AMD X2 Windows XP patch: my opinion

[astaris]

The hotfix solve three problems. Let's see them one by one in order to understand what we really need:

1) Possible decrease in performance during demand-based switching
Windows XP supports processor performace state transitions in order to reduce the energy drawn by the processor. A performance state is a couple of frequency and voltage values: lower the performance state lower the frequency and the voltage lower the energy consumption. The problem here is that in a dual-core system windows can select the wrong performance state; this action of course will result in decreased performance. The new power-saving strategy solve this problem, but it is disabled by default. As we know the new behaviour is enabled by setting the value PerfEnablePackageIdle to 1.

Now the question: do we really need the new performance behaviour?
The answer: NO

Inside the answer:
The performace states of the processor (P-States) can be changed only if:
1) Cool And Quiet is enabled in the Bios.
2) The AMD Driver is installed.
You can see the details on the excellent document "BIOS and Kernel Developer’s Guide for the AMD Athlon™ 64 and
AMD Opteron™ Processors" that can be found on the AMD site.
So if Cool And Quiet is disabled or the AMD driver is not installed, we do not need to enable the new power saving strategy.

2)Correct TSC synchronization
3)Correct C-state promotion and demotion
I'll discusse these 2 points together as they are strictly correlated.
On AMD dual core processors the TSC (Time Stamp Counter) registers of the two cores may drift with respect to one another as a result of P-states change (that does not occur if Cool And Quiet is disabled) or as a result of one core entering the C1 state. One core enter the C1 state when the OS issues an HLT instruction. So the idle process of windows (that uses HLT instructions) causes the core in idle state to enter the C1 state. Now if a game use the TSC values of the cores and the affinity is set to both the CPU it'll get confused by the two different values. The result is that we experience stutter. These 2 points are the important ones in the patch because with this fix windows synchronyze the TSC values on the basis of the the ACPI timer even when the processor enter the C1 state.

As a result of this analysis we can get this little FAQ:

Q) Do we need the hotfix for AMD dual core processor
A) Yes

Q) Do we need to set the value key PerfEnablePackageIdle to 1?
A) Only if we use Cool And Quiet

Q) Is is necessary to install the new AMD driver?
A) This driver is only needed if we use Cool And quiet because it addresses together with the first hotfix a problem specific to P-states transition.

Q) it is necessary to add the string /usepmtimer in the boot.ini?
A) /usepmtimer instructs the Os to use the ACPI timer instead of the TSC timer (default action). Because with the patch windows corrects the wrong TSC behaviour the /usepmtimer switch is not needed.

[shuRe]

thanks for clearing that up, I didnt think it was really needed, but thats just confirmed it.

[Manoj]

I don't know but at first I used to have some problems in games but after the hotfix and installing the AMD drivers and adding "/usepmtimer" in the boot.ini, the games run more smoother.

Things have been great so far.

I can afford to sacrifice a small performance hit for stability. Athlon FX 60 and Geforce 7900 GTX Extreme SLI is already pretty fast.