Responsiveness & Jank Thread

[Cyberbeing]

What is C states and why does it make such a difference in my case? What are the side effects if I disable C states?

C-states are power states, and I wouldn't recommend disabling them all without good reason.

Some of the lower ones like C1/C1E/C2 are often required for TurboBoost and/or SpeedStep to function as expected, both features which you'd usually want to remain functional unless overclocking.

The higher ones like C3/C6/C7 put CPU cores in various levels of reduced functionality deep sleep states (partial core shutdowns), with very very low power draw. Though if you don't care about saving an extra ~1-3W of power draw at idle, these can be safely disabled. It's not unknown for deep sleep states to have an impact on I/O or subsystem performance when active, but it's usually not very significant. It also seems that some older and lower end PSUs don't even support power draws low enough to support the Haswell C6/C7 C-states properly.

[Omega X]

Older Core2s had problems with C-States and memory stability depending on the motherboard.

[Virtual_ManPL]

After reading your discussion about HPET I finally decided to redo a deep test by myself and the results were:

1. TSC = bcdedit /deletevalue useplatformclock + HPET Support=disabled in BIOS - the lowest delay values
2. TSC + HPET = bcdedit /deletevalue useplatformclock + HPET Support=enabled in BIOS - higher delay values than TSC
2'. HPET = bcdedit /set useplatformclock true + HPET Support=enabled in BIOS - nearly the same delay values as TSC + HPET
3. LAPIC = bcdedit /set useplatformclock true + HPET Support=disabled in BIOS - nearly low delay values as TSC, but mouse laggs in LatencyMon tests

in BIOS:
CPU Enhanced Halt (C1E) - disabled
CPU EIST Function - disabled
HPET Mode - 64 bit

for tests these software were used:
DPC Latency Checker
LatencyMon (HIGHT_LEVEL_IRQL, CLOCK_LEVEL_IRQL, DISPATCH_LEVEL)
PC Clock Frequencies (QueryPerformanceCounter)

So it looks like the TSC with HPET disabled in BIOS are the best settings in my case.

[avada]

Older Core2s had problems with C-States and memory stability depending on the motherboard.

How does that manifest? Is an e6750 affected?

[Virtual_ManPL]

@ avada - maybe you missed this part

depending on the motherboard.

[avrion]

I'm experiencing a rather annoying thing on my hardware.
With "gfx.vsync.hw-vsync.enabled" enabled my graphics card does not up-clock to its specified overclock setting.
Once I start a game, that game and firefox are actually fighting over the clock speeds, it will randomly jump from default clock to over-clock, thats pretty freaking annoying. Granted, my hardware/drivers are probably also to blame.
Here are my specs:
i7 920 @ 4.2ghx
12gb ram
2x 7970 in crossfire / latest beta drivers

edit: Never mind any of that, the real cause for the clocks not jumping to its over-clocked level is the regular "use hardware acceleration" setting. I had that turned off and update v40 must've turned it back on.

[DanRaisch]

Not sure if there was a question intended there. Seems the obvious answer is to set "gfx.vsync.hw-vsync.enabled" to False. Oh, and maybe try things with the release drivers?

[Ver Greeneyes]

Firefox's vsync implementation doesn't do anything special - it just calls DwmFlush() to fire a timer after every vblank event. If you have the Desktop Window Manager (DWM) turned off, I believe it uses a 16ms timer to get reasonably close to the same timing. So at most, having the vsync implementation enabled will change how frequently and regularly Firefox paints. But that doesn't tell us much about why it causes your drivers to go nuts.

[avrion]

I apologize, I jumped to conclusion, the actual cause of my issues is the regular "Use hardware acceleration" option, I had that turned off and update v40 must've turned it back on.