned the 50-60% yield numbers, but there is a more important number, fully working, IE A8, chips. Yields for those parts are said to be in the 5-15% range, something a little worse than abysmal. To put a positive spin on things, there are a lot more A6s and A4s floating around than A8s, so if you are in the market for one of those, things look good. Back to the negative side, there aren’t many of either in the channel, nor do their look to be soon.
One of the original problems that floated our way was rather intriguing, different design rules. The CPU and GPU side of the chip used different design rules, and initially you could get great yields on one, but not the other. That problem hampered yield improvement, but it was worked through long ago. What was an OK initial yield after that has not been showing the expected ramp up. This is troubling, and the source of all the problems between AMD and GloFo.
To make things more interesting, SemiAccurate moles tell us that the problems are related to Llano in particular, not the 32nm process. Bulldozer and Trinity, the next two 32nm parts coming out of GloFo, are yielding much better, far above Llano levels. Why does one design have massive issues while another not? Some people have said that the GPU is to blame, but it isn’t, Trinity doesn’t suffer from the same ills. Short story, no one is talking yet.
Where does that leave us? If you chipped in on the Rick Bergman gift basket, we will be refunding your money, All $9.72 of it. We had until noon Saturday to cancel, so the Synaptics announcement worked out perfectly. For AMD though, things are just bad. Wall Street is currently punishing them, and unless Llano yields jump up soon, things are not looking good for Q4 or Q1 either. About the best you can say is that things could be a bit worse.S|A
Updated: September 29th, 11pm.
More sources have come forward and said that Trinity is indeed having yield problems, but not as bad as Llano. Trinity is said to have started off better than Llano, and still have 2Q or so of optimizations to go before launch. That said, it is still more problematic than Bulldozer, but it is not the fault of the GPU.
Latest posts by Charlie Demerjian (see all)
- Thing go bump(gate) in the night for Nvidia’s GP100 Pascal GPU - May 3, 2016
- Cavium’s Octeon TX blends compute and packet moving - May 2, 2016
- Qualcomm releases Zeroth API to developers - May 2, 2016
- Another detail about Qualcomm server SoCs revealed - Apr 27, 2016
- AMD finally really honestly launches the dual Fury - Apr 26, 2016