P95 still effective at RAM OC testing?

[AlwaysHope]

Been out of the loop with p95 & all its meant' to be in this day & age but here goes...
My question is if the latest version of it is actually effective at detecting RAM oc errors on rocket lake platforms?
I use to use TM 0.12 with custom config. when I was on a Zen+ platform but that was then, not sure if that app is still applicable to rocket lake still?
So far I've been able to do Memtest86 & even the Win10 inbuilt memory diagnostic app with success so far (although it was running in basic)
But too properly test RAM oc in windows with rocket lake, what is the best app these days to do this?
Thanks.

 

[Regeneration]

Effective only to detect drastic RAM instability. Multithreaded RAM stress test utilities like MemTest64 are better tools to test RAM stability.

 

[AlwaysHope]

Effective only to detect drastic RAM instability. Multithreaded RAM stress test utilities like MemTest64 are better tools to test RAM stability.

Which one, Memtest86 or MS app? You do know the MS app has 3 states it can run in & even run with cache disabled? I've used MemTest64 before but not certain if its updated enough for Win10 21H1?

 

[lexluthermiester]

My question is if the latest version of it is actually effective at detecting RAM oc errors on rocket lake platforms?

It depends on which test you run.

Effective only to detect drastic RAM instability. Multithreaded RAM stress test utilities like MemTest64 are better tools to test RAM stability.

This is not entirely correct. Memtest is a very good tool for testing RAM, however it is not the end-all-be-all tool. Prime95 has a very useful feature for test CPU, IMC and RAM stability all in one shot. See below.

Ok so if you want to test your RAM stability open up P95 and select the shown "Blend" test;

That's a bit of a basic test though. If you really want to stress out your RAM to ensure it's rock solid see below;

Selecting custom, we set 4 in the "Min FFT size (in K)" field, we set 8192 in the "Max FFT size (in K)" field.

The next input setting depends on your total system RAM. In the "Memory to use (in MB)" field I put in 31000 because I have 32GB of RAM. We have to leave 1GB of RAM for the OS to run in. So if you have 8GB of system RAM, put in 7000 for 7GB to use.

And last we have the "Time to run each FFT size (in minutes)". The default is 3 minutes, but that is not enough time for the custom calculations to complete full cycle, so an extra minute is required.

With this configuration, any instabilities in RAM will be made a readily apparent. So if you run the "Small FFT" and "In-place large FFT's" and everything is fine, but you get a crash on the "Blend" and Custom test, you know your RAM is not perfectly stable.

 

[chrcoluk]

memtest86 is ok at detecting defective ram, but unstable timings/clocks are best detected with in OS testing.

For Windows I suggest either HCI or Karhu.
For Linux, google stressapptest, the best out of the three, if this one passes, you are good to go. The lack of spaces is deliberate to match the name.

 

[Regeneration]

This is not entirely correct. Memtest is a very good tool for testing RAM, however it is not the end-all-be-all tool. Prime95 has a very useful feature for test CPU, IMC and RAM stability all in one shot. See below.

Unstable memory may still pass Prime95 blend without errors.

Which one, Memtest86 or MS app? You do know the MS app has 3 states it can run in & even run with cache disabled? I've used MemTest64 before but not certain if its updated enough for Win10 21H1?

MemTest64, HCI MemTest (several instances), TestMem5.

 

[AlwaysHope]

It depends on which test you run.

This is not entirely correct. Memtest is a very good tool for testing RAM, however it is not the end-all-be-all tool. Prime95 has a very useful feature for test CPU, IMC and RAM stability all in one shot. See below.

Ok so if you want to test your RAM stability open up P95 and select the shown "Blend" test;
View attachment 224474

That's a bit of a basic test though. If you really want to stress out your RAM to ensure it's rock solid see below;
View attachment 224475
Selecting custom, we set 4 in the "Min FFT size (in K)" field, we set 8192 in the "Max FFT size (in K)" field.

The next input setting depends on your total system RAM. In the "Memory to use (in MB)" field I put in 31000 because I have 32GB of RAM. We have to leave 1GB of RAM for the OS to run in. So if you have 8GB of system RAM, put in 7000 for 7GB to use.

And last we have the "Time to run each FFT size (in minutes)". The default is 3 minutes, but that is not enough time for the custom calculations to complete full cycle, so an extra minute is required.

With this configuration, any instabilities in RAM will be made a readily apparent. So if you run the "Small FFT" and "In-place large FFT's" and everything is fine, but you get a crash on the "Blend" and Custom test, you know your RAM is not perfectly stable.

Thanks, I get that about custom runs of p95, I think I'll run large ffts & blend to begin with. I did some testing yesterday @ 3733MHz with my G.Skill kit as described in my system specs & the major timings for XMP locked in, the system automatically turns on gear 2 mode. 2.5Gb of ram should be reserved for win 10, at least that's what I can see from my end & I run literally a vanilla OS ( no customisation cause' MS keep breaking that with each OS update...) but I'd like to know your view on the Microsoft memory integrity app that runs prior to the OS loading? I've used it before & haven't had any issues with tweaked ram in win 10

Unstable memory may still pass Prime95 blend without errors.

This is true, I've seen posts on numerous other PC enthusiasts forums complaining about running p95 variety of vanilla tests for over 12 hrs or more & yet when gaming or something else get BSOD or other types of errors. Although I am aware of the devs who write the code for p95 updating the app frequently for newer platforms.

 

[tabascosauz]

Been out of the loop with p95 & all its meant' to be in this day & age but here goes...
My question is if the latest version of it is actually effective at detecting RAM oc errors on rocket lake platforms?
I use to use TM 0.12 with custom config. when I was on a Zen+ platform but that was then, not sure if that app is still applicable to rocket lake still?
So far I've been able to do Memtest86 & even the Win10 inbuilt memory diagnostic app with success so far (although it was running in basic)
But too properly test RAM oc in windows with rocket lake, what is the best app these days to do this?
Thanks.

Prime95 Large FFT is decent for testing Infinity Fabric and memory controller stability. It still kinda works for memory, but no one serious about memory OC will use it for memory stress testing. It's not that it's necessarily bad (after all Large FFT does still have significant memory load and you will see similar DIMM temps), it's just that it gives you no info whatsoever as to what went wrong. Not to mention with 16/24/32 threads it's hard to keep track of errors if you're standing next to the PC the whole time.

With HCI you can at least see how far you got in coverage. With TM5 you can see what test specifically you failed on, and some error codes to hopefully point you in the right direction if you're not already aware.

Honestly I'm not sure why you would walk *away* from TM5. In OC circles it's pretty much the holy grail. It's also pretty heavy on IF as well (to the point where I would consider P95 Large FFT to be basically redundant), and hammers the CPU if you have Ryzen 5000.

If you want a complement to TM5 for longer term testing, HCI is okay but the free version is a hassle and it's generally not as heavy on CPU as TM5. HCI Pro is convenient for endless coverage into the 2000-3000% region though. Then there's LinpackXtreme, in a lot of ways it just does what TM5 does, but it does give you a GFLOPS number to gauge performance.

Blend FFT just runs a mix of stuff. I'm not sure why it still gets recommended - if you want to torture the cores you use Small, if you want to torture the memory controller you use Large, you don't use something that weakly tries to do both and regularly gives the respective hardware a break.

As to earlier mentioned "in-place Large FFT", that doesn't have much utility because the point of Large FFT is to run in RAM.

 

[AlwaysHope]

Prime95 Large FFT is decent for testing Infinity Fabric and memory controller stability. It still kinda works for memory, but no one serious about memory OC will use it for memory stress testing. It's not that it's necessarily bad (after all Large FFT does still have significant memory load and you will see similar DIMM temps), it's just that it gives you no info whatsoever as to what went wrong. Not to mention with 16/24/32 threads it's hard to keep track of errors if you're standing next to the PC the whole time.

With HCI you can at least see how far you got in coverage. With TM5 you can see what test specifically you failed on, and some error codes to hopefully point you in the right direction if you're not already aware.

Honestly I'm not sure why you would walk *away* from TM5. In OC circles it's pretty much the holy grail. It's also pretty heavy on IF as well (to the point where I would consider P95 Large FFT to be basically redundant), and hammers the CPU if you have Ryzen 5000.

If you want a complement to TM5 for longer term testing, HCI is okay but the free version is a hassle and it's generally not as heavy on CPU as TM5. HCI Pro is convenient for endless coverage into the 2000-3000% region though. Then there's LinpackXtreme, in a lot of ways it just does what TM5 does, but it does give you a GFLOPS number to gauge performance.

Blend FFT just runs a mix of stuff. I'm not sure why it still gets recommended - if you want to torture the cores you use Small, if you want to torture the memory controller you use Large, you don't use something that weakly tries to do both and regularly gives the respective hardware a break.

As to earlier mentioned "in-place Large FFT", that doesn't have much utility because the point of Large FFT is to run in RAM.

Thanks, I'm not waking away from any RAM testing apps. I've used TM5 before but only on AM4 rigs not a rocket lake platform, I'm not sure if its compatible & works effectively at detecting errors on that later platform. I've used LinpackXtreme many times before but again on AM4 rigs only at this point in time. In any case though, I'll fire up TM5 & LinpackXtreme, see how though go on rocket lake.

 

[tabascosauz]

Thanks, I'm not waking away from any RAM testing apps. I've used TM5 before but only on AM4 rigs not a rocket lake platform, I'm not sure if its compatible & works effectively at detecting errors on that later platform. I've used LinpackXtreme many times before but again on AM4 rigs only at this point in time. In any case though, I'll fire up TM5 & LinpackXtreme, see how though go on rocket lake.

afaik no real platform exclusivity stuff going on with memory tests, going back to coffee lake. Just use TM5 with the usual configs, anta extreme1 or 1usmusv3.

If anything, Prime95 is the one that's kinda fallen out of favour depending on what you're using it for (though not its fault). Small FFT basically doesn't work on Ryzen 3000/5000 because new AGESA recognizes it as a power virus and throttles boost clocks hard. All non-AVX P95 still seems to work fine, the corecycler type tests for curve optimizer usually use some form of SSE P95 default.

 

[AlwaysHope]

afaik no real platform exclusivity stuff going on with memory tests, going back to coffee lake. Just use TM5 with the usual configs, anta extreme1 or 1usmusv3.

If anything, Prime95 is the one that's kinda fallen out of favour depending on what you're using it for (though not its fault). Small FFT basically doesn't work on Ryzen 3000/5000 because new AGESA recognizes it as a power virus and throttles boost clocks hard. All non-AVX P95 still seems to work fine, the corecycler type tests for curve optimizer usually use some form of SSE P95 default.

​

p95 30.7 build 8 was running yesterday on my stock i7-11700k with large ffts & RAM oc to 3733Mhz was working well. Not sure if that version is non-avx though.

 

[chrcoluk]

To give an idea, when I moved my 2600x to a 4 dimm setup, I let memtest86 run overnight and no errors.

When I fired up karhu it found an error in under 3 seconds.

Whilst it was idling at the desktop after karhu stopped it had a ram error BSOD.

At 2933mhz it was able to pass 3 hours of karhu, and windows was seemingly stable. But after putting on proxmox (intended use case), I started getting occasional page errors crashing VM's.

Stressapptest then found a very low rate of errors but were there, dropping it to 2866mhz it now passes that test and not a single page fault since, for whatever reason linux is more sensitive to timings than windows.

 

[AlwaysHope]

I'm still new to this LGA1200 platform with Z590, after running Aida64 for 3600 & then 3733MHz ram & forcing the bios to boot in gear 1 mode for 3733, its a no boot at that speed. When booting at 3733 & gear ratio set to auto (gear2) the system boots happily. Running Aida64 benchmark report indicates less RAM scores than 3600 but very slight increase in most CPU tests....This is with the latest bios too for my mobo. Think best to stick with 3600 & tighten major timings for now with RAM testing.
One of the peculiarities of rocket lake.

 

[AlwaysHope]

Ok, finally decided to take the dive & get HCI's memtest pro. v7.0 is very easy to use & works just fine on win10 21H1.
Can't be bothered running p95 for hrs & hrs even though the devs keep it updated.

 

[tabascosauz]

Ok, finally decided to take the dive & get HCI's memtest pro. v7.0 is very easy to use & works just fine on win10 21H1.
Can't be bothered running p95 for hrs & hrs even though the devs keep it updated.

Nothing quite like HCI Pro, good balance of demanding and 1-click easy. Just make sure Low Priority Threads is always unchecked. I usually set same # of threads as CPU, then leave capacity on Auto and it'll decide the right capacity per thread to fill your RAM on its own.

600-700% per thread is okay for a daily system, if you want to be extra sure run it overnight into the 2000-3000% region.

 

[The red spirit]

Please don't use memtest86 for stability testing, my own experience with it was poor ad it wasn't even detecting clearly unstable RAM. Prime95 reasonably decent at RAM testing. My favourite quick and dirty RAM test is MemTest64. It's good at finding stability problems, but it needs Windows loaded.

 

[AlwaysHope]

Please don't use memtest86 for stability testing, my own experience with it was poor ad it wasn't even detecting clearly unstable RAM. Prime95 reasonably decent at RAM testing. My favourite quick and dirty RAM test is MemTest64. It's good at finding stability problems, but it needs Windows loaded.

The best scenario is to use a combination of apps. That is, pre windows & after. Imo, memtest86 v9.3 is fine for pre windows & at least 2 passes minimum, then I do MS memory diagnostics app in standard form (but have to load windows first to fire it up) then on to something with OS running in background like HCI memtest pro or similar. The pro version of that even has a feature to run it in the background whilst doing other activity in OS (gaming, etc..) . A sure fire way to detect in real time memory errors from OC for ALL of the ram.

 

[The red spirit]

The best scenario is to use a combination of apps. That is, pre windows & after. Imo, memtest86 v9.3 is fine for pre windows & at least 2 passes minimum, then I do MS memory diagnostics app in standard form (but have to load windows first to fire it up) then on to something with OS running in background like HCI memtest pro or similar. The pro version of that even has a feature to run it in the background whilst doing other activity in OS (gaming, etc..) . A sure fire way to detect in real time memory errors from OC for ALL of the ram.

I disagree. You only need something quick and dirty between tunes and only on final tune for each setting you should then run more software or just run what you have longer. If you want to tweak RAM timings with all subtimings, then it's going to take a shit ton of time. Still, most of the time best stability test is whether Windows boots or not. That's actually a good test very often. Not very often I saw problems past boot.

Anyway, you seem to mention HCI test's background RAM checking functionality. I would really advise you to not do anything like that and just accept that your PC is going to be not usable. This does impact RAM stability testing results, can potentially crash your system or produce some silent data corruption.

Memtest86 is usually slow as hell. Besides poor stability testing capabilities, it can take hours to test each change. Just don't use it. MemTest64 and P95 are fine, both are faster than memtest86 and test stability better.

 

[AlwaysHope]

I disagree. You only need something quick and dirty between tunes and only on final tune for each setting you should then run more software or just run what you have longer. If you want to tweak RAM timings with all subtimings, then it's going to take a shit ton of time. Still, most of the time best stability test is whether Windows boots or not. That's actually a good test very often. Not very often I saw problems past boot.

Anyway, you seem to mention HCI test's background RAM checking functionality. I would really advise you to not do anything like that and just accept that your PC is going to be not usable. This does impact RAM stability testing results, can potentially crash your system or produce some silent data corruption.

Memtest86 is usually slow as hell. Besides poor stability testing capabilities, it can take hours to test each change. Just don't use it. MemTest64 and P95 are fine, both are faster than memtest86 and test stability better.

Disagree all you like, but that's my method of testing cause' I have the time to do all that. Memtest Pro is safe to use when OS is running with other apps simultaneously > https://hcidesign.com/memtest/manual.html

 

[tabascosauz]

Disagree all you like, but that's my method of testing cause' I have the time to do all that. Memtest Pro is safe to use when OS is running with other apps simultaneously > https://hcidesign.com/memtest/manual.html
View attachment 225078

To be fair, the HCI "low priority mode" seems to be just there for convenience in case you're busy with stuff. There really isn't much benefit to it if you want to do test properly. It should always be the same routine - free up as much RAM as possible, and let the system focus solely on HCI or TM5 for overnight/several hours.

I understand where the real-time testing sentiment comes from (because it's very true when it comes to Infinity Fabric stability), but it's not necessary. If you have DIMM temp sensors, you'll see that no game or real application will place nearly the same amount of sheer stress on your RAM that HCI or TM5 will (unless you have a 3090 and downright horrible airflow) - and honestly, the vast majority of CPU-bound games will have a hard time matching what TM5 does to CPUs, even.

 

[The red spirit]

Disagree all you like, but that's my method of testing cause' I have the time to do all that. Memtest Pro is safe to use when OS is running with other apps simultaneously > https://hcidesign.com/memtest/manual.html
View attachment 225078

I never said unsafe, it simply won't test RAM as good. That's defeats the point o testing RAM to me. You want the best stability test, not something that just seemingly works. If you aren't willing to do that, then perhaps just don't mess with RAM at all. You won't see big gains from tweaks anyway. Or stick with only certain tweaks. That more or less leaves you with speed, CL and command rate.

 

[chrcoluk]

The best scenario is to use a combination of apps. That is, pre windows & after. Imo, memtest86 v9.3 is fine for pre windows & at least 2 passes minimum, then I do MS memory diagnostics app in standard form (but have to load windows first to fire it up) then on to something with OS running in background like HCI memtest pro or similar. The pro version of that even has a feature to run it in the background whilst doing other activity in OS (gaming, etc..) . A sure fire way to detect in real time memory errors from OC for ALL of the ram.

Good choice to use HCI.

I would only do light usage during testing though, just in case its not a stable configuration and you dont lose any work.

Using the PC whilst HCI is doing its work shouldnt crash a stable configuration, but what you dont want is either a OOM or to have it start swapping. As HCI needs to be allocated as much as your free ram as possible.

 

[GerKNG]

i literally just finished a 1000% memtest run with 0 errors. (12600k 32GB DDR4 3600mb/s CL16 @ 4000 CL16)
got a BSOD in P95 448-4096 with around 85% of the ram allocated within 3 minutes

 

[Space Lynx]

I find AIDA64 system stability test works best. It lets me know fairly quick if there will be issues in the future or not. I usually run 12 mins of AIDA64, if it passes and no errors, I leave that OC alone.

 

[phanbuey]

So with ram OCs, especially ones at the edge - it's better to run multiple tests - something like memtest to 1000%-2000%, kahru to 2000%, testmem 5 for 30 mins Prime and some games.

A cocktail of tests for shorter times seems more effective than just one test for forever.

Aida Stability Test missed a bad stick of ram for 15 hours - karhu found it in under a minute, so I don't really use it anymore.