AMD Bulldozer is, to say so, unusual in its modular dual-core block architecture compared to the current X86 processors. To extract more performance out of it, Microsoft now has a kernel patch for for its client and server OS.
Ever wondered why some benchmark results on Bulldozer and its server version, Interlagos, look not just too low but outright weird? Part of the explanation was that the fancy modular architecture, where two integer and one FP block share one L2 cache and instruction scheduler, wasn't really supported that well by the current operating systems. Add to it the Pentium 4-style long pipeline and high latencies, and you got a problem.
Microsoft now has a kernel update solution that solves one part of the problem, at least – if you go to http://support.microsoft.com/kb/2592546/
knowledge base page, you'll notice that there is a patch for Windows 7 and for Windows Server 2008 R2.
This patch updates the thread management logic in the OS at the kernel level for the Bulldozer, and – interestingly – proclaims it as a multithreading feature. With this update, Windows sees each BD dual core block as A SINGLE MULTITHREADED CORE, contrary to AMD's official positioning. Therefore, BD is seen as a 4-core, 8-thread chip, rather than a fully 8-core processors. This update, nevertheless, brings a definite single-digit % performance jump in lightly threaded applications, but it is still good though as it helps the CPU fight higher speed grades of Intel processors in benchmarks. So, naming or no naming confusion, go ahead and have it, if you're one of the early Bulldozer users.
Leave a comment