AMD Bulldozer architecture design was somewhat different from than in the current mainstream X86, it’s neither CMP (Chip MultiProcessors) nor SMT (Simultaneous Multithreading), but CMT (Cluster Multithreading) design. The architecture is innovative, however, innovation always comes with cost, let’s not speak of the performance capability, the coordination between software and application is a tricky issue.

The existing operating system still can’t adapt to AMD’s CMT architecture, thread scheduler prevents CPU from deliver desired performance, which causes performance decreasement, and therefore, many people hope that new hotfix will make some different on this issue.

Such Bulldozer hotfix indeed exists, Microsoft has conducted some improvement on multi-thread scheduler in its upcoming Windows 8, earlier we have reported that Bulldozer performance got improvement under Windows 8, later AMD and Microsoft planned to release optimized coding in the form of hotfix, last December, the hotfix was available for download, but some tested IT media found the patch was of little help, Microsoft admitted that the patch was not an immature one, the final patch would come out in the first quarter of 2012.

But we don’t have to wait that long, on January 11th, Microsoft let out a pair of Bulldozer hotfixes—— KB2646060 and KB2645594, which are compatible with Windows 7 and Windows 2008 R2 OS, support desktop FX, server edition Operon 4200 and 6200 series processors.

Let’s see the characters of two hotfixes firstly.

KB2645594 is the main hotfix for performance improvement, its function is to tell the scheduler that AMD Bulldozer CPU contains dual-core in each module. Take dual-thread as example, if two threads are close relevant, they will be arranged priorly for running in a module, then both of them could share resource better. On the contrary, if they are hardly relevant, scheduler will assign them to different module for avoiding scrambling.

Picture courtesy of Extremetech

The hotfix is most effective with 1-4 threads application , as distribution efficiency is most likely to meet issue in less thread condition, while in multi-thread condition, all threads could be assigned to every module/core, hence the hot fix is nothing useful.

For energy-saving multi-core GPU always decrease part CPU core to energy-saving dormancy state which includes C1, C3 and C6. The C6 is the deepest dormancy power state, in which condition CPU core nearly closes completely, it needs longest time to be waken up. In addition, AMD and Intel Turbo rely on CPU energy-saving technology, the more CPU cores enter C6 power state, the higher working efficiency of rest cores will have. More flexible than previous-generation TurboCore, AMD’s FX-8150 could be accelerated to 3.9GHz or even 4.2GHz.

And then let’s have a understanding of the function in KB2646060. In fact, it may be not considered as performance boost hotfix, as it just avoids AMD processors enter C6 power state quickly, as we know, it needs some time for CPU cores to be shut down/waken up, the latencies arising from it will cause some unfavorable effect on CPU performance. KB2646060 can make CPU core keep in full speed to avoid entering C6 power state prematurely. Nonetheless, the hotfix has somewhat negative effect, as it may result in increased CPU power consumption in lightly-threaded environment (like 1-4 threads) when keeping full-speed operation.

In addition, Microsoft stresses that, it requires to install KB2645594 firstly if you would like to install KB2646060, surely, you can choose not to install it, but the latter hotfix can’t be uninstalled upon installation.

Let’s go on to the next page to see whether the two hotfixes can make a big difference on optimizing the performance of AMD’s Bulldozer processors.

Page 1: Long-Awaited Bulldozer Finally Coming
Page 2: Bulldozer Performance Optimization: Better Than Nothing

Pages: 1 2