AM4 Help thread - stuttering and crash fixes, tips and tricks.

[Mussels]

We've had a lot of AM4 users popping by with similar problems lately, felt like it deserved a dedicated thread.

It's key to remember that what works for one person may not work for you, as the smallest difference in hardware or software can change everything.
No one here can help you with information you didn't share.
There is a lot of generic, universal causes for stutter that aren't covered here - that's why you're trying to test with a clean OS and as little extra hardware and software as possible.
Nothing here will help you if you've got wallpaper engine changing your aero theme and stuttering, or your smart TV is spamming your network
Make sure it's not something like a GPU overclock or undervolt, run stock clocks while troubleshooting.

Before you start:
#1. Find out the details of your PC's hardware.
Put them here, in your system specs. Change the dropdown to "yes" so we can see them in every post you make.

https://www.techpowerup.com/forums/account/specs

We use the "at" symbol to show overclocks, such as "DDR4 3600 @ 3866"

Spoiler: Useful/required software

CPU-Z is great for identifying your CPU, motherboard and RAM.
GPU-Z is great for identifying every possible detail about your GPU, but is less likely to be relevant to this thread.
Zentimings shows you the super fine details of how your RAM is currently running, including voltages your BIOS may not show you
HWinfo64 is a top tier program for its 'sensors only' mode because it shows minimum and maximum values - you can use it to spot something that overheated for even a millisecond
Cinebench R23 is a fantastic CPU benchmark program, and a very small download
Win-Shift-S opens the windows snipping tool for click-and-drag screenshots in your 'pictures' folder that you can directly paste into the TPU forums.

#2. Using those programs, find out what the hardware is *actually* running at.
DDR4 3600 at 2133 isn't helping you, and using HWinfo64 to see the maximum CPU clocks under heavy load is a fantastic way to identify if it's reaching it's full speeds or not.

It's important you take the HWinfo screenshot while stress test programs are running as Zen3 CPU's can poll as fast as every 1ms - 1000 times a second. They can cool down or change clocks a dozen times before you can blink.

Getting a screenshot of maximum values after a game can be useful for something like fan speeds, temperatures and maximum clock speeds - but effective clocks and average clocks will be meaningless.

#3. What you can try do before asking for help:

• Load optimized defaults in your BIOS, disable PBO and XMP for now. You never know what "Auto" means on some boards.
• Use a spare SSD for a truly clean windows 10/11 install, if you can.
• Debloat your OS. Use task manager and disable startup programs, uninstall background software like RGB software, wallpaper changers, anything that runs in the background. One drive syncing things to the cloud could cause ping spikes uploading your save games while still gaming, for example.
• Install the latest AMD chipset drivers
• Disable Vsync or run an FPS limiter under your refresh rate, at least for the duration of testing.
• Make sure your RAM is installed in the correct slots - counting from the CPU, it's usually slots 2 and 4.
• Disconnect from your network and test your issue. Local network traffic can cause issues too, so disconnect from the NETWORK, not the internet.
• Look and listen for any obvious problems - broken fans, intakes clogged with dust, etc.
• Don't forget the obvious - Heat! a CPU stress test doesn't have a GPU's heat as well, so do game testing as well.

Common causes for stuttering on AM4:

To access any advanced PBO features turns *all* of them on.
This often leads people to thinking they changed just one value, when they changed a dozen - enabling curve undervolting also turns on the PBO power limiters, and they often default to maximum balls-to-the-wall values.
This can lead to overheating the CPU or the motherboard VRMs or triggering a known bug in the AGESA code with a too-high EDC value.
(The key to fixing that bug is an EDC limit of 90A or lower on 65W CPUs and 140A or lower, for higher wattage CPUs)


You can test your VRM's and CPU following the guide below, including how to post the relevant information here.
Test with PBO disabled first, then enabled (with custom values set!) so we can compare the results. The heat and wattage may not be worth it.

Spoiler: CPU and VRM overheating testing

The 5800x, 5900x and 5950x all use the same default values with PBO off, but the 5950x can jump from 140W to over 250W just by enabling PBO.

Test with PBO disabled first, and save your screenshots. The score matters too, so you can compare before and after performance.

Keep HWinfo64 open while running Cinebench R23. Make sure to tick "advanced benchmark" and use at least the 10 minute test.
While the test is running, you want to see some specific values and get them all screenshot at the same time from HWinfo64
Make sure to enable snapshot polling in the settings on the right.

These arrow keys let you expand into more columns, so everything fits on screen. Don't let important information get cut off! Make sure your CPU's model is visible, for example.

What matters here is the *CURRENT* temperatures, clock speeds, wattages and effective clock speeds while Cinebench is still running.
Normally the motherboard section will include a VRM temperature sensor even if it's not labelled as such, so it's important to show those.

If your CPU is over 80C, your chipset over 70C, or your effective clocks and current clocks do not match - you're throttling.
Ask for help, and we can figure out why it's throttling and go from there.

From this, we can see my 5800x3D was running at 4.45GHz (full speed for this CPU), effective clocks matched up at 70c and 113W PPT.
There's no signs of any throttling or issues here.
Motherboard and DRAM temps are low too, but not all DRAM has those temp sensors.


Once you've got your results saved, you can now re-test with some custom PBO values, to allow you access to the other PBO features
You can still use these on a 5950x - they'll just limit your multi threaded performance close to PBO disabled values.
You can tune the values higher later, once you know these aren't triggering any thermal throttling or EDC bugs.


My 5800x values are:
PPT: 120W
TDC: 75A
EDC: 110A

If you want full gaming performance with far lower temperatures and wattages (great for weaker motherboards)
PPT: 95W
TDC:60A
EDC: 90A

For a 65W CPU like the 5600x, these are pretty much maxing it out before that AGESA bug hits you. If you have a really bad motherboard VRM setup, enable eco mode instead.
PPT - 88W
TDC - 60A
EDC - 90A

USB disconnects, black screen crashes, crackling audio, GPU driver resets.

The short version:
PCI-E 3.0 riser cables forced to 4.0 speeds and high-density, high-speed RAM are the two primary causes for this with high power draw USB devices a close third.
Any errors on the IO Die can cause it to reset, dropping out any connected device - the errors can be seemingly unrelated.

Out of the box, Ryzen 5000 CPU's are only guaranteed to reach 3200MT/s with two single rank sticks. Any more is overclocking the infinity fabric and straining your SoC.
Just because the motherboard says it can do higher doesn't mean the CPU can.

With overclocking and voltage tweaking -yes XMP is overclocking- Zen 3 (Ryzen 5000 series) can normally reach 3600 to 3800, with above that being extremely rare.

If you're having trouble with stability, post ZenTimings and CPU-Z screenshots.
CPU-Z shows JEDEC and XMP values, while ZenTimings shows what things are really running at.
You can see here that i've overclocked from 3600MT/s (1800MHz) to 3866MT/s (1933MHz) with 1.15v SoC (drooping to 1.13v)

(My RAM normally doesnt have a 3800 XMP profile, I added that myself with an SPD editor and you should pretend it's not there, that's a topic for another thread.)

Spoiler: The long version

Ryzen CPU's have two seperate kinds of chiplets on the CPU itself- the IO die, and the CPU dies.
Together, it's called an SoC - System on Chip.
The CCX has the CPU cores you're familiar with, while the IO die or IOD has the PCIe lanes, the memory controllers, the SATA ports, the connection to the chipset, and helps control power delivery. In all AM4 generations, AMD pairs one of its IO dies with up to two 8-core chiplets.

Any errors on one part of the IO die, can cause one or all of the other parts to error to drop out.

The most common setup problem on Ryzen systems since it's launch is issues with memory compatibility and common misunderstandings of supported RAM speeds.
AMD Has improved their publicly available information on this recently, listing supported RAM configurations on the CPU product pages.
This is from the 5800x3D's information on the AMD website

What this means is that the 5800x3D, the latest and most recent AM4 CPU only officially supports 3200MT/s at best with just two memory sticks.
Any higher is overclocking, no matter what your motherboard or RAM packaging says.

Ryzens Infinity fabric raises speed in sync with DRAM speed, meaning that higher RAM speeds draws more power and risks more SoC instability.

At Ryzens launch, it was found that odd-value timings are harder to run such as ram with 15-15-15-17 timings so AMD introduced "Gear down mode" (GDM) to round up the values without performance loss, but this mode doesn't automatically enable under 2667MT/s - you may need to manually enable this, especially if you aren't using XMP.

How to fix this?
Depending on the cause of your problem the solutions are usually simple enough.

• If you're using a PCI-E riser cable for your GPU, lock your GPU to PCI-E 3.0 in the BIOS
• Remove high power draw devices from your USB ports, including internal ones on the motherboard headers.
• Set up your RAM correctly. Enable GDM and XMP then lower the speed for testing. This lets the timings and DRAM voltage be correct, regardless of the speed.
• Raise your SoC voltage to 1.10v, or 1.15v for more demanding RAM setups.
• Raise the other voltages if necessary, as listed below. Defaults vary between motherboards, so manually setting them can help a lot.

I'll quote this excellent source from reddit under a spoiler tag for an explanation on the voltages

Spoiler: voltages

SoC Voltage: 950-1150mv (Suggestion: 1060mv)
Should not exceed 1200mv for long term usage. However, the SoC can experience instability from either high voltage (usually starting around 1150mv+, due to additional VRM signal noise) just as it can experience instability from too little voltage. Normal range for SoC voltage is usually 950-1100mv. Some chips want more SoC voltage for higher FCLK, some want less SoC voltage for higher FCLK. You'll need to determine yourself what your CPU likes. Normal ranges for SoC voltage at 1800Mhz is 950-1100mv, with most chips are happy in the 1050-1090mv range, it's rare that any chip benefits from 1150mv+.

CLDO VDDG I/O: 900-1100mv (Suggestion: 980mv)
As the name suggests, this is a low dropout voltage, which is derived from the SoC voltage via linear voltage regulator. This is the most useful voltage for stabilizing higher FCLK. Can approach but not exceed SoC voltage. Most chips prefer this to be somewhere in the range of 50-100mv less than SoC voltage, although some continue to scale FCLK right up until 10mv below SoC voltage. Like SoC voltage, either too high or too low can cause instability.

CLDO VDDG CCD: 900-1100mv (Suggestion: Auto/940mv)
Signal voltage for core communication across the IF. You rarely need to touch this unless trying to max FCLK. Most, but not all chips can run CCD at 900mv at 1800MHz for slightly reduced temperatures. Can approach but not exceed SoC voltage. Trying various combinations and spreads between VDDG I/O, CCD, and SoC can be useful when pushing FCLK/MCLK to 1900MHz+. Most chips like a 20-50mv spread between VDDG CCD and I/O when maxing FCLK.

CLDO VDDP: 900-1100mv (Suggestion: Auto/900mv)
DDR4 PHY signal voltage. 900mv is generally good for up to 1866MHz. Above 1866MHz, 996mv works for many chips up to 2000MHz FCLK.

SoC instability: Manifests in several ways. Severe instability may result in a failure to boot or crashing while trying to load windows. Minor instability may show up in Windows Event Viewer as WHEA errors. WHEA errors can be Fatal (usually results in BSOD's, reboots, or application crashing), Corrected (you won't notice anything, but performance may be lower due to PCI-E error correction kicking in), or Incorrectable (often accompanied by things like onboard audio distortion, USB dropouts, SATA/NVME dropouts, network connectivity issues).