Prime 95 operates without issues but causes a system crash when exiting
Prime 95 operates without issues but causes a system crash when exiting
Hello,
I've been trying to configure my setup to run an OC'd Ryzen 5 1600 with a stock cooler and decided to conduct a stress test using Prime95 for the first time. It was a Blend test that I repeated several times and consistently produced the same outcome. After the test ends, the program halts, reports no errors, but then my system freezes abruptly.
Here’s what I currently have installed:
- Ryzen 5 1600 CPU at 3.4 GHz during stress testing, now back to 3.2 GHz
- G.Skill Ripjaws 2x4GB RAM at 2667 MHz under stress, now at 2133 MHz
- MSI B350 PC MATE motherboard
- EVGA GTX 1050 ACX 2.0 graphics card
- Seagate BarraCuda 1 TB hard drive
- Patriot Burst SSD 120GB
- Asus AC56 Wi-Fi adapter
- LG Blue Ray player
- 650 W power supply unit
- Windows 10 Professional installed on both drives with identical product key
According to the information I found about Ryzen processors, my temperatures stayed below 76°C even after six hours of testing. My system remained operational during the test but would crash almost instantly afterward.
I’m curious about possible causes for these random freezes and how I can begin troubleshooting them. This issue seems consistent across my build, including the unexpected reboots. It’s definitely a challenge.
I hope your advice helps prevent more problems.
Did the application execute "sensors only"? If yes, you need sensor readings in the sensors only section for core temperatures, package temperatures, GPU sensors, hard drive sensors, virtual memory, physical memory, and similar details. It’s important to select the sensor checkbox only when launching and pressing Run. In the sensor view, scrolling down should display Core temperatures, TJmax distance, and package temperatures, all grouped together. Green indicates core temps, blue shows TJmax distance, and yellow represents the package temperature.
Initially, Blend isn't the ideal choice for assessing stability or thermal compliance. Moreover, it's possible that you're using a Prime version that isn't commonly advised for this task because newer versions utilize AVX instructions. Unless you're certain you'll mainly be working with applications that rely heavily on AVX, Prime95 version 26.6 remains the preferred approach for most scenarios.
In terms of hardware, the PC Mate boards are budget models not designed for overclocking, even if capable. They typically feature weak power phases, basic VRMs, and various issues, particularly when attempting overclocks. A quick search reveals numerous discussions about problems with these boards across different generations, with Ryzen models adding to the concerns. For memory or CPU overclocking on Ryzen, a more robust board would be better suited.
That said, the first step should be verifying that your BIOS is at version 7A34vA7, released on or after September 19, 2017.
Secondly, regardless of opinions, using a standard cooler—even a Wraith model—during overclocking isn't advisable, especially on such a basic board lacking adequate power phases and VRMs. Additionally, if you're dual booting with those drives, running the same Windows version and partitions on both, that could also be an issue.
I'm unsure why anyone would dual boot the same operating system, but it seems like a personal choice. Still, it might be unnecessary.
Also, avoid using 'core Temp' on Ryzen or any AMD CPU. The preferred metric is the distance to TJmax. What tool are you using to track thermal data?
I suggest skipping HWmonitor or Open hardware monitor. CoreTemp is a better option—make sure to enable the 'distance to TJmax' feature in its settings. HWinfo is also useful, offering more precise readings than HWmonitor and displaying both core temperature and TJmax distance. You should never allow the TJmax reading to fall below 15-20°C during stress tests or other operations.
Prime95 version 26.6 is widely accepted for most stability and thermal testing using the Small FFT method.
You can find detailed information about Intel CPU architectures and specs at this link, which comes from discussions with Computronix—the author of the Intel temperature guide. This guide is also available on the Intel website.
For AMD systems like Zen/Ryzen, this resource provides similar details, though it's less comprehensive.
A Ryzen overclocking guide is frequently referenced and its principles apply across different generations and platforms.
The Ultimate Overclocking Guide offers a comprehensive overview and is equally applicable for both Intel and AMD systems.
Initially, Blend isn't the ideal choice for assessing stability or thermal performance. Moreover, it's possible that your system is using a version of Prime that isn't usually advised for this task because newer releases include AVX instructions. Unless you're certain you'll mainly run applications that rely heavily on AVX, Prime95 version 26.6 remains the best option for most scenarios.
Firstly, the PC Mate boards are budget models not designed for serious overclocking, even if capable. They typically feature weak power phases, basic VRMs, and various issues, particularly when attempting overclocks. A quick search online reveals numerous threads discussing problems with these boards across different generations, with Ryzen models adding their own set of challenges. If you're planning to push memory or CPU speeds on a Ryzen, consider upgrading to a more capable system.
These boards are basic in nature, lacking advanced features and built with lower quality components.
That said, the top priority should be ensuring your BIOS is at version 7A34vA7 or later.
Secondly, regardless of opinions, using a stock cooler—even a Wraith model—is not advisable, especially on a low-end board lacking strong power phases and VRMs. The VRMs on these boards are often inadequate.
Also, if you're dual booting with those drives, using the same OS version and partitions on both, that could be another issue. It might be worth revisiting your setup.
I don't understand why someone would dual boot the same operating system unless it's a personal preference. It seems unnecessary.
Additionally, avoid using "Temp" core temperature on Ryzen or AMD CPUs. The distance to TJmax is the better approach. What tool are you using for thermal monitoring?
CoreTemp is recommended over HWmonitor or Open hardware monitor. CoreTemp (with the option to display distance to TJmax) is more accurate and gives a clearer picture than just core temperatures. HWinfo is also useful, offering both temperature and TJmax readings, but it's important that this value stays between 15-20°C during stress tests or other operations.
Prime95 version 26.6 is the standard method for most stability and thermal testing tasks.
For more details on Intel CPU architectures, you can refer to the Intel temperature guide, which was provided by Computronix, the author of this guide.
The Intel guide is comprehensive, while AMD systems (Zen/Ryzen) offer similar but less detailed data.
A Ryzen overclocking guide is widely referenced and its principles apply across generations.
Prime95 v26.6 is the go-to for baseline stability and thermal testing.
You can find more about Intel CPU specs at the link provided, which includes a thorough overview of their architecture.
For AMD systems, the guide should give you a similar level of insight, though with less depth.
The Ultimate Overclocking Guide is a comprehensive resource applicable to both Intel and AMD platforms.
Running Memtest86 or Memtest86+ makes sense even on a brand new system. It's wise to do so when no issues are detected, just to verify that any potential errors won't lead to silent data loss. This includes unstable CPU or memory overclocking.
Technically, exceeding 2133mhz (2400mhz on certain modules with default SPD settings) is considered an overclock. However, such occurrences are rare. Most memory-related problems usually generate warnings or noticeable changes, like the issues observed. It's more likely a module fault or configuration mistake, such as insufficient memory voltage, especially during intensive tasks like Blend.
Blend also tends to rely on virtual memory, making it unsuitable for thermal or stability checks. Therefore, Small FFT is suggested instead.
Memtest offers superior memory testing. You're likely using the BIOS version I recommend. Confidence in the newer release (released Jan 29) grows after considering recent microcode updates for Spectre. If you trust the BIOS and no early warnings appear, upgrading could be worthwhile.
For testing, aim for at least five passes, preferably seven. Errors may emerge later; immediate crashes during exit often signal problems sooner.
Repeat the test on Small FFT using version 26.6 to compare results with other versions.
The dual-boot scenario can cause complications due to multiple boot partitions, potentially confusing the system. It's preferable to have a single partition unless you need distinct boot files for different OSes. If one partition is just for backup, disconnect it when not in use and vice versa.
I've personally encountered HWmonitor causing crashes; it's unreliable for some chipsets. HWinfo is more accurate. I'd suggest uninstalling HWmonitor, installing HWinfo, and using the "sensors only" mode. It's a far superior tool. Consult others with more experience for further guidance.
Regarding temperatures, package temp gives a general idea of average core temperature, but individual cores must be monitored closely. Any core exceeding thermal limits is concerning. Uneven heat distribution or poor paste application can lead to overheating, even if overall temps seem normal. Checking each core helps identify such issues early.
Darkbreeze :
Running Memtest86 or Memtest86+ makes sense, even on a brand new system. It's a safe practice to verify stability when no obvious issues appear, since any errors—no matter how minor—can lead to silent data loss, including unstable CPU or memory overclocking. Anything above 2133mhz (2400mhz on certain modules with default settings) is typically considered an overclock. However, such occurrences are rare because most memory problems generate warnings or noticeable effects. It's more likely a module issue or a configuration mistake, like insufficient memory voltage, especially after running Blend, which is very memory-heavy. Also, Blend often relies on virtual memory, making it unsuitable for thermal or stability checks. Therefore, Small FFT is the better choice.
Memtest offers superior memory testing. You're using the BIOS version I'd suggest. It would be ideal to trust the newer release from January 29th, but given recent microcode fixes for Spectre, I'm hesitant until more time passes for independent verification. The latest version claims improved compatibility and performance, so it might be worth trying if you confirm the BIOS is stable and no earlier issues persist. Most of these concerns pertain to Intel systems, but I'd still err on caution and stick with what you have. I'll seek further confirmation before proceeding. If everything checks out, updating to the newer version would be wise. I'm not certain yet due to limited testing.
Regarding the Memtest process, aim for at least five passes, preferably seven. Errors may appear later, but immediate crashes during exit suggest a problem early on. I recommend repeating the test with version 26.6 on Small FFT and comparing results to the previous version.
The dual-boot scenario with installations on separate drives can cause boot partition confusion, though it's less likely if both OSes and partitions are identical. It's preferable to have a single boot partition unless you need different boot records for distinct operating systems. If one partition is just for backup, disconnect it when not in use and vice versa. This setup carries some risk, but it's better than potential instability.
I've personally encountered HWmonitor causing crashes on some chipsets; its accuracy varies. I'd suggest uninstalling it and using HWinfo instead, enabling the "sensors only" option for more reliable data. Other users have shared similar experiences, so this change could significantly improve system stability.
From a thermal perspective, package temp gives a general idea of average core temperatures, but individual cores must be monitored closely. If one core is much hotter than others, it may indicate uneven heat distribution or poor thermal paste application. This could point to air bubbles or thickness issues, requiring reapplication for optimal cooling.
Did the application execute "sensors only"? If yes, you need sensor readings in the sensors only section for core temperatures, package temperatures, GPU sensors, hard drive sensors, virtual memory, physical memory, and similar details. It’s important to select the sensor checkbox only when launching and pressing Run. In the sensor window, scrolling down should display Core temperatures, TJmax distance, and package temperatures, all grouped together. Green indicates core temps, blue shows TJmax distance, and yellow represents the package temperature.
I now grasp what you're referring to by TJM distance. Your Hwinfo shows more statistics than mine. On my end, I have the one with the current temperatures listed. It could be related to my board or possibly because I haven't set up Hwinfo correctly. I'll keep searching for a fix.
On the other hand, I reworked my thermal compound and unplugged that extra drive. I also downloaded and ran Prime 95 on small FFTs for about an hour; it didn't crash, so I believe we've resolved that issue at least.
Cool idea, try CoreTemp instead of HWinfo.
After setup, launch CoreTemp, navigate to Options, Settings, and ensure the box next to "display distance to TJmax in temperature fields" is checked. You can switch it on and off to test if your temps match AMD recommendations and then compare with regular core readings for any unusual differences.
Also, consider removing any unused or outdated monitoring tools, including bundled utilities from your motherboard, unless you really need them. I've noticed problems when multiple such programs conflict, especially bundled ones like AI Suite or MSI control panel.