The following is not in any particular order and is based on a mix of my opinion and opinions of others that I often see floating around PC sites.
-- "short" = quick, fast test, usually 1- 5 mins
-- "medium" = minimum test, I'd be fairly comfortable if the RAM passes medium without error
-- "long" = when you really want to be extra certain of no RAM errors. Lots of people only use the medium tests (with a few different programs though) and have success that way. Some people even say the long tests are overkill. If you really want to be extra certain, and don't mind the extra time (do it while you sleep, are busy with other things, are out for the day/night, etc.), then do the long tests. Most errors should be picked up with the "medium" tests though. Also, errors
not picked up in medium but
picked up in "long" can be purely related to heat and nothing else, which can then deceive you into thinking your RAM OC settings are bad, so monitor your RAM temps!
Karhu RAM Test
- medium: 6400% (Karhu says 6400% has 99.41% error detection rate)
- long: 10,000% - 20,000%
HCI Memtest / Memtest Pro
- medium: 400% - 600%
- long: 1000 - 2000%
- TM5 Anta777 Absolut (replaced Anta777 Extreme)
- medium: 6 cycles
- long: 18-24 cycles
- TM5 Usmus v3
- medium: 6 cycles
- long: 12-18 cycles
Linpack Xtreme
- medium: 10-15 runs of each stress test
- long: 25 - 50 runs of each stress test
- Note: because I have no idea which test to do (I think the #s are 2 GB, 3, 4, 6, 10, 30 or something), I just do every one. Eg. 10 runs of 2 GB, 10 runs 3 GB, and so on. The 2 GB tests last like 5 seconds each, the 3 GB like 15 seconds, etc. so they're not too time-consuming. The most time-consuming test (30 GB) is around 5-ish mins each
Y-Cruncher
- not sure which medium or long stress tests to run, ask around because this is another great program. See below for using Y-Cruncher for a great short test
OCCT
- 30 min w/ AVX
- 30 mins w/ out AVX
Notes
- apparently TM5 Usmus v3 and Karhu hit the IMC harder so can potentially detect IMC-related errors quicker
- apparently Y-Cruncher can potentially detect VCCSA voltage-related errors quicker
- for a short test to check VCCSA, run Y-Cruncher with the following settings 0 [hit enter], 1 [hit enter], 8 [hit enter]. This test should last around 3-ish mins. If you error/fail this test, consider adjusting VCCSA (up or down) before adjusting any other settings (other voltages, RAM frequency, RAM timings, CPU core, CPU cache, etc.)
- me, personally, when I try new RAM settings before I go into the medium or long stress tests, I do some short tests, those being: 1 or 2 of the Y-Cruncher VCCSA tests I explained in my previous point, then 3-5 Linpack Xtreme runs using stress test and the second or third lowest GB option. If I pass both short tests, I then use a few programs to benchmark the new RAM settings to see the speeds, scores, latency, etc.: Programs I use for this are as follows (in no particular order):
- Passmark Performance Test (set to 2 tests, short, keep best result, all Memory Tests)
- Aida64, the 4 memory tests (I do the latency test 3-5 times and use the best result since it can be really inconsistent)
- Intel MLC GUI - just the first screen's bandwidth & latency test, I do it 2 times as the latency test is more consistent than Aida64
- Linpack Xtreme: Stress Test, 3 tests of your desire (I choose 3 GB option)
- Y-Cruncher: upon starting, hit 0, enter, 1, enter, 8, enter, use the two times as test score
- OCCT Memory Benchmark
- PyPrime: a nice quick test but good because scales well with everything (like Linpack Xtreme)
- Geekbench 3: faster than GB4, use the memory (single- and multi-threaded) scores
- I not only do this out of curiosity and excitement to see possible gains in performance but to also see if there are any weird scores. If you have weird test results like lower scores, lower bandwidth, higher latency, compared to your previous, slower RAM settings, then that can indicate an issue (ie. instability). I then go on to the medium stress tests. I personally start with the following 3 in no particular order: TM5 Usmus v3, TM5 Anta777 Absolut, Karhu Ram Test. If all 3 of those give me no errors, then I move on to a medium run of HCI Memtest and medium Linpack Xtreme. If all those pass, I then either move on to the long tests or use my PC as normal and do the long tests when I feel like it (when I'm sleeping, not home, don't need the PC, etc.)
- I always have HWInfo open during medium and long stress tests to monitor temps
- I always have my case side fans on max, my door open, and WIndow open (if I'm home) to keep the RAM as cool as possible in order to get heat out of the equation as much as possible
- If you pass all the medium and long test yet start having instability in games, day-to-day stuff, etc. and you notice that the RAM temps are way higher than during the tests, then you can be 99.9% sure the instability is purely heat related.
- I usually have a conservative clock on my CPU and CPU cache (AKA ring, uncore). For eg, if I know I'm stable at 1.300v 5.2 GHz core, 4.9 GHz cache, I'll keep the same VCore but drop cpu clock down 100 MHz and cache down 100-200 MHz in order to get those out of the RAM & IMC equation. If I pass the Y-Cruncher and Linpack Xtreme short along with the medium tests, I then put my CPU and CPU cache back to my normal clocks and then start with the long tests. Some people say to not do this but instead to just always test the RAM with your max/normal CPU clocks from the beginning. They may be right. If you're 100% sure that your CPU and CPU cache clocks are 100% stable then that may indeed be a better strategy. Sometimes a normally stable core or cache clock may need more voltage once you push your RAM harder due to the IMC being under more stress so you could indeed get instability which you may think is from the RAM (since you think you know your CPU core & cache are 100% stable) when it could, in fact, actually be your CPU cache being too high (and possibly just needing a bump in voltage) due to your RAM OC putting more stress on the IMC. That's why I like to back off the CPU & CPU cache clocks 100-200 MHz at first.
P.S. If anything is incorrect, let me know and I'll correct it ASAP
). I earlier downloaded tm5. Is that also a program able to quickly sus out instabilities in the system, if any?
github.com
