In non-GAAP terms, AMD actually pulled a $92 million profit for Q1. This shows us that AMD is moving in the right direction in terms of how they are running their company. The big loss was due to the money given to GF, as well as the liquidation of AMD’s stake in that company. In GAAP terms, having a loss of $590 never looks good. But this does appear to be money well spent, and the investment will hopefully pay off in 2013 when AMD releases their 28 nm APUs.
Rory Read has really cleaned house at AMD. When many of the top (and familiar) people at AMD were either let go or resigned, my gut reaction was that Rory was simply pandering to Wall Street. After seeing how AMD has handled getting away from partial ownership in GLOBALFOUNDRIES, how it has dealt with TSMC and its 28 nm woes, and how it has aggressively redefined itself in the past 9 months, I am firmly behind what Rory appears to be doing. Rory’s stated goal is to turn AMD into an execution machine.
History does not look to be kind to Hector Ruiz’s time at the helm of AMD. When Jerry Sanders, III retired and handed over the reins, AMD was in a tremendous position. It had an aggressive roadmap for its processors, it was not all that far behind Intel in terms of process technology (about 10 months rather than the current 18 months), and the company was gaining marketshare in mobile, desktop, and server spaces. Apparently Hector came in and made some sweeping changes that derailed AMD from that roadmap. The changes did not allow AMD to be flexible enough to respond to Intel’s Core and Core 2 technologies. Instead users received the Athlon X2 Rev. E which added DDR-2 support. Then they had to deal with the original Phenom which never lived up to expectations. Dirk Meyer was an engineer first, and while he helped AMD get back on track in terms of execution, he appeared to be far too conservative in planning for the future. While we are still early in Rory’s tenure as CEO, the changes we have seen so far appear to be quite positive. Time will tell what kind of foundation he is laying and if he can do the job he was hired for.
Rory knows that AMD is the only real competitor to Intel in the processor field, and that very fact gives AMD a certain guaranteed market share. Rory feels that the graphics expertise that AMD currently possesses is their crown jewel, and they need to aggressively push that technology up and down their product roadmap. AMD is currently saddled with the underperforming Bulldozer architecture, and will be for the near future. This does not mean that it cannot be further optimized to decrease TDPs and increase IPC. Piledriver will be the first iteration of an improved Bulldozer, and while I am not expecting miracles in the IPC department, it does look like many of the negatives in terms of power consumption have been solved. Steamroller will be the next and is scheduled for a 2013 launch. In keeping with AMD’s increased pace of release, I would expect the first Steamroller based APUs to be released one year from now. These will initially be 28 nm parts, but if GF can get 22 nm out in 2H 2013 then we could expect those versions to be out by the Holiday season 2013. The 28 nm refresh of Piledriver should have the GCN graphics component, and those will again be released in early 2013. As per AMD’s conference call this quarter, they will be aggressively ramping 28 nm production at GF with these parts in 2H 2012 (hence the new exclusivity wafer agreement with GF).
Looking further ahead we see that AMD will slowly phase out their non-GPU enabled CPUs. AM3+ will get a refresh later this summer with a Vishera based unit that will share silicon with server based products. After that though, AMD will be moving towards APUs exclusively. We can assume that they will do the same for the server side and aggressively promote GPGPU and OpenCL applications for these server based APUs. One of AMD’s goals with GCN is to more tightly knit the CPU and GPU together. This will help simplify development, and in moving towards the future with GCN and later graphics technologies, AMD is hoping to carve out a larger marketshare in not only the traditional notebook/desktop/server spaces, but also in the fast growing mobile space. Expect to see AMD start to push out lower TDP products for tablets and eventually units which can fit into handheld devices. The improved Bobcat architecture can theoretically move into those TDP spaces.
In closing, AMD had a good quarter even though the GAAP loss of $590 million does put a damper on enthusiasm. That money is an investment in the future rather than a material loss that is gone forever. AMD has restructured itself quite nicely. Although the reduction in staff has left a bad taste for some, it is a much leaner entity going into the future. They also are appearing to execute at a higher level than what we saw last year at this time. The complete change to a 28 nm based graphics architecture went off nearly without a hitch. Trinity is on time and performing at expectations. Brazos 2.0 is an interesting adjustment considering the 28 nm versions of that architecture were cancelled and AMD made some major changes to the current 40 nm product offerings that apparently have made up for it.
One final metric to comment upon is that of ASPs. AMD does not comment nearly as much on ASPs as they used to, and this is because the overall product mix they are offering makes ASPs far less meaningful than they were 10 years ago. For example, Brazos APUs and their attached FCH (Fusion Control Hub), sell for perhaps $50 as a combo for the highest level SKU (in this case the E-450). In terms of raw die cost each Brazos APU looks to be around $10, and the FCH likely around $7. It will take around $5 per chip to make a finished product (wafer cutting, processing, and application to a substrate). So for each combo we would expect a finished APU/FCH combo to be $27 (not counting shipping costs to and from the packaging facility). This is a VERY rough estimate, but we can see that AMD has a goodly amount of margin for a sub-$100 processor. AMD has sold 35 to 45 million of them in the past year. So for the past year these products have added materially to AMD’s bottom line, as they are inexpensive to produce and have a solid margin. This does degrade the ASP for the past year, even though by all measure Brazos has been a great success and a constant source of profit for the company. Stepping away from the ASP metric is a good thing, and I certainly hope more analysts will realize this.
It does feel like AMD is back on track and is working to be a strong competitor to Intel once again. But that particular journey has just started and is filled with pitfalls, and AMD has a lot to prove between now and their first 28 nm APU release in 2013. Still, the numbers are overall very positive, and AMD looks to have some very competitive products throughout the rest of the year in terms of APUs and GPUs. Time will tell if the Piledriver based units for AM3+ and the server space will live up to expectations though.