AMD unlikely to gain ground on Intel at 32 nm. Does it matter?
Opinion – AMD’s 2008 Financial Analyst Day, held yesterday, saw a much more upbeat AMD than last year. The first 45 nm CPU is out of the gate, there is a fantastic graphics chip lineup, a new and compelling stream computing agenda and a thoroughly updated roadmap that had only one major gap – netbook processors. No doubt, AMD has gained strength and confidence, but there was a strangely inconsistent message about the migration to 32 nm. That was preceded by executives stressing that the Shanghai CPU was released early, which was not entirely correct. I would not care when AMD transitions to another process technology, but AMD seems to be making a big deal out of nanometers these days. Three executives talked about the 32 nm migration yesterday and gave three different time frames within three hours. Now I wonder: Should we care?
Let’s step back first and let me briefly explain some background on AMD’s process migration claims. Back in the Q3 2006 AMD conference call, then president and COO Dirk Meyer told analysts and press that AMD had accelerated its manufacturing roadmap and was planning on shipping 45 nm processors within 18 months, which translated to June or July 2008. The goal was to cut the distance to Intel’s manufacturing advantage (12 months at the time) in half. These days we often hear the company talking about the fact that Shanghai has been delivered early, which is about 50% true. The Shanghai chip itself was planned for a Q1 2009 rollout, according to AMD, but if we remember that mid-2008 45 nm promise, the CPU is about five or six months late. So there was not really any gain on Intel.
A couple months ago, I was talking to AMD senior vice president Randy Allen - after he claimed during a pre-IDF AMD briefing that Shanghai would be delivered early – and asked him to clarify AMD’s “early” statement. He told me that he would have to look back at Dirk Meyer’s claims. However, Allen admitted that AMD has not gained any ground on Intel, in terms of manufacturing technology, with 45 nm. But he mentioned that he would expect to decrease the distance with following product generations. 32 nm, 22 nm and so on.
Back then, I wrote that very few would care about the time when AMD is transitioning to a new process (exception: financial analysts). In the end I am not buying CPU nanometers when I am buying a new PC. And if AMD can squeeze more time out of 65/45 nm than Intel - especially through their immersion lithography advantage - good for them. But that is not what we hear.
The problem really is that AMD continues to make a big deal out of the whole manufacturing process message and, if you listen closely, that message tends to get screwed up and changed along the way. That was no different during today’s 32 nm news, which had some very interesting and surprising tidbits about the introduction and technology the company uses. And I was interested if AMD would actually be able to catch up with Intel, as indicated by Allen.
32 nm, take 1
Very early in the presentation (slide 16 for those who kept count and followed the presentations; you can see the slides including roadmap charts in our slideshow), a slide shown by president and CEO Dirk Meyer mentioned that 32 nm products will be finalized for a 2010 production. If you have followed AMD for some time, then you may have noticed that the company has kept a fairly consistent product introduction pattern in terms of server/desktop products. AMD server CPUs are often rolled out first during major technology refreshes (exception: 65 nm introduction), while the desktop chips follow 60 – 90 days later. But the news here is that AMD 32 nm CPUs in fact are coming in 2010. Let’s see if other speakers spill more information.
32 nm, take 2
Randy Allen was up next, but strangely enough, the server roadmap described by the executive did not highlight any nanometers and the 2010 timeframe did not include any 32 nm processors as far as we are aware of. The chart showed the Magny Cours and Sao Paulo 8- and 12-core CPUs, which are generally believed to be 45 nm parts. So, the obvious question was: Will AMD miss the 2010 timeframe with its server products? I continued to listen. Allen showed the desktop roadmap on slide 75. The audience was told that the 2011 Orochi processor will be the company’s first 32 nm client processor. Wait a second: 2011? More than 12 months behind Intel? And does that mean a server chip that was not mentioned on the server roadmap will precede Orochi?
Earlier in the day, a press release stated that the Deneb 45 nm (desktop) quad-core (Phenom X4 II) will be released in Q1 2009. That would mean that the 32 nm desktop chip will still be more than one year behind Intel, as it was the case with 65 nm and it is now with 45 nm. The good news, of course is, that if Orochi will be released in early 2011, a server product could make it into 2010, if the Barcelona and Shanghai product introduction strategy will repeat itself. Or, if we remember the 65 nm introduction (in which desktop CPUs preceded server CPUs), the server CPU could miss 2010. Allen had no further information on this topic, so I hoped for more information from the next speaker.
32 nm, take 3
Doug Grose, who will head up the recently introduced Foundry Company (as opposed to AMD, “The Product Company”), which will primarily produce semiconductors for AMD, noted on slide 118 that the company will focus on developing a 32 nm process in 2009 and that 32 nm chips could be in production in the first half of 2010 (slide 122). He noted that a 32 nm SRAM chip was planned for a Q4 2008 production (which is usually the first test chip for a new manufacturing process). 32 nm tape-outs are scheduled for 2009 and a production ramp is planned for “early 2010” – which, however, referred to bulk processors and ATI graphics chips and not AMD CPUs. This reminded me of Meyer’s earlier slide statement of a 2010 32 nm production. Did Meyer refer to bulk (=GPU) or SOI/high-K production (=CPU)? He did not refer to anything as he skipped the 32 nm note entirely during his presentation (But I guess he meant CPUs since there was a 40 nm GPU note as well).
If everything goes to plan, The Foundry Company will be able to begin production of “high-performance” processors in the Q1/Q2 2010 time frame. 32 nm bulk chip production capability will be achieved in Q4 2009 / Q1 2010 and low power high-K processors will follow in Q1/Q2 2010, Grose said.
So, when will we see AMD 32 nm processors and will AMD be able to cut down the distance to Intel? Obviously 32 nm was a big deal for AMD executives during the day, even if the information of a 32 nm server chip was completely left out of all presentations. During the following Q&A, Allen was asked when a “completely new core” will be released by AMD. He referred that it would really depend on the point of view what a completely new core is and in AMD’s view, even the 45 nm Shanghai has been re-engineered and could be described as a new core. But he ended up noting that new (32 nm) CPUs (Bulldozer and Bobcat) with a “new pipeline” and a “radical change” would be released in 2011. Nathan Brookwood from Insight 64 also jumped on the new core and 32 nm question and asked whether we are talking about a 2010 or 2011 introduction date of 32 nm. He was told that 2010 would be the time and the fact that it appeared like 2011 in the presentations was because of the “calendar breaks” in the presentation graphics.
I’d be lying if I said I wasn’t confused at this point. 2010? 2011? And if 2010, when in 2010? And what will we see in 2011? When can we expect the server chip?
I decided it was a good idea to contact AMD’s Jon Carvill, who will head up PR for The Foundry Company, and ask him for help to clear up what we should expect from a 32 nm migration. Carvill got quickly back to me stating that the first 32 nm client product (Orochi) will ramp in the second half of 2010 and see a late 2010 or early 2011 introduction date. Volume production is planned for the first half of 2011, Carvill said. “With these dates in mind, the 32 nm product transition is following a similar cadence of both 65 nm and 45 nm,” he said. Of course, that means that AMD expects to keep the two year update in place, which is in line with Intel, but it also means that AMD is unlikely to gain any ground on Intel.
Carvill said that AMD has “not specified process technology in the next-gen server portfolio.”
So, is this good news or bad news? For financial analysts it might be considered bad news as a transition of process technologies also impacts the economy of scale. But what about the consumer? The fact is that it really does not matter. 2010 or 2011, who cares?
I admit, I have been very picky with the 32 nm message in this article, but if AMD makes a big deal about 45 nm and the 32 nm transition, if it promises that it will catch up with its rival and three months later says it won’t, shouldn’t we make a big deal about it as well? If we hear that certain chips are early when in fact they are late, shouldn’t we expect a more consistent message? Somehow it appears that the message of 32 nm isn’t exactly aligned with every executive in the company and a possible introduction date of 32 nm processors is all over the 2010 plate, reaching even into 2011.
Or AMD simply got tired of being pinned down by journalists and questions that start with “but back then you said …”. What we seem to know now, however, is that AMD will not be able to decrease the manufacturing distance to Intel with the 32 nm generation of products either, as indicated by Randy Allen a few months ago, at least if Intel can keep its current late 2009 introduction date. I wouldn’t go as far as saying that the 32 nm chips are slated for a late introduction already, and frankly, I don’t care whether it is 2010 and 2011. But as long as AMD puts such a great focus on it and pairs it with a relatively inconsistent message, I am sure that journalists and analysts will make a big deal out of it as well.
What are your thoughts? Do you care about AMD's 32 nm message? Do you care when AMD will transition to 32 nm? Let me know by writing a comment below,