AMD Q2 2010: Financials, Fusion, and Bulldozer
More Bulldozer and Closing
Bulldozer still appears on track for a 2011 release, but much of this depends on GLOBALFOUNDRIES ramp of their 32 nm process. AMD did not specifically say when Bulldozer taped out, but did say that they expect to give their partners samples in late Q3 and Q4 of this year. While GF is not yielding 32 nm where they hoped it would be, this does not preclude sending Bulldozer based wafers through the process to get initial products into AMD engineers’ hands. I would be surprised if AMD did not have initial working samples by August.
The next bit of good news is that AMD is pulling forward their Ontario based Fusion product. This is a CPU based on the low power Bobcat core, which is aimed at the Atom CPU from Intel. AMD expects this product to have “lower notebook” CPU type performance, which I would personally categorize as a per clock performance on par with the original dual core Athlon 64 X2 parts clocked at 1.8 GHz and above. Each CPU core is relatively tiny, around 25 to 30 mm squared. The chip will be fabricated at TSMC on their 40 nm process.
Bobcat is a highly simplified, but efficient design which should match (and perhaps exceed) the performance of the first dual core Athlon 64 parts from AMD back in 2004/2005.
This chip was actually demoed at Computex earlier this summer, running a scene from the DX 11 based “Aliens Vs. Predator” game, at a decent resolution and framerates. The graphics portion has not been categorized in terms of stream units or performance, but I would guess that it would be comprised of 2 x 80 stream processor SIMD units. This would make it faster than the 55 nm based 785G/880G integrated chipset, which features one 80 core SIMD unit. The entire chip, with a dual core Bobcat processor, would still be less than 110 mm squared at 40 nm. This is a pretty compact part considering its potential performance and featureset. These parts should be introduced in sub 20 watt TDP packages.
AMD was initially going to introduce Ontario in 1H 2011, but due to issues with Llano, and the opening up of more 40 nm space at TSMC, AMD focused their efforts on getting this part out the door as soon as possible. While the CPU performance will not astound, the graphics performance in such a package and at the expected price will be second to none. This could be a very disruptive part for AMD as compared to what else is on the market at this time, and certainly it could revolutionize the graphics performance (and video playback) of netbooks and low end notebooks. The idea of a 7 hour battery life while watching 1080P HD movies could appeal to a broad range of consumers.
Llano is the mainstream/desktop Fusion product based on a tweaked Phenom II architecture and a 6 SIMD graphics core. This product is to be produced on GLOBALFOUNDRIES’ 32 nm SOI/HKMG process. Design work for this part has been done for some time, and there are rumors of samples floating around. However, these samples have not been released to partners, and are internal to AMD. This was to be AMD’s foil against Intel’s upcoming Sandy Bridge, but unfortunately for AMD they will not be able to release the part in a timely manner to offset Sandy Bridge.
We were told in the conference call that Llano would be pushed back “internally” due to the 32 nm yield curves not being where they were expected to be. At first this looked to be near catastrophic news for AMD, but after digging for more information a few tidbits have come to light. Some time ago, internal to AMD and GF, the timetable for Llano was pushed forward. The conversation apparently went much like this:
“Ok then! Let’s do it! Change the internal roadmap to show Llano being out in late Q4 2010!”
Needless to say, the reckless optimism of the time has bit AMD in the short term. GLOBALFOUNDRIES is right around where they expected to be with their 32 nm ramp in terms of yields, but it is not anywhere near where it needs to be to have a successful launch of Llano in the aggressive timeframe they were hoping for. So, while to the outside it appears as though Llano is going to be 2 to 3 months late, in reality it is right where they initially expected it to be. AMD’s official stance is that Llano will be released in 1H 2011, and that has not changed in their publicly available roadmaps.
Another reason for the “delay” of Llano has to do with the unheralded “tic-toc” method that AMD has been using for over a decade. If you can remember back to the heady Athlon days, we saw a very conservative approach to new process nodes from AMD. The original 250 nm Athlon was essentially unchanged when it went to the copper based 180 nm process. It was not until sometime later that we saw the integrated cache Athlons on that process released. We saw this again at the 130 nm node when the Thoroughbred cores were released, and were functionally identical to the 180 nm Palomino cores (though the cache had a new placement to maximize die space). And again with the jump from the 130 nm Athlon 64 “Clawhammer” to the 90 nm “Winchester”, which were again relatively unchanged from each other.
Llano will be a tweaked Phenom II architecture with 1 MB of L2 cache per core, no L3 cache, and a boatload of DX11 stream processors.
Llano is a complex, new part on a complex, new process node. As such, it makes many of the engineers at AMD and GF quite nervous. This slightly more conservative approach is in place to make sure that we don’t see the Fermi-type issues that NVIDIA had with their new, large, complex architecture on a new process node. AMD cannot afford to miss its release by 6 to 9 months, and so they are taking the extra time to insure that they have working product out in the necessary timeframe and at sustainable yields.
All in all, AMD had a very good quarter with some very good progress towards being independent and profitable. Its stake in GLOBALFOUNDRIES is now down to 28%, but they still have a very close working relationship with that company due to AMD being their only major client so far. The graphics portion of AMD is again very competitive with its arch-rival, NVIDIA. It is a profitable section for AMD, and one that could have a huge impact on AMD’s future technical prowess in CPUs, especially in regards to massively parallel stream processing.
Progress is being made on Bulldozer, but for the time being Magny-Cours will be the flag bearer for the server market for AMD. Thuban (Phenom II X6) has also allowed AMD to stabilize their ASPs on the desktop, and we should expect a few more SKUs based on that part through the rest of the year. The Athlon II series is the real breadwinner for AMD on the desktop and notebooks, and these inexpensive, yet still well performing cores has allowed AMD to retain marketshare even in the face of Intel’s superior Nehalem based parts.
AMD is hoping that its edge in graphics performance will make Llano more attractive than Intel’s Sandy Bridge. We see an interesting matchup here, with Sandy Bridge having more CPU horsepower, but we are assuming it to have lesser graphics performance and potentially not be able to do GPGPU type work. While on the other side we will have lower CPU performance from Llano (tweaked Phenom II design), but with a far superior graphics unit and AMD’s growing expertise with GPGPU type workloads. Remember, AMD’s OpenCL software utilizes both CPU and GPU resources.
Which product will gather the most support? Well, I don’t think that Llano will upset Intel’s marketshare to any major degree, but it could allow AMD to sell more units overall, and at a higher margin, than what they were able to do up until that point. AMD is obviously very excited about Fusion, but they may not have a completely comparable part to Intel until Bulldozer variants are produced in the Fusion line.
The important thing to take away here is that AMD is still competitive with Intel, even though they don’t have a 32 nm product shipping at this time. The combination of a solid CPU lineup at decent prices, leading edge chipset design and feature integration, and a graphics portion which is firing on all cylinders for both integrated and standalone parts, gives AMD a solid foundation to build upon. Ontario will give AMD a much needed boost in the mobile sector, and it could be a class leading product upon introduction. The main push for now is to get 32 nm production to where it needs to be, as soon as possible. Once that occurs, AMD will be in far better shape against Intel for the near future.