Stress Test Lowers DPC

[yeesecretalias]

I found an interesting behavior, if you run a CPU stress test such as Aida64, your DPC latency goes down significantly.
On my system it goes from 0.27 > 0.16.
Not all stress tests do this, I've only found Aida64 and HeavyLoad to have this behavior.

Anyone have a good idea or explanation of why this occurs?
Is this something that can be used without the stress test portion to improve DPC latency?
Or is it part of the stress test mechanic?

[Shade7]

I found an interesting behavior, if you run a CPU stress test such as Aida64, your DPC latency goes down significantly.
On my system it goes from 0.27 > 0.16.
Not all stress tests do this, I've only found Aida64 and HeavyLoad to have this behavior.

Anyone have a good idea or explanation of why this occurs?
Is this something that can be used without the stress test portion to improve DPC latency?
Or is it part of the stress test mechanic?

If you wait a while- then test your DPC latency again- Does it stay low?

When does the effect revert- when you restart PC?

It's possible the stress test brings the CPU out of some kind of idle power mode and wakes everything up.

[yeesecretalias]

I found an interesting behavior, if you run a CPU stress test such as Aida64, your DPC latency goes down significantly.
On my system it goes from 0.27 > 0.16.
Not all stress tests do this, I've only found Aida64 and HeavyLoad to have this behavior.

Anyone have a good idea or explanation of why this occurs?
Is this something that can be used without the stress test portion to improve DPC latency?
Or is it part of the stress test mechanic?

If you wait a while- then test your DPC latency again- Does it stay low?

When does the effect revert- when you restart PC?

It's possible the stress test brings the CPU out of some kind of idle power mode and wakes everything up.

The DPC latency is decreased as long as the stress test is running. Stopping it reverts the latency back to normal.
My initial thought was also that the tests do some kind of CPU power saving configuration. Something fairly hidden since I have all the usual BIOS power saving / windows power plan culprits set correctly.

Also just to add further, this has been tested on multiple PC's for multiple people, and has the same effect.

[MisterS0L0]

Maybe using the full CPU etc stops any of the power saving techs doing what they do and you get better latency ?

[imprecise]

I found an interesting behavior, if you run a CPU stress test such as Aida64, your DPC latency goes down significantly.
On my system it goes from 0.27 > 0.16.
Not all stress tests do this, I've only found Aida64 and HeavyLoad to have this behavior.

Anyone have a good idea or explanation of why this occurs?
Is this something that can be used without the stress test portion to improve DPC latency?
Or is it part of the stress test mechanic?

Core Parking is enabled by default, have you turned it off?

Efficacy of Disabling Core Parking
Empirical evidence shows that disabling CPU core parking can make a tangible improvement in system performance. There are many factors that will determine precisely how effective it will be for a given situation. However, generally, Windows is too aggressive in its core parking, resulting in high latency during bursting CPU loads, stemming from the overhead of having to unpark CPU cores. Since bursting CPU loads are the most common type for many workloads, core parking can be a substantial drag on system performance and responsiveness.

Source:
https://bitsum.com/parkcontrol/

[Chief Blur Buster]

It's a very common power management technique. Latency/DPC often worsens when there's a lot more things sleeping/parked/etc.

Stress tests on CPU and GPU can suddenly lower latencies by waking up the power management behaviors.

Keeping power pegged up there is tough, given the energy efficiency drives (quite desirable if office work, quite undesirable in esports).

[dannyoceanic]

I found an interesting behavior, if you run a CPU stress test such as Aida64, your DPC latency goes down significantly.
On my system it goes from 0.27 > 0.16.
Not all stress tests do this, I've only found Aida64 and HeavyLoad to have this behavior.

Anyone have a good idea or explanation of why this occurs?
Is this something that can be used without the stress test portion to improve DPC latency?
Or is it part of the stress test mechanic?

Core Parking is enabled by default, have you turned it off?

Efficacy of Disabling Core Parking
Empirical evidence shows that disabling CPU core parking can make a tangible improvement in system performance. There are many factors that will determine precisely how effective it will be for a given situation. However, generally, Windows is too aggressive in its core parking, resulting in high latency during bursting CPU loads, stemming from the overhead of having to unpark CPU cores. Since bursting CPU loads are the most common type for many workloads, core parking can be a substantial drag on system performance and responsiveness.

Source:
https://bitsum.com/parkcontrol/

Any idea if unpacking cores on AMD processor help with latency problems in valorant?

[imprecise]

Any idea if unpacking cores on AMD processor help with latency problems in valorant?

Yes, I expect it to help. I always notice the difference after disabling core parking, tested across many different games.

Edit: I'm talking about input latency, not internet latency (internet ping), just to sort out any possible confusion.

[dannyoceanic]

Any idea if unpacking cores on AMD processor help with latency problems in valorant?

Yes, I expect it to help. I always notice the difference after disabling core parking, tested across many different games.

Thanks for the answer will test the program you suggest and try it out later with valorant, hoping that maybe it will make it run smoother

[schizobeyondpills]

I found an interesting behavior, if you run a CPU stress test such as Aida64, your DPC latency goes down significantly.
On my system it goes from 0.27 > 0.16.
Not all stress tests do this, I've only found Aida64 and HeavyLoad to have this behavior.

Anyone have a good idea or explanation of why this occurs?
Is this something that can be used without the stress test portion to improve DPC latency?
Or is it part of the stress test mechanic?

CPU doesnt leave C0 state, OS doesnt cause CPU to sleep(and no disabling idle in power settings doesnt disable OS placing cpu in C1/OS sleeping, its more like a hint to reduce it a bit, its still happening), OS scheduler works better, everything has 200+ layers of power savings since most CPUs/GPUs would pull 600W+ and be impossible to cool if they didnt throttle in milli,micro,nano,pico second ranges (all combined at all times)