How reliable is Memtest 86?
How reliable is Memtest 86?
Hello, I recently got some new RAM and am trying to adjust them slightly. For this, I use the Ryzen DRAM Calculator, but the results are quite inconsistent (over 100 errors in the first run). The RAM performs stably (no errors) when XMP is enabled and everything runs in auto mode. I found a helpful guide here: https://tinyurl.com/ydhs8gc3 (the original post includes a document at https://www.reddit.com/r/overclocking/co..._ryzen_oc/).
Following the advice, I started manually adjusting timings while keeping voltages, impedance values (CAD BUS, procODT, RTT) as the calculator suggested. Then I entered the original XMP settings but without auto mode. Everything remained stable.
Later, I changed my tCL from 15 to 14 and saw a massive error count in the first 15 minutes of testing. After following troubleshooting steps from the guide, I reduced the DRAM voltage from 1.35 to 1.38, but the issue persisted. The guide mentions adjusting the address/command drive strength (ADR) or resistance values on the CPU pins—like address, RAS, CAS, WE, bank, parity. Increasing these values could help with tighter timing stability. So I raised the CAD impedance from 20 ohms to 30 ohms.
The next step was running Memtest 86 for over two hours (three passes, no errors). After stopping it briefly, I resumed testing. About two hours later, I ran the test again and saw exactly three errors on the very first pass. Since BIOS hadn’t changed, I decided to stop further testing and looked into possible causes. I found nothing concrete.
This raises my question: How dependable is Memtest 86? If it occasionally reports errors randomly, a quick run with three passes (no errors) followed by one pass with a few errors seems normal. If it’s trustworthy, how long should I run it to confirm stability? With over 40 timings to tweak and already doing stress tests, I worry I might cause more damage than benefit. Plus, it takes a lot of time.
When is the best time to run Memtest 86 and how many passes should I use to be sure everything works? Also, are there other methods or tools you recommend for overclocking memory? Thank you.
It's very dependable since it solely creates data, stores it in RAM, and checks it against expected results. Any discrepancies indicate an issue in that memory area. They usually run several iterations to confirm the problem persists, as external factors can sometimes affect performance. If the error keeps occurring, it may point to a hardware fault or incorrect configuration.
This describes how unstable memory behaves during overclocking. It randomly crashes due to repeated bit flips—which is actually a standard industry issue. Most repair shops need to confirm the RAM can handle repeated stress without errors before labeling it as a problem. Damage occurs only if you apply excessive voltage beyond the recommended level (usually around 1.45V). If you think overclocking is wasting your time, avoid it. Simply enable the XMP profile and, if needed, manually adjust the speed on the motherboard and fine-tune timing slightly.
Errors may occur even in ideal conditions, which is why ECC memory was developed. You want to see steady issues repeated over several attempts. This consistency points to a real problem.
For instance sector 3466 generates 44 errors across three attempts, each with three passes, yet no issues appear in the fourth attempt—this suggests it might be acceptable now, as the fourth was just fortunate. That's why I generally use a scale of six and base decisions there.
You're asking about the behavior of Memtest86 when it detects errors. The message you received indicates a discrepancy between reported errors and actual results—specifically, three passes showing no issues but a second test revealing errors in the first run. It seems the tool is flagging something that isn't present during subsequent checks. Regarding the number of passes, increasing from four to six might help catch intermittent faults, though it could also increase noise. As for identifying problematic sectors, the error details usually point to specific addresses or memory regions where issues occur. The format you mentioned (e.g., 0xxxxxxxxxxxxx) helps pinpoint the location, making it easier to focus troubleshooting on the relevant area. Frequency settings matter too; staying within recommended ranges (like 3200 MHz for your CPU) prevents instability and crashes. Adjusting DRAM, SOC voltage, or timing parameters can improve performance, but be cautious of over-tweaking. If errors appear only in certain passes, treat them as transient issues rather than signs of a fundamental problem.
The exact address 0xxxxxxxxxxx points to the faulty memory spot. If it keeps appearing, the device is defective. Because the outcomes aren't clear, try running it for 10 cycles. If only one cycle shows a few errors, the memory is likely fine. You're essentially confirming your point. If the issue doesn't repeat, it probably isn't the RAM but some external interference, suggesting everything is normal.
From my perspective, prioritizing stability and reliability means simply leaving XMP settings alone and proceeding. RAM overclocking presents a significantly greater challenge compared to CPU overclocking. The chances of undetected data corruption during RAM overclocking are considerably higher. I haven’t encountered memtest producing misleading positive results. The observed differences in performance across runs usually indicate underlying instability. I rarely relied on the calculator, as it often suggested overly optimistic memory values even for safe configurations. It’s also important to remember that memtest isn’t the sole method for assessing RAM stability. In the past, I experienced a situation where memtest continued to run smoothly without any issues. Using Prime95 instead would have been more effective, focusing solely on RAM performance. The stress on RAM comes from Prime95’s workload, but real-world scenarios like Prime95 can apply varying patterns to the memory controller, leading to inconsistent outcomes. You’re not just adjusting RAM speed; the memory controller must also adapt effectively.
Thank you for clarifying. To be honest, I wasn't adjusting the timings; I only followed the initial settings from the calculator. When I noticed it wasn't functioning properly, I switched to a manual approach. For that, I rely on the instructions in the guide along with my own calculations. I use the calculator to obtain values for impedances (CAD Bus, ProcODT, RTT) and various voltages, since beyond vDIMM and vSOC I'm not sure about the best methods or settings. I've tested with a 95V source for about half an hour, and so far there have been no issues.