The job cuts that AMD CEO Rory Read promised a few weeks back have now begun. AMD has closed its Operating System Research Centre (OSRC) in Dresden, Germany. The programmers at the OSRC were responsible for a number of code improvements to Linux, as well as for supporting features like PowerNow and Turbo Core. AMD was 17th on the Top 20 list of Linux kernel contributors through version 3.2, with 2,510 accepted changes or roughly 1 per cent of the total. The OSRC was also apparently working on virtualisation support in Linux 3.6.
From 2008 to 2010 the group was quite active, with a number of presentations and papers on multi-core and many-core processing, thread synchronisation, transaction locks, and various improvements to AMD’s hardware virtualisation support. Public disclosure on these initiatives appears to have ended in 2010; it’s unclear if the company was still pursuing these lines of research under the radar.
Obviously AMD has to make cuts somewhere, and having just fired some 1,500 people last year, it’s safe to assume the company has long since trimmed the easy fat. Nonetheless, these types of choices make us uneasy. True, they fit with Read’s declaration that the company’s future is in reusable IP and semi-custom work, but that attitude doesn’t give us much reason to think AMD will continue to offer a unique value proposition.
Consider x86-64 (aka AMD64). It’s a poster child for how AMD was able to build a standard that now dominates the x86 computing industry, but it didn’t happen overnight. AMD64 began as an alternative to Intel’s IA-64 (Itanium) instruction set. It was announced in 1999, the full specification was finished in 2000, and the first CPU that could use it arrived in 2003.
In July 2010, Microsoft announced that the number of PCs using a 64-bit version of Windows had finally broken the 50 per cent barrier. That means it took a decade for AMD64 to go from an idea to the status quo. In 2009/10, AMD was clearly working on some of the same concepts that Intel plans to debut in Haswell next year, like transactional extensions and lock-free programming. I’m not saying Read personally killed those initiatives – the company stopped publishing its work on them long before he was hired – but it’s an example of an area where AMD was doing cutting-edge work, and apparently isn’t anymore.
New communication strategy needed
AMD’s stock price is stuck at $2.00 in part because investors aren’t satisfied with vague promises about semi-custom SoCs, a more agile AMD, and future viability. Instead of trying to bury doubters in an avalanche of optimism, Read and his executive team need to stand up, acknowledge the real impact their changes will have on AMD, and take a page from Dirk Meyer’s playbook.
Meyer, in retrospect, clearly made mistakes with Bulldozer’s design, but his tenure at AMD was marked by a return to fundamentally timely roadmap execution. Phenom II, Thuban, the Radeon 4000 and 5000 family, and the various server updates all dropped when he said they would. The one exception to this was Llano, which was pushed back due to manufacturing problems at GlobalFoundries. Even in that case, Meyer was able to soften the impact by bringing Brazos forward. Llano ultimately did meet its revised launch period, though supplies remained tightly constricted through 2011.
AMD’s current roadmap is a shambles. Kabini is set to launch in the first half of next year, but the follow-up APU to Trinity, codenamed Kaveri, hasn’t taped out yet. AMD still won’t admit that it’s manufacturing Kabini at TSMC. There’s no information on whether or not Socket AM3+ has a future beyond the recently launched Piledriver CPUs, or if the company will bring Kaveri’s CPU (Steamroller) to market as a standalone server/enthusiast part.
That’s not to say that every development is negative. In the past six months, AMD has bought itself a server vendor, a new mesh fabric, launched an interesting mobile compatibility initiative, announced plans for 64-bit ARM servers, and begun working more closely with game publishers to promote AMD hardware.
The problem here is that, devoid of an overarching explanation, AMD looks like it’s flailing. The only way for Read & Co to stabilise analyst perceptions of the company is to explain, in detail, why certain departments are being closed, explain what products are being prioritised, and which are not, and when the new products are going to ship. Then ship them, on time.
If the computer market was healthy, AMD might have more room to manoeuvre, but Q3 system sales fell a whopping 9 per cent year-on-year. That’s the biggest decline in over a decade and it’s not expected to reverse in the fourth quarter. Macroeconomic indicators remain weak and Read’s own forecasts don’t predict a return to profitability until late 2013 or early 2014. A clear, well-executed roadmap is one of the few cards AMD could play.
If AMD executives don’t step forward to pre-emptively own the layoff narrative, it’ll be written piecemeal. Since the only people left at AMD are the ones working in key areas, layoffs that affect those areas will be seen as crippling to the company’s chances of recovery. Once that stone starts rolling, it won’t matter if Read had a tenable plan or not.