IO causing micro-stutters / mini freezes in all games. How to solve ?

[Spip]

Been basically trying everything the past 6 weeks*. Lately, I'm suspecting an I/O cause for the stuttering in all games:

• In Cyberpunk 2077, whenever the game autosaves, there's a micro-freeze, and the same happens when I take an in-game screenshot.
  This should not happen: Writing a 4MB file (autosave) or 300KB image (screenshot) is nothing for modern I/O subsystems.

• In AC Valhalla, the framerate goes all over the place and drops multiple times close to 0 while it's clearly loading new textures / effects / environment (have a video demonstrating it).

During these stutters, the CPU frametime spikes and some P-Cores usage drops to 0%, as if interrupted by high priority IO messages. Tested with games on: SATA HDD, SATA SSD and NVMe. Same results.

Lookingt at default MSI (Message Signal Interrupts, not the company ): I/O controllers are by default set to "High" priority and have loads of low "IRQs" (from -2 to -37, then comes the NIC -39 to -63 and finally the graphics card at -65); and their limits are high too.

Unfortunately, tweaking those settings does not improve the situation, and can even cause the NVMe to stop working properly.

Any ideas or suggestions on how to prevent IO to completely interrupt the CPU ?

Additional note: This occurs in all games, be they CPU intensive or not, GPU intensive or not. Neither are overheating or being throttled in simple games, but the micro-stutters will occur.

Specs:

13900K with a Noctua NH D15, about 70° max package temp while gaming
Asus Z790 Hero
RTX4070 Dual at 1440p with a G-sync native screen


PS: Is there a way to post images ? As I have several screenshots illustrating the observations.


* For those interested, a non-exhaustive list of things and combinations I tried, where none solved the issue:

• BIOS: HT on/off, Resizebar on/off, E-Cores on/off etc.

• CPU: Affinity setting, underclocking, setting power limits...

• Windows 10 and 11: HAGS, VRR, Game mode, Power settings (disabling parking etc.)

• Nvidia: With/without frame limit, Vsync, GSync, low latency...

• Extensive HW testing of RAM, CPU, GPU via stress-testing, and stability (long runs) testing

• Completely fresh windows install, with only the drivers and one game

etc.

[andrelip]

Advice:

1) Forget about interrupts/IRQs/priorities and focus on disk access, including writing and reading speeds and latency.
2) SSDs become extremely slow when the disk is full (70% or more). Ensure this is not the case here.
3) Run a disk benchmark and ensure that your values are on par with online results for the same disk. If not, run trim and defragmentation and test again.

Return with the results in order for us to help you troubleshoot the issue.

[Spip]

Thanks for your quick reply and advice. Please find some answers / comments below:

• System NVMe is 35% full (2 TB)

• NVMe where games are installed is ~60% full (4TB capacity). This is with overprovisioning enabled (10%)

• Benchmark results for the 2 NMVe are in line with expected results (7000 MB/s read, 6800 MB/s write, 1400000 random IOPS read, 620000 random IOPS write). These are the low end results.

• TRIM is enabled

Both NVMe have a heatsink. Highest temp is 58° under heavy load, more around 52° when gaming ; so thermal throttling doesn't kick in.

In terms of latency, I've launched LatencyMon and tested the following: lots of background programs running, Winamp playing music, and then playing a CPU/GPU intensive game. Results where all in the green (highest interrupt: 518µs, highest ISR 152µs, highest DPC 807µs).

[damon69]

Hey, Spip! I have been experiencing very similar symptoms to those you describe. So far I have only run into the micro-stutter/freezing while playing Cyberpunk 2077 update 2.11 and only while Process Lasso and Park Control are running. Before I figured out that turning off these two programs eliminated the freezing I optimized pretty much everything else. I turned off the GF Experience overlay, configured Afterburner and RTSS to both OC and cap FPS in order to reduce latency. I returned my CPU to stock frequencies and the RAM to XMP. I reinstalled the game and later Windows11. For the last year+ I have used Reshade to fix the raised Black Level when HDR is enabled but I tried running the game with RS removed. Cyberpunk has always been a finnicky-ass game but I do not remember this particular type of issue playing the game pre 2.11 update. In fact, Lasso was the thing that seemed to get the game running smoothly after the 2.0/PL update. I suppose that the new Hetero-Core game setting could be the culprit but I have run the game with it on and with it off and still only run into the freezing if Lasso and CP are active. BTW I've found that when the game freezes I can unfreeze it by pressing ALT-TAB, which only adds to my confusion but there it is lol! I have run the game in Windowed and Fullscreen and even tried disabling G-sync. But so far the problem only bites when Lasso and PC are active. Obviously, I can just disable these apps to play the game. But I'd really like to know the specific cause of the freezing so I can change settings or whatever and fix it. I'll put my specs below and I'll post any new experiences or discoveries I come across.

14900k (with a truly shitty V/F Table)
Z790 Apex White
32 GB of G.Skill DDR5 7800
Gigabyte 4080 OC
LG C2 42 inch OLED
Corsair AX1600i PSU
Corsair ICUE Link h170i 420
2x Samsung 990 Pro NVME SSD
Windows 11 latest, latest BIOS, Drivers etc.