CSGO stuttering every 5 seconds mystery (ex. 5800x3d stutter topic)

[tonypivo]

Have you tested with RTX 2060 and older drivers?

nope, did not have a chance

I found this post exactly because I switched from a 10600kf to a 5800x3d and started experiencing these crashes. Previously with the 10600KF I didn't have micro stutters, at least it wasn't noticeable despite having about 120 fps less. I only play pubg, my memories are 4000mhz CL16 b die (I used them at 10600kf), now in 5800x3d I use normally at 4000mhz with infinity fabric 2000mhz. I downloaded warzone and some other games, and everything runs fine... I've tried everything and in pubg the stutters continue, they are not all the time, but they are annoying and the gameplay is not smooth. I believe it is a cpu problem with the game, watching some videos on youtube you can see that the 1 percent low is always low. I believe that for PUBG intel it is a better choice, for example the 13700k, it has even lower performance in AVG fps, but in gameplay smoothness it is superior. If I had known this earlier, I would not have taken the 5800x3d, maybe a 13700kf and a b660 and the value would be very close.
if anyone who has the 5800x3d knows anything to improve, or say if the experience is similar, I would be grateful.. I keep following

do you also see the same behaviour in cs go?

 

[Bloax]

I like how the first page has a bunch of esoterica

Type "Exploit Protection" into the search box in win 10.

Set Control Flow Guard (CFG) to Off by default. Restart and test.

Also you can set the GPU to MSI mode with High interrupt setting and it'll change the GPU from queue polling to direct.
Get msi tool from here, run as admin, check the box for your gpu if it isn't already, set priority to high, apply in top right and restart.

meanwhile not a single person has had the thought cross -
"yo, could you show your ZenTimings?"

ZenTimings

ZenTimings is a simple and lightweight app for monitoring memory timings on Ryzen platform.

zentimings.protonrom.com

Because I certainly know a thing that could cause stuttering on AM4 - just look at this "stock" behaviour on some boards;

That some-THING is fabric overvoltage - which causes microstutters in the form of aggressive error correction - combined with desyncing the memory/fabric/memory controller (m-, f-, u- clk) frequencies, for extra bad and inconsistent performance.

If you're running DDR4-3200, then you can simply set the fabric clock to 1600, and enable VDDG voltage control - then make sure you're running 1.0v SOC, 0.9v IOD, 0.8v CCD - and it should usually "just" work.
I say "make sure" because my Asrock board wouldn't actually apply the SOC voltage via the "SOC Voltage" option, you had to go to advanced voltage controls and set the SOC VRM voltage to manual, and then set the appropriate voltage there.
Lower FCLKs (under 1800) are not super demanding in terms of dialing in the 0.01v precise combo of SOC+IOD+CCD voltages (yes, all three at once) to make the CPU not choke on its own vomit constantly. FCLKs above 1800, do exhibit this behaviour.

If you're curious on how to dial in high FCLKs, then this is a short summary:

[Official] AMD Ryzen DDR4 24/7 Memory Stability Thread

www.overclock.net

I should also note that there is relatively little gain above 1800 FCLK, so unless you actually enjoy watching paint dry - you can save yourself a lot of hassle and not run 2000 FCLK like a silly person.
if you're for some reason "someone who knows better", then I would like to inform you that the "silly person" running 2000 FCLK and figuring out "it actually performs the same" as well as learning the "why"
|
that silly person is I. (and yes later on I even figured out how to reduce the RRD/WTR values by cranking VDD18 and lowering procODT)


Why is there little gain? Internal, invisible power throttling, due to power limits which FCLK/UCLK/MCLK/vSOC/vIOD/vCCD all have an impact on. (obviously also CPU vCore + frequency)
This "invisible" limit is the same reason that you won't actually see gains from raising CPU frequency above 4700-4800 MHz, unless you run the CPU sub-ambient - thus reducing electrical resistance, meaning less power draw, more headroom until you hit invisible power limits ..

p.s. VDDP voltage is memory controller voltage, and should be perfectly content around 0.85v (or 850 mV if your motherboard uses that scale) for DDR4 3200-3600 memeory
you can see a bunch of values for it on the zen3d/zen3 sections of this spreadsheet:

Zen RAM OC Leaderboards

docs.google.com

they mostly stay below 1.0v, and never 1.1v, which boards for some demented reason usually default to

 

[lucasre5]

nope, did not have a chance

do you also see the same behaviour in cs go?

No, in CS go everything runs fine bro. Just PUBG.

I like how the first page has a bunch of esoterica



meanwhile not a single person has had the thought cross -
"yo, could you show your ZenTimings?"

ZenTimings

ZenTimings is a simple and lightweight app for monitoring memory timings on Ryzen platform.

zentimings.protonrom.com

Because I certainly know a thing that could cause stuttering on AM4 - just look at this "stock" behaviour on some boards;
View attachment 284579

That some-THING is fabric overvoltage - which causes microstutters in the form of aggressive error correction - combined with desyncing the memory/fabric/memory controller (m-, f-, u- clk) frequencies, for extra bad and inconsistent performance.

If you're running DDR4-3200, then you can simply set the fabric clock to 1600, and enable VDDG voltage control - then make sure you're running 1.0v SOC, 0.9v IOD, 0.8v CCD - and it should usually "just" work.
I say "make sure" because my Asrock board wouldn't actually apply the SOC voltage via the "SOC Voltage" option, you had to go to advanced voltage controls and set the SOC VRM voltage to manual, and then set the appropriate voltage there.
Lower FCLKs (under 1800) are not super demanding in terms of dialing in the 0.01v precise combo of SOC+IOD+CCD voltages (yes, all three at once) to make the CPU not choke on its own vomit constantly. FCLKs above 1800, do exhibit this behaviour.

If you're curious on how to dial in high FCLKs, then this is a short summary:

[Official] AMD Ryzen DDR4 24/7 Memory Stability Thread

www.overclock.net

I should also note that there is relatively little gain above 1800 FCLK, so unless you actually enjoy watching paint dry - you can save yourself a lot of hassle and not run 2000 FCLK like a silly person.
if you're for some reason "someone who knows better", then I would like to inform you that the "silly person" running 2000 FCLK and figuring out "it actually performs the same" as well as learning the "why"
View attachment 284582|
that silly person is I. (and yes later on I even figured out how to reduce the RRD/WTR values by cranking VDD18 and lowering procODT)


Why is there little gain? Internal, invisible power throttling, due to power limits which FCLK/UCLK/MCLK/vSOC/vIOD/vCCD all have an impact on. (obviously also CPU vCore + frequency)
This "invisible" limit is the same reason that you won't actually see gains from raising CPU frequency above 4700-4800 MHz, unless you run the CPU sub-ambient - thus reducing electrical resistance, meaning less power draw, more headroom until you hit invisible power limits ..

p.s. VDDP voltage is memory controller voltage, and should be perfectly content around 0.85v (or 850 mV if your motherboard uses that scale) for DDR4 3200-3600 memeory
you can see a bunch of values for it on the zen3d/zen3 sections of this spreadsheet:

Zen RAM OC Leaderboards

docs.google.com

they mostly stay below 1.0v, and never 1.1v, which boards for some demented reason usually default to

here is my zentimings, what could i do to improve? decrease the frequency and the IF? my memories are samsumb b die 4000mhz CL16, and both using them as well as going down to 3600mhz (cl14) and IF 1800 the small crashes in pubg continue.

 

[tonypivo]

I like how the first page has a bunch of esoterica



meanwhile not a single person has had the thought cross -
"yo, could you show your ZenTimings?"

ZenTimings

ZenTimings is a simple and lightweight app for monitoring memory timings on Ryzen platform.

zentimings.protonrom.com

Because I certainly know a thing that could cause stuttering on AM4 - just look at this "stock" behaviour on some boards;
View attachment 284579

That some-THING is fabric overvoltage - which causes microstutters in the form of aggressive error correction - combined with desyncing the memory/fabric/memory controller (m-, f-, u- clk) frequencies, for extra bad and inconsistent performance.

If you're running DDR4-3200, then you can simply set the fabric clock to 1600, and enable VDDG voltage control - then make sure you're running 1.0v SOC, 0.9v IOD, 0.8v CCD - and it should usually "just" work.
I say "make sure" because my Asrock board wouldn't actually apply the SOC voltage via the "SOC Voltage" option, you had to go to advanced voltage controls and set the SOC VRM voltage to manual, and then set the appropriate voltage there.
Lower FCLKs (under 1800) are not super demanding in terms of dialing in the 0.01v precise combo of SOC+IOD+CCD voltages (yes, all three at once) to make the CPU not choke on its own vomit constantly. FCLKs above 1800, do exhibit this behaviour.

If you're curious on how to dial in high FCLKs, then this is a short summary:

[Official] AMD Ryzen DDR4 24/7 Memory Stability Thread

www.overclock.net

I should also note that there is relatively little gain above 1800 FCLK, so unless you actually enjoy watching paint dry - you can save yourself a lot of hassle and not run 2000 FCLK like a silly person.
if you're for some reason "someone who knows better", then I would like to inform you that the "silly person" running 2000 FCLK and figuring out "it actually performs the same" as well as learning the "why"
View attachment 284582|
that silly person is I. (and yes later on I even figured out how to reduce the RRD/WTR values by cranking VDD18 and lowering procODT)


Why is there little gain? Internal, invisible power throttling, due to power limits which FCLK/UCLK/MCLK/vSOC/vIOD/vCCD all have an impact on. (obviously also CPU vCore + frequency)
This "invisible" limit is the same reason that you won't actually see gains from raising CPU frequency above 4700-4800 MHz, unless you run the CPU sub-ambient - thus reducing electrical resistance, meaning less power draw, more headroom until you hit invisible power limits ..

p.s. VDDP voltage is memory controller voltage, and should be perfectly content around 0.85v (or 850 mV if your motherboard uses that scale) for DDR4 3200-3600 memeory
you can see a bunch of values for it on the zen3d/zen3 sections of this spreadsheet:

Zen RAM OC Leaderboards

docs.google.com

they mostly stay below 1.0v, and never 1.1v, which boards for some demented reason usually default to

hey,
thanks for the detailed message.

here are my zen timings. what should be changed here?

Maybe I need to change the motherboard to a decent one that is proven to work fine with the 5800x3d? I

 

[3x0]

here are my zen timings. what should be changed here?

here is my zentimings, what could i do to improve?

Everything is OK there

 

[tonypivo]

what a mystery...

 

[3x0]

what a mystery...

I thought we concluded it's 99% nV drivers in your case?

 

[tonypivo]

I thought we concluded it's 99% nV drivers in your case?

I can't prove otherwise, but it is strange that there are no similar reports in nvidia forums for this.
Anyways, it's a guessing game at this point since I have no other option to test.

Hopefully will update this post with the fix someday

 

[3x0]

I can't prove otherwise, but it is strange that there are no similar reports in nvidia forums for this.

Maybe they are all on other forums blaming other components?

 

[tonypivo]

Maybe they are all on other forums blaming other components?

lol yea, I have a thread there as well.
I just wanna game in peace and put an end to this issue

 

[Bloax]

hey,
thanks for the detailed message.

here are my zen timings. what should be changed here?

View attachment 284686

Maybe I need to change the motherboard to a decent one that is proven to work fine with the 5800x3d? I

At 1800 FCLK, the usual range of vSOC requirements is 0.98-1.05v, IOD is usually 50-100 mV below SOC (e.g. 0.98vSOC, 0.93vIOD), and CCD is shockingly low on single-CCD CPUs. (easily -200 mV compared to vSOC, i.e. 0.78-0.85)

So at first I would shoot it with 1.04v SOC, 0.94v IOD, 0.79v CCD - and see if that improves things.

If you feel adventurous, you can also try setting just tFAW to 16 - as it conveniently runs tRRD_s 4 already - but let's not get too lost in memory overclocking of mystery Hynix or Micron chips hehehehehehehehe...

No, in CS go everything runs fine bro. Just PUBG.



here is my zentimings, what could i do to improve? decrease the frequency and the IF? my memories are samsumb b die 4000mhz CL16, and both using them as well as going down to 3600mhz (cl14) and IF 1800 the small crashes in pubg continue.

1.2 VDDP is pretty bad for your memory controller ("it's getting fried" bad), so setting that to 0.98v would be a start, or 0.90v if you're gonna run ddr4-3600 / 1800 FCLK

Besides that, the advice for 1800 FCLK above remains the same - although since you're running Sammy B sticks (obvious by tRCD 16 at DDR4-4000), you can whack in these pretty safely:

tRAS 32 (tCL + tRCD)
tRC 48 (tRAS + tRP)
(""there are no magic formulas for these"" yeah but it works consistently and is easy to follow, so i don't care)
tRRD_s -> 4
tRRD_L ->6
tFAW -> 16
tWTR_s -> 4
tWTR_L -> 12
RDRD and WRWR _SCL, 4
RDWR and WRRD are fine at 8 and 3

and tRFC 336, tRFC2 250, tRFC4 154
[running tRFC as 7x tRC, (tRFC/1.346), (tRFC/2.1875) - because it's convenient and tends to work good]
tWR and tRTP are two timings that require a bunch of manual testing to find which values those DIMMs want, but their impact on performance is hard to observe.

if you're gonna pay for fast memory, you should be running the fast memory as fast memory, cough cough
and if you're curious why I just spit that out, then it's because practically anyone running those memory chips runs similar settings.

There's a high chance that from ddr4-4000 -> 3600, you drop close to 0.12vDIMM and thus should be at around 1.33-1.35v running 3600 16-16-16, but obviously that requires testing.

 

[3x0]

@lucasre5 Which GPU are you using?

 

[tonypivo]

At 1800 FCLK, the usual range of vSOC requirements is 0.98-1.05v, IOD is usually 50-100 mV below SOC (e.g. 0.98vSOC, 0.93vIOD), and CCD is shockingly low on single-CCD CPUs. (easily -200 mV compared to vSOC, i.e. 0.78-0.85)

So at first I would shoot it with 1.04v SOC, 0.94v IOD, 0.79v CCD - and see if that improves things.

I have tried this. I do not see changes in the stuttering of cs go

 

[Huubsterr]

I came across this forum while looking for this exact issue. Looks like there is no sulution yet..

I just 'upgraded' from a perfectly working 5800x to a 5800x3d and also experience the stuttering. Never experienced this before in my former setup.

I use a Strix 4080 on a x570 Mobo, the Asus ROG Crosshair Dark Hero. Just inserted a hardware TPM module too after reading about it, but the stuttering is still there.

I do have the stuttering in Warzone 2.0, quite often actually, short but multiple times per minute. I did not go back to the 5800x yet, will probably do so soon and I hope it's gone then. If so, the 5800x3d will go back to sender unfortunately.

 

[3x0]

I came across this forum while looking for this exact issue. Looks like there is no sulution yet..

I just 'upgraded' from a perfectly working 5800x to a 5800x3d and also experience the stuttering. Never experienced this before in my former setup.

I use a Strix 4080 on a x570 Mobo, the Asus ROG Crosshair Dark Hero. Just inserted a hardware TPM module too after reading about it, but the stuttering is still there.

I do have the stuttering in Warzone 2.0, quite often actually, short but multiple times per minute. I did not go back to the 5800x yet, will probably do so soon and I hope it's gone then. If so, the 5800x3d will go back to sender unfortunately.

Can you post CapFrameX results from a session of Warzone? I'll post mine to compare, running a 6600XT. Only stutter I get is server lag, which is common for WZ.

Edit: old results, from Season 1 in december
5600X, 3 runs


5800X3D, 3 runs


Settings I used are medium/high in these runs, which are difficult for my 6600XT at 1080p.

And again, it's possible nV drivers are to blame. All of the people in this thread are using nV...

 

[Huubsterr]

Can you post CapFrameX results from a session of Warzone? I'll post mine to compare, running a 6600XT. Only stutter I get is server lag, which is common for WZ.

Edit: old results, from Season 1 in december
5600X, 3 runs
View attachment 285062View attachment 285063View attachment 285064

5800X3D, 3 runs
View attachment 285065View attachment 285066View attachment 285067

Settings I used are medium/high in these runs, which are difficult for my 6600XT at 1080p.

And again, it's possible nV drivers are to blame. All of the people in this thread are using nV...

I've never done these type of recordings actually. With my 5800x Warzone was just running smooth as butter on my Gsync monitor. It's purely based on subjective experience, but with the 5800x3d that's clearly not the case anymore. Online I see stutters of 0.2 or 1s stutters, mine are shorter though, just quick short microstutters, but quite often.

Regarding drivers, I've been reading about it and encountered this issue already reported in early 2022, so I'm not sure I'm willing to wait for a fix on this, while I can still return my brand new 5800x3d..

I really have my doubts whether it is NVidia or not. Due to the fact that few lucky people have found random solutions, I can't find a pattern in the fixes, apart from disabling fTPM or using a hardware TPM chip, which both didn't work for me..

 

[tonypivo]

Funny thing, I have been playing warzone fine for some time now until yesterday. A few days after the update it was not stuttering, then yesterday I started stuttering, not often, about once in 5 mins. So sick of this bs

 

[tabascosauz]

I've never done these type of recordings actually. With my 5800x Warzone was just running smooth as butter on my Gsync monitor. It's purely based on subjective experience, but with the 5800x3d that's clearly not the case anymore. Online I see stutters of 0.2 or 1s stutters, mine are shorter though, just quick short microstutters, but quite often.

Regarding drivers, I've been reading about it and encountered this issue already reported in early 2022, so I'm not sure I'm willing to wait for a fix on this, while I can still return my brand new 5800x3d..

I really have my doubts whether it is NVidia or not. Due to the fact that few lucky people have found random solutions, I can't find a pattern in the fixes, apart from disabling fTPM or using a hardware TPM chip, which both didn't work for me..

Did you clean install Windows after upgrading? I had widespread stuttering across all games including MW19 when I upgraded from 5900X to 5800X3D. No other performance issues showed up anywhere in benchmark scores or behaviour. Logic dictates that it should not be necessary, but the difference was plain as day.

The fTPM problem is more "lag" than stuttering, impossible to mistake one for the other.

 

[tonypivo]

Did you clean install Windows after upgrading? I had widespread stuttering across all games including MW19 when I upgraded from 5900X to 5800X3D. No other performance issues showed up anywhere in benchmark scores or behaviour. Logic dictates that it should not be necessary, but the difference was plain as day.

The fTPM problem is more "lag" than stuttering, impossible to mistake one for the other.

Not asking me but. I have tried windows 10 and windows 11 and have stutters still. maybe I need to get some kind of old windows update and try

 

[LabRat 891]

I don't recall seeing any mentions of checking power connections and whether there's a riser involved in the build or not.
I've had a Vega 64 act really weird (regular, momentary, 'performance crash'/stutter), when it was on a long riser w/ it's own separate PSU. I assumed there was a 'ground loop / ground-plane disparity' issue or the external PSU couldn't handle the rapid changes is current-load and/or transients. (we need some PSU info, nV's high end 30xx and 40xx cards are infamous for their milisecond-scale extreme current draw)
BTW, The Vega went back to normal performance when inserted into my X570's Chipset-conn. x4 slot. While not ideal, have you tried a different slot? -for troubleshooting's sake.

I've also had similar issues caused by bad power extensions, and poorly-seated power connectors. [You'd be surprised how much a little bit of extra impedance/resistance can affect a power-hungry GPU.]

I like how the first page has a bunch of esoterica



meanwhile not a single person has had the thought cross -
"yo, could you show your ZenTimings?"

ZenTimings

ZenTimings is a simple and lightweight app for monitoring memory timings on Ryzen platform.

zentimings.protonrom.com

Because I certainly know a thing that could cause stuttering on AM4 - just look at this "stock" behaviour on some boards;
{/truncated quote]

I adore finding those 'esoteric' "proposed solutions". More than a few times, many years after such is posted, those 'oddball' suggestions have been helpful in fixing things.

 

[tonypivo]

I have rechecked my connections and everything seems ok. I have reseated the GPU several time, as well as testing another 2060 GPU and still stuttered. I was thinking my 750w is not enough for a 4080, but it should handle a 2060

 

[izy]

You should try ram at 3200mhz.
Install the game on a different drive, i had shutters because of a bad ssd.

 

[tonypivo]

You should try ram at 3200mhz.
Install the game on a different drive, i had shutters because of a bad ssd.

why 3200?

 

[izy]

why 3200?

To be sure it is not a RAM / IC related problem.

 

[3x0]

I've never done these type of recordings actually. With my 5800x Warzone was just running smooth as butter on my Gsync monitor. It's purely based on subjective experience, but with the 5800x3d that's clearly not the case anymore. Online I see stutters of 0.2 or 1s stutters, mine are shorter though, just quick short microstutters, but quite often.

Here's an entire match on Ashika Island, apart from a few spikes to 30-ish miliseconds, I see no stutters. The only way to figure the problem out is with concrete evidence, so download CapFrameX and post your data for comparison.