Micro-stuttering / hitch & Frametime spikes issue
-
- Posts: 7
- Joined: 11 Oct 2020, 10:28
Micro-stuttering / hitch & Frametime spikes issue
Hi everyone,
I want to share an issue I have since a long long time ago. This issue has obsessed me. I've tried everything to solve it and, although it got better, is still there torturing me and preventing me from enjoying gaming.
The issue is some kind of micro-stutter or hitch that freezes the image (not the sound) just a fraction of second but enough time to break the immersion... frametime spikes.
Let's see, since I've done thousands of tests with different hardware/drivers combinations I'm just exposing current RIG, where the issue still persists but it's the best experience I've been able to achieve. I can say that I've tested almost every from a default clean Windows 10 installation to a custom installation with these specs/settings/configurations:
- i7 8700 5.1 GHZ Delidded (tested also before delid, with stock setting, manual OC, bios profile OC ...)
- Trident Z Neo 32 [2x16] GB RAM DDR4 3600 Mhz CL16-16-16-36 1.35v B-DIE (tested also 1 DIMM, 2 DIMMs, swapped, stock, XMP, manual OC, same with 16 [2x8] GB from Corsair Dominator DDR4 4000 Mhz)
- Asus Maximus X Apex last bios 2402 (tested with bios stock settings, custom settings...)
- Asus Rog PG278Q 2560x1440 144hz (tested also with Asus Rog PG348q 3440x1400 100hz)
- 1x GTX 1080 ti ASUS ROG (tested OC, non-OC, tested with another GTX 1080 ti ASUS ROG in SLI, single, swapped)
- Samsung NMve M.2 970 Pro 512GB (tested in 4 different SSD: samsung, kingston...)
- PSU BeQuiet 1500W Dark Power 11 (tested OC six 12V rails, non-OC one massive 12V rail)
- Water cooling Fractal Design s36 Kelvin
[Note: all configurations have been tested and stressed, no thermal CPU/GPU throlling found]
- Windows 10 Pro 2004
* Opera browser installed
* Ethernet devices disabled
* Audio devices disabled
* GPU/Audio default drivers uninstalled with DDU (safe mode)
* Nvidia Drivers 456.71 installed (tested with 456.55 and olders)
* Chipset driver installed
* ACHI driver installed (when tested in SDD)
* NVM Samsung drivers installed (when tested in NMve M.2)
* MSI Afterburner installed
* Steam installed (overlay disabled)
* GPU-Z installed
* CPU-Z installed
* XBox Game Bar disabled
* Game Mode enabled (tested with it disabled)
* VRR and GPU scheduled enabled (tested with them disabled)
* Windows update disabled
* Windows antivirus disabled (each time I reboot as I haven't been able to disable it forever)
* Firewall disabled
* HPET in device manager disabled
* CMD:
bcdedit /set useplatformclock no - Reiniciar
bcdedit /set useplatformtick yes - Reiniciar
bcdedit /set disabledynamictick yes - Reiniciar
* All not needed windows 10 features from configuration disabled (privacy...)
* All not needed windows 10 applications uninstalled (office, spotify...)
* Several not needed windows 10 services disabled (SysMain, windows update, printers mangement...)
* Process Lasso and Park Cores installed (cores unparked and games with high priority class and I/O priority, CPU affinity to all cores/threads, Application power profile to Bitsum Highest Performance)
* Intelligent standby list cleaner installed and current timer resolution set to 0.5)
* MSI Util v3 installed and set GTX 1080 ti to MSI mode and Interrupt Priority to High
* Interrupt Affinity Policy Tool (intPolicy_x64) installed and set GTX 1080 ti to Core 4 and USB HUB to core 2 (first core is 0)
* Latency Monitor installed
Steps before starting a game:
NVidia Control Panel:
GSYNC, Force V-Sync, Prefer maximum performance, preferred refresh rate highest available (however, I've tested a lot of combinations of v-sync nvidia/in-game, fast, adaptative, no-gsync...)
MSI Afterburner:
Cap FPS to 100 (or 60, 90, 98, 120, 142... depending on the monitor and the test I want to do. Also tested with NVCP Framelimiter)
Monitoring/On screen display: CPUx usage, framerate, frametime)
Power Limit 120%, temp limit 90º and GPU Boost
Issue:
It seems that sometimes when the game has to manage something (load a texture, pop up a new item or npc, show an explosion, do a physics calculation...) there's a micro-stuttering/hitch, a spike in frametime.
In some games, like The Witcher 3 (occasionally when riding fast but always when opening trunks, boxes or loot), with all previous configuration the result is quite good and I see isolated framespikes from time to time. In others, like Batman Arkham Knight (specially driving the car or each time a radio conversation appears during game) or Deus Ex: Manking Divided (specially Prague), there are framespikes.
If I cap FPS to 60 spikes disappear in games (however, spikes when opening boxes, for example, are still present, engine maybe?). It makes me think that is something binded to game engines (awful ports maybe?) since when FPS is capped to 90 or 100, for example, none of the cores nor the GPU is reaching 100% nor close. I don't know if its something related to latency that, although i've got pretty good results with Latency Monitor, is not perfect and something is stealing CPU cycles causing interruptions that do not allow CPU to mantain 90 or 100 fps when has to do extra stuff (load texture from SSD, do a physics calculation...).
The situation get worse with Hyperthreading off. Resolution or graphics settings seems to have no impact.
Sometimes it seems to me like some kind of unsync between monitor Hz, v-sync, game engine, cpu ... I don't know... maybe I am driving myself insane.
Here you can see several images that shows the behaviour I've described:
If you need any specific MSI Afterburner monitoring graph, any kind of extra information or want me to test again something that I've already listed, I will do with pleasure.
The fact is that I wanted to buy a brand new RTX 3090 but having the same issue also with this card scares me a lot.
If I cannot find any fix, at least I hope my post helps somebody to improve his performance as I got my best results with these settings.
Thank you very much in advance.
My happiness is in your hands.
I want to share an issue I have since a long long time ago. This issue has obsessed me. I've tried everything to solve it and, although it got better, is still there torturing me and preventing me from enjoying gaming.
The issue is some kind of micro-stutter or hitch that freezes the image (not the sound) just a fraction of second but enough time to break the immersion... frametime spikes.
Let's see, since I've done thousands of tests with different hardware/drivers combinations I'm just exposing current RIG, where the issue still persists but it's the best experience I've been able to achieve. I can say that I've tested almost every from a default clean Windows 10 installation to a custom installation with these specs/settings/configurations:
- i7 8700 5.1 GHZ Delidded (tested also before delid, with stock setting, manual OC, bios profile OC ...)
- Trident Z Neo 32 [2x16] GB RAM DDR4 3600 Mhz CL16-16-16-36 1.35v B-DIE (tested also 1 DIMM, 2 DIMMs, swapped, stock, XMP, manual OC, same with 16 [2x8] GB from Corsair Dominator DDR4 4000 Mhz)
- Asus Maximus X Apex last bios 2402 (tested with bios stock settings, custom settings...)
- Asus Rog PG278Q 2560x1440 144hz (tested also with Asus Rog PG348q 3440x1400 100hz)
- 1x GTX 1080 ti ASUS ROG (tested OC, non-OC, tested with another GTX 1080 ti ASUS ROG in SLI, single, swapped)
- Samsung NMve M.2 970 Pro 512GB (tested in 4 different SSD: samsung, kingston...)
- PSU BeQuiet 1500W Dark Power 11 (tested OC six 12V rails, non-OC one massive 12V rail)
- Water cooling Fractal Design s36 Kelvin
[Note: all configurations have been tested and stressed, no thermal CPU/GPU throlling found]
- Windows 10 Pro 2004
* Opera browser installed
* Ethernet devices disabled
* Audio devices disabled
* GPU/Audio default drivers uninstalled with DDU (safe mode)
* Nvidia Drivers 456.71 installed (tested with 456.55 and olders)
* Chipset driver installed
* ACHI driver installed (when tested in SDD)
* NVM Samsung drivers installed (when tested in NMve M.2)
* MSI Afterburner installed
* Steam installed (overlay disabled)
* GPU-Z installed
* CPU-Z installed
* XBox Game Bar disabled
* Game Mode enabled (tested with it disabled)
* VRR and GPU scheduled enabled (tested with them disabled)
* Windows update disabled
* Windows antivirus disabled (each time I reboot as I haven't been able to disable it forever)
* Firewall disabled
* HPET in device manager disabled
* CMD:
bcdedit /set useplatformclock no - Reiniciar
bcdedit /set useplatformtick yes - Reiniciar
bcdedit /set disabledynamictick yes - Reiniciar
* All not needed windows 10 features from configuration disabled (privacy...)
* All not needed windows 10 applications uninstalled (office, spotify...)
* Several not needed windows 10 services disabled (SysMain, windows update, printers mangement...)
* Process Lasso and Park Cores installed (cores unparked and games with high priority class and I/O priority, CPU affinity to all cores/threads, Application power profile to Bitsum Highest Performance)
* Intelligent standby list cleaner installed and current timer resolution set to 0.5)
* MSI Util v3 installed and set GTX 1080 ti to MSI mode and Interrupt Priority to High
* Interrupt Affinity Policy Tool (intPolicy_x64) installed and set GTX 1080 ti to Core 4 and USB HUB to core 2 (first core is 0)
* Latency Monitor installed
Steps before starting a game:
NVidia Control Panel:
GSYNC, Force V-Sync, Prefer maximum performance, preferred refresh rate highest available (however, I've tested a lot of combinations of v-sync nvidia/in-game, fast, adaptative, no-gsync...)
MSI Afterburner:
Cap FPS to 100 (or 60, 90, 98, 120, 142... depending on the monitor and the test I want to do. Also tested with NVCP Framelimiter)
Monitoring/On screen display: CPUx usage, framerate, frametime)
Power Limit 120%, temp limit 90º and GPU Boost
Issue:
It seems that sometimes when the game has to manage something (load a texture, pop up a new item or npc, show an explosion, do a physics calculation...) there's a micro-stuttering/hitch, a spike in frametime.
In some games, like The Witcher 3 (occasionally when riding fast but always when opening trunks, boxes or loot), with all previous configuration the result is quite good and I see isolated framespikes from time to time. In others, like Batman Arkham Knight (specially driving the car or each time a radio conversation appears during game) or Deus Ex: Manking Divided (specially Prague), there are framespikes.
If I cap FPS to 60 spikes disappear in games (however, spikes when opening boxes, for example, are still present, engine maybe?). It makes me think that is something binded to game engines (awful ports maybe?) since when FPS is capped to 90 or 100, for example, none of the cores nor the GPU is reaching 100% nor close. I don't know if its something related to latency that, although i've got pretty good results with Latency Monitor, is not perfect and something is stealing CPU cycles causing interruptions that do not allow CPU to mantain 90 or 100 fps when has to do extra stuff (load texture from SSD, do a physics calculation...).
The situation get worse with Hyperthreading off. Resolution or graphics settings seems to have no impact.
Sometimes it seems to me like some kind of unsync between monitor Hz, v-sync, game engine, cpu ... I don't know... maybe I am driving myself insane.
Here you can see several images that shows the behaviour I've described:
If you need any specific MSI Afterburner monitoring graph, any kind of extra information or want me to test again something that I've already listed, I will do with pleasure.
The fact is that I wanted to buy a brand new RTX 3090 but having the same issue also with this card scares me a lot.
If I cannot find any fix, at least I hope my post helps somebody to improve his performance as I got my best results with these settings.
Thank you very much in advance.
My happiness is in your hands.
Re: Micro-stuttering / hitch & Frametime spikes issue
It's asset/shader loading from the HDD/SSD.Rachmaninoff wrote: ↑11 Oct 2020, 11:38Issue:
It seems that sometimes when the game has to manage something (load a texture, pop up a new item or npc, show an explosion, do a physics calculation...) there's a micro-stuttering/hitch, a spike in frametime.
There's nothing you can do to eliminate it. It's known, it's normal, and it varies in severity and frequency per game.
(jorimt: /jor-uhm-tee/)
Author: Blur Busters "G-SYNC 101" Series
Displays: ASUS PG27AQN, LG 48C4 Scaler: RetroTINK 4k Consoles: Dreamcast, PS2, PS3, PS5, Switch 2, Wii, Xbox, Analogue Pocket + Dock VR: Beyond, Quest 3, Reverb G2, Index OS: Windows 11 Pro Case: Fractal Design Torrent PSU: Seasonic PRIME TX-1000 MB: ASUS Z790 Hero CPU: Intel i9-13900k w/Noctua NH-U12A GPU: GIGABYTE RTX 4090 GAMING OC RAM: 32GB G.SKILL Trident Z5 DDR5 6400MHz CL32 SSDs: 2TB WD_BLACK SN850 (OS), 4TB WD_BLACK SN850X (Games) Keyboards: Wooting 60HE, Logitech G915 TKL Mice: Razer Viper Mini SE, Razer Viper 8kHz Sound: Creative Sound Blaster Katana V2 (speakers/amp/DAC), AFUL Performer 8 (IEMs)
Author: Blur Busters "G-SYNC 101" Series
Displays: ASUS PG27AQN, LG 48C4 Scaler: RetroTINK 4k Consoles: Dreamcast, PS2, PS3, PS5, Switch 2, Wii, Xbox, Analogue Pocket + Dock VR: Beyond, Quest 3, Reverb G2, Index OS: Windows 11 Pro Case: Fractal Design Torrent PSU: Seasonic PRIME TX-1000 MB: ASUS Z790 Hero CPU: Intel i9-13900k w/Noctua NH-U12A GPU: GIGABYTE RTX 4090 GAMING OC RAM: 32GB G.SKILL Trident Z5 DDR5 6400MHz CL32 SSDs: 2TB WD_BLACK SN850 (OS), 4TB WD_BLACK SN850X (Games) Keyboards: Wooting 60HE, Logitech G915 TKL Mice: Razer Viper Mini SE, Razer Viper 8kHz Sound: Creative Sound Blaster Katana V2 (speakers/amp/DAC), AFUL Performer 8 (IEMs)
- Chief Blur Buster
- Site Admin
- Posts: 12051
- Joined: 05 Dec 2013, 15:44
- Location: Toronto / Hamilton, Ontario, Canada
- Contact:
Re: Micro-stuttering / hitch & Frametime spikes issue
That's one possible effect. Newer games are designed to try to make this invisible (e.g. interstital such as an elevator or an intermediate room between two doors to two large spaces) but sometimes it's hard -- there are definitely no way to eliminate ALL hitching.jorimt wrote: ↑11 Oct 2020, 11:57It's asset/shader loading from the HDD/SSD.Rachmaninoff wrote: ↑11 Oct 2020, 11:38Issue:
It seems that sometimes when the game has to manage something (load a texture, pop up a new item or npc, show an explosion, do a physics calculation...) there's a micro-stuttering/hitch, a spike in frametime.
There's nothing you can do to eliminate it. It's known, it's normal, and it varies in severity and frequency per game.
You did correctly use large amounts of memory (e.g. 32GB RAM+ to help disk caching) combined with a lower-latency SSD (NVMe M.2 SSD such as Samsung Pro series or one of those higher $/GB gumsticks), to try to minimize the hitching from asset loads. So that is covered.
In some cases, SSD can unexpectedly fragment sufficiently to have noticeable performance slowdowns
Your SSD is small -- 512MB. Thus, it probably is near full. Doublecheck unexpectedly-high SSD fragmentation too. Also, if your 970pro is 95% full, write amplification effects may lead to sufficient fragmentation (1000-count fragment to big video game files) whereupon SSD fragmentation really slows down things. A fresh reformat and a policy of keeping ~20+% disk space free, will help ensure extremely fast read/writes without write amplification / fragmentation factors. SSD do not suffer from fragmentation nearly as much as HDD but if 1 megabyte is split up to tons of 64 kilobyte fragments (installing game to a nearly full SSD + write amplification + other factors), it affects benchmarks much like 64KB block read bechmarks instead of 1MB block read benchmarks. You know how SSDs slow down with trying to read thousands of tiny blocks -- the "super-bad-fragmentation" situations of a nearly-full SSD can lead to such a similar performance degradation if your game assets is splintered into many thousands of tiny pieces, creating longer hitching instead of shorter hitching. But if you did a fresh install of the game into a mostly-free SSD (during your SSD testing), then this is probably not your factor. Just an error margin to consider.
Yes, SSD are much more immune to fragmentation slowdowns than HDD, but not 100% immune especially in the refresh rate race to retina refresh rates, and the associated Vicious Cycle Effect (higher Hz making even smaller hitching more visible). Amazing how few games try to concurrentize asset loads (like PlayStation 5 game design does) to make SSD loads more similar to QD32 benchmarks (more fragmentation immune) instead of QD1 (VERY fragmentation sensitive).
In the future, try to buy more SSD than you really need, so you avoid fragmentation-related SSD slowdowns (caused by trying to install a game on a nearly-full SSD, creating game-load benchmarks sometimes similar to SSD 4K-QD1 instead of SSD 1M-QD1) and the associated hitching -- QD1 benchmarks are much slower than QD32 benchmarks, and many older games loaded assets in a Queue Depth 1 speed, so a very heavily fragmented SSD really slows down that particular game-asset-loading behavior.
Too many M.2 SSDs in M.2 slots using processor with too few PCI-X lanes
Make sure you've got enough PCI-X lanes to run your NVMe SSDs at full speed, some motherboards will slow down SSDs if all M.2 slots are filled, be warned. This might not be your factor, unless it's a large asset load (e.g. like between large views), and you have lots of memory so that probably isn't your issue.
Overclocking Threasholds pushed too hard?
Another possibility is overclocking thresholds being pushed a bit hard -- try temporarily backing down about 5 / 10% / 20% / 25% and seeing if hitching reduces. Your framerate might fall a bit but the hitching might suddenly become much shorter (far beyond than the percentage). There's sometimes a degradation-cliff that occurs to certain benchmarks (e.g. I/O or PCI Express or other thing) when things are pushed to their overclock limits, caused by various factors ranging from error correction retransmit logic to thermal throttling, etc. One benchmark may remain fast (e.g. CPU) but another benchmark (e.g. SSD) might degrade, so weird degradations can unexpectedly occur if it's overclocked-nearly-to-crash. So test all your benchmarks everytime you raise overclocking, to make sure no stone is unturned.
If the hitching is more engine-related (e.g. decompressing textures, rendering related, game physics related) rather than disk I/O related, certain things such as variable refresh rate might make the hitching invisible if they're the type of hitching that leads to single-framedrops. I haven't seen the hitching you see, if it's a sustained freeze ("less than one second" seems quite bad and can't be erased by VRR), or if it's a single-framedrop type freeze (those are much more easily hidden by VRR).
So an ultra-high-Hz variable refresh rate monitor may help to hide much of the hitching. Use G-SYNC + VSYNC ON in NVIDIA Control Panel, while concurrently using VSYNC OFF in the game-menus (the opposing-settings technique) to sometimes maximize the reliability of de-stuttering of most game engines -- some engines use more VRR-compatible logic when the in-game menus are configured to VSYNC OFF.
Special Note about Overclock-induced unexpected performance sensitivities
Related to overclocking (since overclocked systems are much more EMI-sensitive), longshot factor might be EMI (99%+ of the time, not the problem) but this is an increasing problem due to the large amounts of error-correcting-and-retransmit systems built into a modern PC. A PCI-Express bus and even GDDR6X is increasingly behaving more packetized with error-detect-and-retransmit, and like an Internet connection with 50% packet loss, heavy overclocks can cause lots of data loss that no longer crashes a computer because of the error correction built into modern systems -- even NVIDIA mentions and acknowledges this. That's why I recommend you try test-backing-off overclocks slightly. Make sure your computer tower is not inches away from power wires, power supplies, appliances, or radio-emitting appliances like WiFi routers, especially if your computer has a transparent window (less shielding), as you know the LG 5K monitor weirdnesses with nearby routers -- that's not isolated. It happens way more often, I've actually seen this happen personally (at much smaller sub-recall-advisory scales like affecting only 1 in 1000 people -- too few to cause a product recall -- but frequent enough that 1000 users out of 1 million are affected) to various computer devices. This happens often when interfering devices are too close to unshielded computer parts (like a WiFi router sitting in front of transparent computer window to an unshielded NVMe M.2 SSD inches away from the router's antennas). It takes strong nearby EMI to do this, but this can cause some subtle benchmark slowdowns since computers sometimes slow down (at edge of limits) before they finally suddenly crash. Imagine a PCI Express bus losing 10% of attempted transfers (like 10% packet loss on a bad Internet connection), since newer cards/buses error-correct missing data instead of simply crashing sooner. Even newer memory such as GDDR6X on RTX 3080 has some things in common with a packet-data connection in that it's error-detect-and-retransmit memory (rather than simple parity ECC). This is probably NOT your problem (it's usually a wild goose chase to many red herrings), but I mention this simply because the increase in error correction has made this begin to happen more frequently for overclocked-to-the-limit computers.
Are you using variable refresh rate?
I assume that you are. Variable refresh rate (VRR) aka G-SYNC / FreeSync is an excellent Magic Eraser of single-framedrop hitching (Demo animation: www.testufo.com/vrr where framerate changes are stutterless), though it can't erase those sustained multi-refresh-cycle disk-load freezes. But typical framdrops such as 60fps-59fps-60fps -- that type of framedrop is rendered completely invisible with VRR. If you are not yet using VRR, you might find that some of the minor hitching disappears, for example, allowing you to better tolerate the remaining longer-hitching effects. Test multiple framerate caps, since sometimes caps occasionally slightly reduce the severity of stutters, since the transition to "equal to above VRR" to "framerate inside VRR range" occasionally has more visible hitching (with certain sync technology combinations) than "capped inside VRR range". Test multiple cap margins (3fps below, 10fps below, 20fps below) to avoid leaving stones unturned with specific games.
A lot of things I have said are longshots and probably won't help since you're probably using VRR -- and your multiple-SSD means you've installed your game freshly to mostly empty space at least a few times -- but I see your heavy overclocks, so I've mentioned a few extras just in case.
Faster GPU will help....to a certain extent
An RTX 3080 can improve hitching by speeding up load-related processing (e.g. faster texture decompressing after loading from SSD). It won't speed up the SSD or the QD1 bottleneck, but it can speed up the processing, so that hitching that lasted 0.25 second (distracting) might now last only 0.1 second (less distracting). Some games will have NO hitching difference, while other games will have MAJOR hitching reductions.
Head of Blur Busters - BlurBusters.com | TestUFO.com | Follow @BlurBusters on: BlueSky | Twitter | Facebook

Forum Rules wrote: 1. Rule #1: Be Nice. This is published forum rule #1. Even To Newbies & People You Disagree With!
2. Please report rule violations If you see a post that violates forum rules, then report the post.
3. ALWAYS respect indie testers here. See how indies are bootstrapping Blur Busters research!
Re: Micro-stuttering / hitch & Frametime spikes issue
Right. I myself have delved into this subject too many times to bother deep diving into blind troubleshooting again, so I'm just skipping to the part where they all end; you can't eliminate all stutter in every instance.Chief Blur Buster wrote: ↑11 Oct 2020, 14:05That's one possible effect. Newer games are designed to try to make this invisible (e.g. interstital such as an elevator or an intermediate room between two doors to two large spaces) but sometimes it's hard -- there are definitely no way to eliminate ALL hitching.
Going by his complete post, he feels he should never see a hitch, which isn't possible unless you limit yourself to playing emulated 8-bit sidescrolling titles (and even then, only if they're properly configured).
Even the latest bleeding edge open world AAA titles are still prone to asset load-related stutter, and no matter how many settings are played with, or how optimized or powerful the system, you're going to get a hitch here or there.
Part of the problem, is this issue simply isn't prioritized on the dev-side. That, and any benefits faster SSDs would provide in reducing asset load stutter aren't taken advantage of at the engine level...yet. Games just see them as plain old HDDs where real-time asset transfer speeds and caching mechanisms are concerned.
Hopefully that changes with the next-gen console ports that should potentially focus more on higher speed loading of assets from the SSD. Right now, the only thing SSD improves in PC games is load times.
The one game I know of that does prioritize SSD performance for real-time asset loading on PC is Star Citizen, and the issue there is it makes HDDs effectively unusable.
(jorimt: /jor-uhm-tee/)
Author: Blur Busters "G-SYNC 101" Series
Displays: ASUS PG27AQN, LG 48C4 Scaler: RetroTINK 4k Consoles: Dreamcast, PS2, PS3, PS5, Switch 2, Wii, Xbox, Analogue Pocket + Dock VR: Beyond, Quest 3, Reverb G2, Index OS: Windows 11 Pro Case: Fractal Design Torrent PSU: Seasonic PRIME TX-1000 MB: ASUS Z790 Hero CPU: Intel i9-13900k w/Noctua NH-U12A GPU: GIGABYTE RTX 4090 GAMING OC RAM: 32GB G.SKILL Trident Z5 DDR5 6400MHz CL32 SSDs: 2TB WD_BLACK SN850 (OS), 4TB WD_BLACK SN850X (Games) Keyboards: Wooting 60HE, Logitech G915 TKL Mice: Razer Viper Mini SE, Razer Viper 8kHz Sound: Creative Sound Blaster Katana V2 (speakers/amp/DAC), AFUL Performer 8 (IEMs)
Author: Blur Busters "G-SYNC 101" Series
Displays: ASUS PG27AQN, LG 48C4 Scaler: RetroTINK 4k Consoles: Dreamcast, PS2, PS3, PS5, Switch 2, Wii, Xbox, Analogue Pocket + Dock VR: Beyond, Quest 3, Reverb G2, Index OS: Windows 11 Pro Case: Fractal Design Torrent PSU: Seasonic PRIME TX-1000 MB: ASUS Z790 Hero CPU: Intel i9-13900k w/Noctua NH-U12A GPU: GIGABYTE RTX 4090 GAMING OC RAM: 32GB G.SKILL Trident Z5 DDR5 6400MHz CL32 SSDs: 2TB WD_BLACK SN850 (OS), 4TB WD_BLACK SN850X (Games) Keyboards: Wooting 60HE, Logitech G915 TKL Mice: Razer Viper Mini SE, Razer Viper 8kHz Sound: Creative Sound Blaster Katana V2 (speakers/amp/DAC), AFUL Performer 8 (IEMs)
- Chief Blur Buster
- Site Admin
- Posts: 12051
- Joined: 05 Dec 2013, 15:44
- Location: Toronto / Hamilton, Ontario, Canada
- Contact:
Re: Micro-stuttering / hitch & Frametime spikes issue
Yep. It's behaving as a QD1. Queue Depth 1 because most games are optimized to that to be compatible with HDDs. Games that optimize for QD32 asset loading (multithreaded asset loading FTW!), are effectively unusable on spinning rust (Star Citizen), since old disks don't like loading 32 files simultaneously. Like Star Citizen you mentioned.
That's why I mentioned SSD fragmentation -- think of SSD slowing down from 1M-QD1 benchmark to 4K-QD1 benchmark, because a game installing to a highly fragmented SSD (caused by write amplification behaviors), can save the game assets in tens of thosuands of 4K-to-64K "memory holes" in the SSD. In this case, asset loading behaves like those 4K-QD1 benchmarks rather than 1M-QD1 benchmarks.
QD32 is more fragmentation-immune, while QD1 is very SSD fragmentation sensitive. (QD32 = request read 32 files at the same time, request read 32 SSD blocks at the same time, request read 32 sections of a large SSD file at the same time, etc)
The Optimize Disks of Windows 10 usually defragments SSDs "just enough" to be fine for most users. It automatically runs already once every month (or week). But a forced manual Optimize pass before AND after installing a game, is usually sufficient to prevent this behavior, especially if you're rapidly testing lots of stuff and reinstalling stuff on new small (512MB) SSDs that almost fill up before finally installing games. Third party SSD defrag software will provide a better fragmentation view to find out whether you've got millions of 4K fragments, or the game data is one contiguous file. Don't defrag often, it wears down the SSD. Pick your poison. Don't worry about moderate fragmentation (e.g. 1GB fragmented into 40 pieces), it's those 1GB game asset files fragmented into 10,000 or 100,000 fragments, that begins to really seriously degrade QD1 performance -- caused by installing a game onto a nearly-full SSD that is already peppered full of holes between files. In this case, asset-loads starts to slow down human perceptibly like QD1-vs-QD32 benchmarks of SSD.
PlayStation 5 will be the first game platform to kinda mandate/force SSD-optimized asset loading -- for virtually hitchless asset streaming. But the vast majority of PC games don't have hitchless SSD asset streaming.
But given the OP have probably installed the games at least once to a nearly-empty SSD (in SSD swaps), so I'm pretty certain fragmentation is not the cause of the lengthening of hitching. The best a SSD defragment (of horribly fragmented game assets when installing on a post-write-amplified SSD) may do is reduce the hitching, from say 1 second freezes to 1/4 second freezes, not completely solve it altogether.
OP did so much troubleshooting already, so I mentioned a few more unturned "Left Field" stones -- just in case. Common advice overlooks SSD fragmentation and overclock-induced slowdowns -- but both exists. Hitching can reduce, but not be completely eliminated altogether.
Expectations are tempered (pun intended: I'm knocking on that clear tempered glass window on the Blur Busters computer...)
Head of Blur Busters - BlurBusters.com | TestUFO.com | Follow @BlurBusters on: BlueSky | Twitter | Facebook

Forum Rules wrote: 1. Rule #1: Be Nice. This is published forum rule #1. Even To Newbies & People You Disagree With!
2. Please report rule violations If you see a post that violates forum rules, then report the post.
3. ALWAYS respect indie testers here. See how indies are bootstrapping Blur Busters research!
Re: Micro-stuttering / hitch & Frametime spikes issue
Yeah, the hope is that it gradually forces a change on PC. That said, it's a risky one, as swapping to such methods will very likely exempt HDDs from being viable performers in such games.Chief Blur Buster wrote: ↑11 Oct 2020, 14:45PlayStation 5 will be the first game platform to kinda mandate/force SSD-optimized asset loading -- for virtually hitchless asset streaming. But the vast majority of PC games don't have hitchless SSD asset streaming.
Many gamers (myself included) rely on small SSDs for OS, and huge HDDs for games, and only load their most played games on their SSDs in rotation. So hopefully it will also make higher capacity, higher speed SSDs the norm, and more importantly, more affordable, because games are certainly not getting any smaller (something else that needs to be addressed; games no longer rely on massive pre-rendered movie files, so all that's really left is increasingly higher res textures. We need better compression methods).
I'm not the least ways oppose to you or anyone else humoring further troubleshooting, and hope if his SSD does need optimizing or overclocks need revisiting, addressing them will alleviate what he's experiencing.Chief Blur Buster wrote: ↑11 Oct 2020, 14:45But given the OP have probably installed the games at least once to a nearly-empty SSD (in SSD swaps), so I'm pretty certain it's not the cause of the lengthening of hitching. The best a SSD defragment (of horribly fragmented game assets when installing on a post-write-amplified SSD) may do is reduce the hitching, from say 1 second freezes to 1/4 second freezes, not completely solve it altogether.
OP did so much troubleshooting already, so I mentioned a few more unturned "Left Field" stones -- just in case. Common advice overlooks SSD fragmentation and overclock-induced slowdowns -- but both exists. Hitching can reduce, but not be completely eliminated altogether.
All I'm saying is, even if it does, as you also mentioned, it won't eliminate hitches, and that's what he and many others seem to be chasing...
(jorimt: /jor-uhm-tee/)
Author: Blur Busters "G-SYNC 101" Series
Displays: ASUS PG27AQN, LG 48C4 Scaler: RetroTINK 4k Consoles: Dreamcast, PS2, PS3, PS5, Switch 2, Wii, Xbox, Analogue Pocket + Dock VR: Beyond, Quest 3, Reverb G2, Index OS: Windows 11 Pro Case: Fractal Design Torrent PSU: Seasonic PRIME TX-1000 MB: ASUS Z790 Hero CPU: Intel i9-13900k w/Noctua NH-U12A GPU: GIGABYTE RTX 4090 GAMING OC RAM: 32GB G.SKILL Trident Z5 DDR5 6400MHz CL32 SSDs: 2TB WD_BLACK SN850 (OS), 4TB WD_BLACK SN850X (Games) Keyboards: Wooting 60HE, Logitech G915 TKL Mice: Razer Viper Mini SE, Razer Viper 8kHz Sound: Creative Sound Blaster Katana V2 (speakers/amp/DAC), AFUL Performer 8 (IEMs)
Author: Blur Busters "G-SYNC 101" Series
Displays: ASUS PG27AQN, LG 48C4 Scaler: RetroTINK 4k Consoles: Dreamcast, PS2, PS3, PS5, Switch 2, Wii, Xbox, Analogue Pocket + Dock VR: Beyond, Quest 3, Reverb G2, Index OS: Windows 11 Pro Case: Fractal Design Torrent PSU: Seasonic PRIME TX-1000 MB: ASUS Z790 Hero CPU: Intel i9-13900k w/Noctua NH-U12A GPU: GIGABYTE RTX 4090 GAMING OC RAM: 32GB G.SKILL Trident Z5 DDR5 6400MHz CL32 SSDs: 2TB WD_BLACK SN850 (OS), 4TB WD_BLACK SN850X (Games) Keyboards: Wooting 60HE, Logitech G915 TKL Mice: Razer Viper Mini SE, Razer Viper 8kHz Sound: Creative Sound Blaster Katana V2 (speakers/amp/DAC), AFUL Performer 8 (IEMs)
- Chief Blur Buster
- Site Admin
- Posts: 12051
- Joined: 05 Dec 2013, 15:44
- Location: Toronto / Hamilton, Ontario, Canada
- Contact:
Re: Micro-stuttering / hitch & Frametime spikes issue
Indeed.jorimt wrote: ↑11 Oct 2020, 15:04Yeah, the hope is that it gradually forces a change on PC. That said, it's a risky one, as swapping to such methods will very likely exempt HDDs from being viable performers in such games.Chief Blur Buster wrote: ↑11 Oct 2020, 14:45PlayStation 5 will be the first game platform to kinda mandate/force SSD-optimized asset loading -- for virtually hitchless asset streaming. But the vast majority of PC games don't have hitchless SSD asset streaming.
Intermediate transition period may be game installers might that rearrange the asset files based on HDD versus SSD [Unpacking Game...27% Complete] as well as as well as have a HDD-optimized asset load mode (infrequent, serialized, major requests) versus SSD-optimized asset streaming mode (frequent, parllized, micro requests).
Intel Optane memory (and its incredible QD1 speeds) also does reduce hitching of legacy games. The OP appears to have the money necessary, so this is an additional suggestion to use a Optane memory as a workaround for QD1-legacy games. Optane isn't much help most of the time, but it does partially mitigate the surgically focussed OP's complaint of QD1-asset-load hitching reductions.
My "No Purchases" prescription is:
- Try not to go beyond ~75-80% full on your SSDs where possible;
- Defrag test. Disk Optimize before/after installing flagship game; so you don't have fragmentation related slowdowns (QD1 sensitivity)
- G-SYNC+VSYNC ON in NVIDIA Control Panel, VSYNC OFF in-game menu, and cap 3fps-below
- Also test RTSS capping instead of in-game capping (might sometimes be laggier but smoother than ingame cap, more hitching reductions occur)
- Potentially minor overclock reduction, *IF* there is a determined overclock-related hitch-amplification factor.
- No spinning rust (except for zip files and backup files), not even a "My Documents" folder on spinning rust, Windows should have zero accesses to the spinning platters. Even a single accidental access to spinning rust because one Windows setting or folder was there, can make/break certain games. Likely not an issue due to 32GB RAM and tons of caching, but...
My "Cost No Object" prescription (in combination to above) is:
- RTX 3080; for faster 3D processing;
- Optane Memory; for fastest QD1 money can buy;
- SSD with fastest QD1 benchmarks you can find. 7 Gbytes/sec is garbage if 4K QD1 is only 100 Kbytes/sec (exaggerating, but... haha)
This WILL NOT completely eliminate hitching, but will reduce it in many games.
Head of Blur Busters - BlurBusters.com | TestUFO.com | Follow @BlurBusters on: BlueSky | Twitter | Facebook

Forum Rules wrote: 1. Rule #1: Be Nice. This is published forum rule #1. Even To Newbies & People You Disagree With!
2. Please report rule violations If you see a post that violates forum rules, then report the post.
3. ALWAYS respect indie testers here. See how indies are bootstrapping Blur Busters research!
- Chief Blur Buster
- Site Admin
- Posts: 12051
- Joined: 05 Dec 2013, 15:44
- Location: Toronto / Hamilton, Ontario, Canada
- Contact:
Re: Micro-stuttering / hitch & Frametime spikes issue
Disappearage of major mid-game spikes is suggestive of:Rachmaninoff wrote: ↑11 Oct 2020, 11:38f I cap FPS to 60 spikes disappear in games (however, spikes when opening boxes, for example, are still present, engine maybe?).
- Potentially reducible via de-overclocking slightly
(capping reduces overclock-related hitching, whether thermal/throttling/RFI related)
- Potentially reducible via Optane memory or RAMdisk
(improved QD1 performance moving hitch noisefloor to higher framerate)
No-Money-Spend Tests
- Test De-overclock. CPU, RAM, GPU.
- Test RAMdrive: Another test is to temporarily reduce your RAM to 16GB, create a 16GB RAMdisk, install your Steam game to the RAMdisk, and test if the hitching reduces. This will help you test-drive Optane before purchasing.
Not 100% will improve, but these test-drives will help guide future money-spends.
Head of Blur Busters - BlurBusters.com | TestUFO.com | Follow @BlurBusters on: BlueSky | Twitter | Facebook

Forum Rules wrote: 1. Rule #1: Be Nice. This is published forum rule #1. Even To Newbies & People You Disagree With!
2. Please report rule violations If you see a post that violates forum rules, then report the post.
3. ALWAYS respect indie testers here. See how indies are bootstrapping Blur Busters research!
Re: Micro-stuttering / hitch & Frametime spikes issue
Ah yes, a similar concept to the more extreme "load the entirety of the game files onto high speed DRAM" (aka RAM Disk) method.Chief Blur Buster wrote: ↑11 Oct 2020, 15:09Intel Optane memory (and its incredible QD1 speeds) also does reduce hitching of legacy games.
EDIT: ha, just saw you mentioned that.
(jorimt: /jor-uhm-tee/)
Author: Blur Busters "G-SYNC 101" Series
Displays: ASUS PG27AQN, LG 48C4 Scaler: RetroTINK 4k Consoles: Dreamcast, PS2, PS3, PS5, Switch 2, Wii, Xbox, Analogue Pocket + Dock VR: Beyond, Quest 3, Reverb G2, Index OS: Windows 11 Pro Case: Fractal Design Torrent PSU: Seasonic PRIME TX-1000 MB: ASUS Z790 Hero CPU: Intel i9-13900k w/Noctua NH-U12A GPU: GIGABYTE RTX 4090 GAMING OC RAM: 32GB G.SKILL Trident Z5 DDR5 6400MHz CL32 SSDs: 2TB WD_BLACK SN850 (OS), 4TB WD_BLACK SN850X (Games) Keyboards: Wooting 60HE, Logitech G915 TKL Mice: Razer Viper Mini SE, Razer Viper 8kHz Sound: Creative Sound Blaster Katana V2 (speakers/amp/DAC), AFUL Performer 8 (IEMs)
Author: Blur Busters "G-SYNC 101" Series
Displays: ASUS PG27AQN, LG 48C4 Scaler: RetroTINK 4k Consoles: Dreamcast, PS2, PS3, PS5, Switch 2, Wii, Xbox, Analogue Pocket + Dock VR: Beyond, Quest 3, Reverb G2, Index OS: Windows 11 Pro Case: Fractal Design Torrent PSU: Seasonic PRIME TX-1000 MB: ASUS Z790 Hero CPU: Intel i9-13900k w/Noctua NH-U12A GPU: GIGABYTE RTX 4090 GAMING OC RAM: 32GB G.SKILL Trident Z5 DDR5 6400MHz CL32 SSDs: 2TB WD_BLACK SN850 (OS), 4TB WD_BLACK SN850X (Games) Keyboards: Wooting 60HE, Logitech G915 TKL Mice: Razer Viper Mini SE, Razer Viper 8kHz Sound: Creative Sound Blaster Katana V2 (speakers/amp/DAC), AFUL Performer 8 (IEMs)
- Chief Blur Buster
- Site Admin
- Posts: 12051
- Joined: 05 Dec 2013, 15:44
- Location: Toronto / Hamilton, Ontario, Canada
- Contact:
Re: Micro-stuttering / hitch & Frametime spikes issue
Optane memory is not as important to most, but for those anal about QD1 hitching in legacy games, Optane is a help for those 0.1% worst-frametime moments (frametimes affected by asset loads).
Benchmark software often miss those. But before spending, the 32GB RAM is a golden opportunity to temporarily test a RAMdisk just for one game, to determine if it erases the specific surgical complaint by raising the hitching noisefloor to higher framerate caps. Find the biggest, baddest, hitchingest game that fits in the RAMdrive space, and test that baby.
Be noted to test with games that has otherwise no performance difference for 16GB-vs-32GB, it is known that 32GB occasionally helps the 0.1% worst-frametimes (asset-load impacted frametimes) in some games thanks to the improved caching especially for doing repeat-operations like a repeat level load.
It's like how benchmarks only improve from say, 123.1fps to 124fps when upgrading RAM+Optane, but bring out the magnifying glass, it's all those reductions in asset-load related hitching moments. Framerate mostly unchanged but the rare hitching has reduced in lengths/frequency/visibility. But sometimes the user complaint is THE hitching. Surgical mitigation FTW!
Many people who spend $5000 on a computer, checks Optane benchmarks, and dismiss them....not realizing what they're picky about (QD1 asset-load-related hitching).
Benchmark software often miss those. But before spending, the 32GB RAM is a golden opportunity to temporarily test a RAMdisk just for one game, to determine if it erases the specific surgical complaint by raising the hitching noisefloor to higher framerate caps. Find the biggest, baddest, hitchingest game that fits in the RAMdrive space, and test that baby.
Be noted to test with games that has otherwise no performance difference for 16GB-vs-32GB, it is known that 32GB occasionally helps the 0.1% worst-frametimes (asset-load impacted frametimes) in some games thanks to the improved caching especially for doing repeat-operations like a repeat level load.
It's like how benchmarks only improve from say, 123.1fps to 124fps when upgrading RAM+Optane, but bring out the magnifying glass, it's all those reductions in asset-load related hitching moments. Framerate mostly unchanged but the rare hitching has reduced in lengths/frequency/visibility. But sometimes the user complaint is THE hitching. Surgical mitigation FTW!
Many people who spend $5000 on a computer, checks Optane benchmarks, and dismiss them....not realizing what they're picky about (QD1 asset-load-related hitching).
Head of Blur Busters - BlurBusters.com | TestUFO.com | Follow @BlurBusters on: BlueSky | Twitter | Facebook

Forum Rules wrote: 1. Rule #1: Be Nice. This is published forum rule #1. Even To Newbies & People You Disagree With!
2. Please report rule violations If you see a post that violates forum rules, then report the post.
3. ALWAYS respect indie testers here. See how indies are bootstrapping Blur Busters research!