Review Index:

AMD Q2 2010: Financials, Fusion, and Bulldozer

Subject: Editorial
Manufacturer: AMD

When is a Profit not a Profit?

    "Can I have some more please?"

    This past week AMD released their Q2 numbers, and just as Intel had a very positive report, AMD is following suite.  Q2 is typically a poorly performing quarter, oftentimes lower than Q1 due to the usual overflow of commerce from the busy Holiday season.  Q3 is usually stronger, primarily due to “back to school” spending.  So that often leaves Q2 eating the dust of the rest of the year.

    Two days before AMD posted their results, Intel showed off their record quarter.  Many were confused if AMD would see the same type of boost that Intel had or not.  As it turns out, the market was very good to AMD. 

    Revenues for AMD were $1.65 billion for the quarter, well above the expected $1.55 billion most analysts were predicting.  AMD also reported that margins had grown up to 45%, which is well above what AMD has seen for the past several years (it once hit a low of around 28% over a year ago).  Depending on what articles have been read, there was some confusion if AMD had pulled out a profit, or had swung to a loss.  The truth of the matter is, they did both.

    Accounting abuses have been common over the past couple of decades (and longer), and the financial industry as adopted what is known as G.A.A.P., or “Generally Accepted Accounting Principles”.  Reporting under G.A.A.P. gives consistent results that can be compared between companies which could have materially different markets.  Most companies also report non-G.A.A.P. results as well, which take into account differences in valuation which are not explicitly defined under G.A.A.P.

    Using G.A.A.P. AMD suffered a $43 million loss due to a $120 million charge connected to continued ownership and participation of GLOBALFOUNDRIES.  Cash on hand has stayed relatively constant from quarter to quarter, with a slight dip due to this loss.  However, AMD was able to reduce its long term debt by approximately $206 million by buying back of Senior Convertible Notes that were due in 2015.  AMD expects to continue to work down their long term debt over the next few quarters, and appears to balance that out with achieving a small profit as compared to reporting larger profits while keeping debt static.

    Non-G.A.A.P. reporting shows a profit of $83 million, or $0.12 cents per share.  This is of course above the $0.06 per share that many analysts were projecting.  While it would have been nice for AMD to show a profit in both G.A.A.P. and non, it has been a particularly strong quarter for AMD.

Breaking It Down

    The compute group had the lion’s share of revenue, coming in at $1.21 billion.  This includes all CPUs and chipsets, plus other odd products that AMD releases under this umbrella.  The graphics group contributed $440 million to the grand total, which is a record for this group under AMD.

    On the CPU side AMD has seen a brisk uptake of its Athlon II series of processors on both the desktop and the notebook.  Notebook growth has been very, very strong for AMD due to its combination of rich featureset (DX 10.1 support, accelerated HD video playback), competitive power consumption, and relatively low cost.  The introduction of the 6-core desktop processors in the previous quarter helped to jumpstart the high end offerings from AMD, though overall ASPs (average selling prices) were down, actual products shipped were up.  AMD also saw a boost from the introduction of the 800 series of chipsets, who’s SATA 6G support has turned into a significant checkbox feature for OEMs, as well as the close bundling of USB 3.0 support for motherboards using the 800 series chipsets.

Propus is the chip used in the Athlon II X3 and X4 processors.  It is relatively small for a quad core, and apparently still has decent margins though they are selling at sub $100 prices.

    The graphics side has undoubtedly been the star performer for AMD.  While the graphics technology in chipsets is relegated to the compute section when reporting financials, the standalone graphics chips have proven to be quite strong.  The basis of this strength is of course being the first to offer top to bottom DX11 based parts.  Unfortunately for AMD, production of these parts since last Summer has been constrained.  In other words, AMD left a lot of potential revenue on the table due to demand that they were not able to satisfy with the supply at hand.  Even with the influx of DX11 cards from NVIDIA at the high end, demand for AMD graphics chips are still at a near all time high.

    Servers were a disappointment for AMD, and they appear to have lost marketshare to Intel in this very important sector.  AMD was hoping for better, but some of its major OEM partners were unable to put out good quantities of products based on the Opteron 4000 and 6000 parts.  Of greater import is that of new Magny-Cours based products that have finally been released in decent quantities for Q3.  This higher margin part is aimed at 2P and 4P servers, with a focus on compute density and low power consumption per core.  AMD expects an uptick in marketshare due to these parts being readily available in Q3 of this year.

Future Products and their Potential Impact

    AMD announced that it had successfully taped out its next generation high performance server and desktop processor core, codenamed “Bulldozer”.  This product will be showcased architecturally at the Hot Chips conference later this summer, though actual product announcements and specifications will be unveiled closer to the actual introduction in 2011.  It typically takes between 9 to 14 months for a product to be released to retail from tapeout.  For example, in January of 2008, AMD announced that it had taped out its first 45 nm quad core processor, which later was introduced as the “Shanghai” based Opteron in October, 2008.  This then saw release on the desktop as the Phenom II X4 940 initially in January, 2009.

A high level diagram of what Bulldozer will bring to the market.

    Bulldozer is big news, as it is a big step away from the Athlon/Phenom architectures we have been seeing since 2004.  From a high level view, each core will actually consist of two integer execution units and a single 256 bit SSE/FPU unit.  So an “8 core” chip will actually contain 4 SSE/FPU units.  AMD believes, and rightly so, that most CPU workloads are integer in nature.  The improved 256 bit SSE unit should be able to handle the floating point operations much more efficiently than the previous 64 bit unit in the Athlon 64, and the 128 bit unit in the Phenom series.  It should be able to pack 4 x 64 bit single precision floats (SSE), or potentially 2 x 80 bit traditional x87 instructions.

    While AMD has not exhaustively gone into the Bulldozer architecture, it is not a stretch to imagine that this will get the Fusion treatment sooner rather than later.  A Bulldozer with a high performance GPU with 480 steam units could potentially be a floating point/parallel processing monster with high integer performance.  Much of this depends on the software infrastructure which would allow workloads to utilize those attached stream units, rather than rely solely on the SSE/FP units of the CPU.