PDA

View Full Version : Memtest86+ v.4.2 skipping last GB of ram?



ped
02-09-2012, 04:29 AM
Hello everyone,

One follow-up question I had related to memtest86+ (v.4.2). So far I've tested three 4GB sticks from my new build, memtest86+ correctly sees I have 4079MB of RAM in each case, and so far they have all passed. Now I'm on my last stick + last slot (decided to test sticks + slots at the same time). I noticed while it was running through each of the tests across the addresses, that after it's done testing the 2048-3064 range, that it jumps to 4096 - 5112?? Meanwhile it still shows it 4079MB as the total memory.

Is that just a simply little coding bug, or is that an indication of another problem?

http://img.tapatalk.com/57e706a8-eb06-fd9e.jpg

Thanks much.
ped

Lsdmeasap
02-09-2012, 02:49 PM
It could be a coding bug, or an incompatibility with some of your hardware, not sure? I know I suggested this program, so sorry it's already making you ask questions! :bawling:

That would be something best answered by the team over at their forum
Memtest86+ Official forum (http://forum.canardpc.com/forums/73-Memtest86-Official-forum)

I've never noticed it before, so it may be normal for some reason? I'll test myself tonight and pay attention to the test to see if I observe the same as you mention. Does it only happen with test 3# as shown in your image, or any test?

It could be because it's a 32 bit program, so it may not be able to access more than 4GB minus reserved amount? I know a x64 version is coming very soon.

*Edit*

Found this directly related comment by one of the programmers, it's not exact but close, maybe they've just changed the area reserved for this. I'll let you know what I observe later tonight.

The 4736MB upper limit is not incorrect, memory from ~4300-4096MB gets remapped to above 4096MB because that region is mapped to PCI/AGP/PCIe peripherals.
http://forum.canardpc.com/threads/52536-4G-memory-e820-bad-memtest-bad-or

ped
02-09-2012, 06:15 PM
It could be a coding bug, or an incompatibility with some of your hardware, not sure? I know I suggested this program, so sorry it's already making you ask questions! :bawling:

That would be something best answered by the team over at their forum
Memtest86+ Official forum (http://forum.canardpc.com/forums/73-Memtest86-Official-forum)

I've never noticed it before, so it may be normal for some reason? I'll test myself tonight and pay attention to the test to see if I observe the same as you mention. Does it only happen with test 3# as shown in your image, or any test?

It could be because it's a 32 bit program, so it may not be able to access more than 4GB minus reserved amount? I know a x64 version is coming very soon.
*Edit*

Found this directly related comment by one of the programmers, it's not exact but close, maybe they've just changed the area reserved for this. I'll let you know what I observe later tonight.

4G memory - e820 bad, memtest bad or ??? (http://forum.canardpc.com/threads/52536-4G-memory-e820-bad-memtest-bad-or)

No worries at all about the recommandation, in fact switched over to it from another source that it was better. So you reinforced the solution. :)

And thanks for that developer's quote. I think that may explain it quite well.
To answer your questions:
- It occured on every test
- i even started it with a full bank of RAM - 16GB total. Whats happening is in this configuration it still skips the 3GB-4GB range, the next range is now 4-6 GB (instead of 4-5), and then after it scans 14-16GB, it does a scan of 16-17GB. Maybe the quote is the answer?

Very interested to see if you see the same thing in the 3-4GB range.

Do you recommend i still post my question on their developers forum?

Cheers.

Lsdmeasap
02-09-2012, 07:30 PM
Ahh ok, well that's good to know! I just felt bad seeing you have issues after I just suggested for you to use it!

I'll test later today about the memory skipping for you, just so we can be sure because I've never really noticed it myself I don't think, or if I did I didn't think much about it.

I'll test though and let you know! I'll try with 2x2GB, 2x3GB, and 2x8GB just to see if it always happens. Yes, I'd still go ahead and ask in their forum, just to see if they also mention what I quoted from their forum.

ped
02-10-2012, 03:24 PM
Ahh ok, well that's good to know! I just felt bad seeing you have issues after I just suggested for you to use it!

I'll test later today about the memory skipping for you, just so we can be sure because I've never really noticed it myself I don't think, or if I did I didn't think much about it.

I'll test though and let you know! I'll try with 2x2GB, 2x3GB, and 2x8GB just to see if it always happens. Yes, I'd still go ahead and ask in their forum, just to see if they also mention what I quoted from their forum.

Ok i guess I'm more curious what you found in your test before I'll post. :)

If not, will post later tonight if nothing else.

Oh, BTW i ran a 22 hour test last night, 10 passes for the 16GB. So curious if anyone recommends a followup software to these tests. Maybe the windows version, also called memtest? Or should i be good barring the 3-4gb range discussion.

profJim
02-10-2012, 04:49 PM
You can download and view a short 39 second Memtest86+ v4.20 video that shows the typical memory range "skipping" with 4GB of more of memory on my P35-DS3L system.

http://www.mediafire.com/download.php?2cktvscgtz0udkb

You should stress test your system using Prime95 and/or HCI Memtest.
With Prime95 I suggest running each of the tests (Blend, Small FFT, and Large FFT) for at least 6 - 12 hours or more. Longer tests are better.
Download links are in the http://forums.tweaktown.com/gigabyte/30530-latest-overclocking-programs-system-info-benchmarking-stability-tools.html thread.

Iskaryotes
02-11-2012, 10:13 AM
I have the same exact problem!!!
My memtest skips 2gb-4gb region of the memory.
When it does skips that portion, it flashes "relocated" message right next to it. Does your do that too?
Have you found the solution to your problem by any chance?

profJim
02-11-2012, 10:35 AM
If you look closely at the video, you will see that memory appears to be skipped from about 3500 - 4096, but the last portion of each test then appears to be testing memory addresses about 500MB higher than the memory's highest address space. This is not a program bug.

@Iskaryotes
If you are using the older Memtest, version 3.x program instead of using the Memtest86+ program, your problem is due to a bug in the older, original version that hasn't been upgraded for quite a while.
I've never seen flashing "relocated" message using Memtest86+.
We don't know anything about your computer hardware, so we can't advise you further.

Iskaryotes
02-11-2012, 11:01 AM
If you look closely at the video, you will see that memory appears to be skipped from about 3500 - 4096, but the last portion of each test then appears to be testing memory addresses about 500MB higher than the memory's highest address space. This is not a program bug.

@Iskaryotes
If you are using the older Memtest, version 3.x program instead of using the Memtest86+ program, your problem is due to a bug in the older, original version that hasn't been upgraded for quite a while.
I've never seen flashing "relocated" message using Memtest86+.
We don't know anything about your computer hardware, so we can't advise you further.

I am actually using Memtest86+ program.
Also, even the video shows "relocated" message. If you look REALLY closely between 15-16 sec of the video (top right, underneath "one's & zeros), you will see "relocated" message flash REALLY quickly.
and that is what I am talking about.

profJim
02-11-2012, 02:23 PM
My own video proves me wrong :cackle:
The message flashes very quickly just as the test memory address range changes from "184 - 2048" to "2048 - 3583".
If you're gonna be that picky, you should be an assembly language programmer. :laughing: :clap:

With 8gb of ram, the memory address test ranges are:
_184 - 2048
2048 - 3583
4096 - 6144
6144 - 8192
8192 - 8704

note:
4096 minus 3583 = 513
8704 minus 8192 = 512

I'll leave it to an Intel memory specialist to explain why it works this way.
I'm confident that Memtest86+ is testing all available memory.

Iskaryotes
02-11-2012, 02:32 PM
My own video proves me wrong :cackle:
The message flashes very quickly just as the test memory address range changes from "184 - 2048" to "2048 - 3583".
If you're gonna be that picky, you should be an assembly language programmer. :laughing: :clap:

With 8gb of ram, the memory address test ranges are:
_184 - 2048
2048 - 3583
4096 - 6144
6144 - 8192
8192 - 8704

note:
4096 minus 3583 = 513
8704 minus 8192 = 512

I'll leave it to an Intel memory specialist to explain why it works this way.
I'm confident that Memtest86+ is testing all available memory.

Hehe. But the problem with mine is that it "relocated" a whole lot of memory.
It goes from 0-2gb to 4-6gb. It skips over 2-4gb really quickly (that's when it flashes "relocated").
I am just worried that that is where all the bad sectors are and the computer automatically skips that portion because it already knows it's bad.
Then I would effectively be using 14gb of RAM instead of 16gb.
Even though 14gb is PLENTY, I paid for 16gb and I am determined to get my money's worth!!! lol
Well, I filed for RMA with newegg and I will run the test with fresh set of RAM and see what happens.

profJim
02-11-2012, 03:11 PM
Skipping memory as you've described is a problem.
Was your system running at stock settings when you were testing.
Usually, if there is a dead memory module or a bad memory slot, the system won't boot or will freeze during the post process.
It would be interesting to see what happens when running the windows memory diagnostic program from a bootable floppy or usb drive.
The are also some live Linux packages with memory diagnostic software that you can burn to an optical drive to test your hardware.

If you have a problem with your new memory, create your own thread and be sure to create a detailed hardware list in your forum signature.

ped
02-11-2012, 04:31 PM
@Iskaryotes
If you are using the older Memtest, version 3.x program instead of using the Memtest86+ program, your problem is due to a bug in the older, original version that hasn't been upgraded for quite a while.
I've never seen flashing "relocated" message using Memtest86+.
We don't know anything about your computer hardware, so we can't advise you further.

Sorry to correct this, and not trying to single you out because I've seen several admins say the exact same thing, however we need to stop telling everyone that "memtest86" is the older version that has not been updated for awhile. It was indeed the case for awhile, but that has changed.

That's exactly what got me. On top of the fact that if you search "best memory testers" you get some guy on "pcsupport.about.com" that has no time/date stamp (that drives me nuts about this stie), and the #1 recommendation is memtest86. Memtest86+ is 4th on his list because "it's identical to memtest86, and it would be like running the same program twice". So to those looking for the latest info (like myself) or people brand new to this, they wouldn't know. And I know about how memtest86+ was a fork from the original memtest86, and now seems both forks are being worked. :/

Couple that with the fact that:
MEMTEST86:
Version 4.0a (Released 20-Aug-2011)

Multi-threaded with up to 16 Cores
Tests Up to 64 GB of memory
MEMTEST86+:
** Memtest86+ V4.20 (25/01/2011) **

Download - Pre-Compiled Bootable ISO (.gz)
Download - Pre-Compiled Bootable ISO (.zip)

You think you are making the better choice since MEMTEST86 is more recent.

Of course what you don't realize is that Memtest86 despite being more recent, does not seem as stable. So in any case, I completely agree that I won't be using the original for awhile, just if I go by release dates alone, won't make the correct choice.

Cheers.

ped
02-11-2012, 04:45 PM
You can download and view a short 39 second Memtest86+ v4.20 video that shows the typical memory range "skipping" with 4GB of more of memory on my P35-DS3L system.

http://www.mediafire.com/download.php?2cktvscgtz0udkb

You should stress test your system using Prime95 and/or HCI Memtest.
With Prime95 I suggest running each of the tests (Blend, Small FFT, and Large FFT) for at least 6 - 12 hours or more. Longer tests are better.
Download links are in the http://forums.tweaktown.com/gigabyte/30530-latest-overclocking-programs-system-info-benchmarking-stability-tools.html thread.

Ok thanks mucho. Will check that forum and go next.

So Prime95 is used also to just test memory performance, or is that more of the stability aspect? Would for example Prime95 potentially uncover memory problems?

And thanks much to confirm that HCI memtest is a good follow-up. If there any tips, most appreciated if they could be past along. Meanwhile I'll look at the forum post.

Thanks to everyone who helped clarify the skipping problem! Seems to be no issue, and simply part of how the addressing works!

Cheers everyone.

Lsdmeasap
02-11-2012, 04:47 PM
Sorry, I never had time to test yet, but sounds like I may not need to now right?

@ Iskaryotes, is all of your memory showing correctly in system properties, and in Task Manager / Performance Tab (Physical Memory Total)?

You are correct Ped! It's finally been updated, I will however still be using and advising 86+ most likely, but I will check out the updated 86

profJim
02-11-2012, 04:51 PM
Thanks for the update, I wasn't aware that the original Memtest has finally been updated.

FWIW, HCI Memtest is a windows program, where you run several simultaneous instances depending upon how much memory you have installed.

My preference is to run Memtest86+ initially with brand new memory or when I'm overclocking on the "ragged edge", but it isn't a stress test in the normal sense.
Memtest86+ will often run error free for many hours, but HCI Memtest or Prime95 can and will often generate errors if your bios settings aren't okie-dokie.

ped
02-11-2012, 04:57 PM
You are correct Ped! It's finally been updated, I will however still be using and advising 86+ most likely, but I will check out the updated 86

Oh geez, let me tell you. I did use the memtest86 originally and I completely freaked out that my motherboard had gone bonkers because 1/2 way through the test it spit up garbage.
http://img.tapatalk.com/57e706a8-c989-2fff.jpg
That's when I got recommended to try memtest86+ and was much more reassured.

However I will indeed download the prime95/HCI memtest and run those, I really do think that the original is bugged for sandy bridge architectures.

OR... maybe my machine really does have some problems... but assuredly they need to work out some bugs.

Cheers.

profJim
02-11-2012, 05:24 PM
You can click on the Prime95 tutorial link in my sig, just make sure that you have "round off checking" enabled before you start the test.
Prime95 puts your cpu and memory subsystem through a high stress workout, but with a "mostly" stable system, each of the three tests should be run for at least 6 - 10 hours or more. If a single error occurs, stop the test, as your system isn't stable. When reporting Prime95 errors, tell us which test was run, how long the test ran, and which FFT value failed.

I once ran HCI Memtest for several hours and the results were identical to my three 10+ hour Prime95 tests; no errors.

Running only 5 - 10 loops with LinX or IBT (IntelBurnTest) is often a good quick and dirty preliminary stress test.
We will report you to the SPCC (society for the prevention of cruelty to cpu's) if you run more than 10 loops :)

Lsdmeasap
02-11-2012, 05:55 PM
HCI is a great tool! I always use HCI memtest x 4-6 windows (Depending on CPU core count) @ 600-800MB memory or more depending on how much installed, I use that in memory and motherboard reviews, along with 5-10 passes of LinX too

ped
02-12-2012, 03:33 AM
You can click on the Prime95 tutorial link in my sig, just make sure that you have "round off checking" enabled before you start the test.
Prime95 puts your cpu and memory subsystem through a high stress workout, but with a "mostly" stable system, each of the three tests should be run for at least 6 - 10 hours or more. If a single error occurs, stop the test, as your system isn't stable. When reporting Prime95 errors, tell us which test was run, how long the test ran, and which FFT value failed.

I once ran HCI Memtest for several hours and the results were identical to my three 10+ hour Prime95 tests; no errors.

Running only 5 - 10 loops with LinX or IBT (IntelBurnTest) is often a good quick and dirty preliminary stress test.
We will report you to the SPCC (society for the prevention of cruelty to cpu's) if you run more than 10 loops :)

Well I don't know if it's the appropriate thread anymore, however I just ran 130% coverage on HCI Memtest with no errors. Ran Prime95 for 6 hours on blend with no errors either. Then I ran OCCT for the GPU, and voila... ran for about 10 minutes at 230+ fps, then dropped to 40... along with a smorgasbord full of errors.

So there's definitely something up. I will use the "rounding check" point you mentione and now start with the Large FFTs (will go backwards) and see what I come up with.

I guess I'm glad to know that my intuition was correct... however it's always nice to be proven wrong in these cases.

Or maybe I'll see if I can stay within the SPCC guidelines and just go right to LinX :), as now I'm wondering if the GPU issue is related to the CPU. Guess this isn't a test I run just before going to bed, eh? :D

Cheers all!

Lsdmeasap
02-12-2012, 04:12 AM
How many instances of HCI are you using, with how much memory? 1xx% isn't very long, I normally run 4-6 instanced @ 600-800MB until at least 500-1000% depending on the time I have to spare.

100% isn't nearly long enough, you want to go to at least 500% to be sure

profJim
02-12-2012, 03:37 PM
@Ped
We can best help you if you add a detailed hardware list to your forum signature, where we can quickly and easily view your hardware setup. Keep in mind that many of us are assisting in many other threads. Are you running the latest version of OCCT? Running its torture test is similar to running LinX (or IBT) and the gpu test at the same time. My preference is to focus on cpu and memory stability first and to test graphics stability later. Testing all of them at the same time makes it much harder to determine where the problem is.

ped
02-12-2012, 04:40 PM
My current specs are as follows (will post in my sig later today):

CPU: Intel i7-2600K (not yet OC'ed)
Mobo: ASRock P67 EXTREME4 GEN3 SATA 6Gb/s USB 3.0 ATX
RAM: Corsair Vengence 16GB 1600Mhz DDR3 - 4 sticks of 4GB
GPU: Sapphire Radeon HD 7970 3GB (not OC'ed)
SSD: Samsung 830 Series 256GB SATA 6.0Gb/s SSD
PSU: SeaSonic Platinum-1000 1000W 80 PLUS Platinum Modular
Optical Drive: SONY - Interal DVD 24x - ad-5280S - sata
Case: Cooler Master Storm Trooper
Cooler: Thermalright SilverArrow CPU Cooler
Sound: Creative Sound Blaster X-Fi Titanium Fatal1ty PCI-E x1 (only because they didn't sell the cheaper option here)
Monitor: ViewSonic 27" vx2753mh-LED running in 1920x1080p @ 60Hz

Pending placement:
HDD: Seagate Barracuda 7200rpm 1TB SATA 6.0Gbps ST1000DM003

The HDD is pending placement given my CM Storm Trooper case was missing the tool-less rails, which are now being sent via snail mail from CM.

In regards to the quick test vs. gold standard: Yes, absolutely realize that I will need to run everything much longer. I wanted to speed things up this weekend while I am here to see if I can get an overall picture. During the week when I work, will run each test during 12 hour cycles, and before I use the rig for gaming, will do all the windows tests (aka not memtest86+) again once more before calling it good. Do appreciate giving the guidelines however... like 500% for HCI MemTest... getting tons and tons of different info everywhere.

Tests thus far - all passed unless otherwise indicated:
Memtest86 (Aug 2011 build) v4.0a
Ran with both full RAM in 16GB, and finally with just 1 stick in the first slot, same result.
Garbage on the screen - as I posted earlier. Was told that it has known issues, and to use 86+ instead.
Memtest86+ v4.2 (Jan 2011 build)
Stick 1, slot 1 - 12 passes
Stick 2, slot 2 - 16 passes (late work day)
Stick 3, slot 3 - 3 passes (know it's short... will likely do it again unless recommended otherwise)
Stick 4, slot 4 - 4 passes (again, know it it's short)
All sticks in all slots (with original placement as they were tested above) - so 16GB with 10 passes.
HCI MemTest
All sticks in all slots
Ran 7 threads of 2GB each + 1 thread of "All Remaining Ram available"
Ran for 130%, except for the 1 single thread which was like 300% due to less RAM it had to test.
Prime95
Ran Blend - 8 threads - 64 bit - for 4 hours (known too short for gold standard)
OCCT GPU v4.1.0 (did not use beta as production installed Direct X 10 which I didn't have yet)
Ran 1 shader, error checking on, got 300k+ (or 3M+) errors, FPS dropped from 230 to 30 after 10 minutes. Near the end of the test (ran 37 minutes), FPS dropped as low as 15. Image attached.
4319
Wanted to run LinX 0.6.4 afew hours before going to bed.
Ran 3 times (took over 24 minutes per run... longer than I thought - given I used full memory the problem size was 44799).
Ran 4 threads only given did not want testing downtime during the evening.

Stopped and kicked off Prime95 again before bed.
Ran Large FFTs - 8 threads - 64 bit - rounding on - for 9h10m

Now currently re-running LinX again
4 threads, stop on error.
Currently up to 6 runs, no errors.
CPU idle temp 30C, under load (LinX) at 44C - 46C max.

There we go, will run LinX up to 10 runs.

Thanks much everyone for their help and interest!

profJim
02-12-2012, 04:52 PM
I wonder if your video card is running too hot and possibly throttling down, but it shouldn't take 20 minutes to cool down.
Nice system!!!

ped
02-12-2012, 05:07 PM
I wonder if your video card is running too hot and possibly throttling down, but it shouldn't take 20 minutes to cool down.
Nice system!!!

Thanks. Got really good advice on the builds from another site. Never built one from scratch before, but wanted it to last for gaming for 2+ years.

I was thinking the same thing too about the heat... however on Radeon HD 7970 forums, they say that some of the problems with the card is that it sounds like a jet engine when it gets very loaded. I did hear the fan kick on at 79C when I came back to it at the 35 minute mark, but the fans should have kicked in much much more before it ever overheated.. no? So I was expecting it to have a much higher fan noise, or perhaps that was the max and it was simply too much for it? Either way I'm concerned then about this card's performance.

I was running it in full resolution 1920x1080p, however that's what my games would be.

Doesn't explain the errors too, unless the card detected it had errors and throttled itself down because of heat?

What do you think?

profJim
02-12-2012, 05:23 PM
I would try stress testing your video card only, using MSI Kombustor or oZone's Furmark.
As I recall, Furmark displays frame rates and gpu temperatures as part of its on screen display (OSD).
You could also test with your most demanding video game for an extended period of time.
Was your OCCT gpu test done alone or were you using the OCCT torture test?

Most newer mid to higher end video cards will self correct by lowering gpu and/or memory speeds if memory errors are detected, but this shouldn't happen if you haven't overclocked your video card.

ped
02-12-2012, 05:38 PM
I would try stress testing your video card only, using MSI Kombustor or oZone's Furmark.
As I recall, Furmark displays frame rates and gpu temperatures as part of its on screen display (OSD).
You could also test with your most demanding video game for an extended period of time.
Was your OCCT gpu test done alone or were you using the OCCT torture test?

Most newer mid to higher end video cards will self correct by lowering gpu and/or memory speeds if memory errors are detected, but this shouldn't happen if you haven't overclocked your video card.

Yeah, I think that's a good route.

To answer your questions:
Was running the OCCT torture test, but alone? LOL. I guess I can't tell the difference. I know I ran the OCCT GPU test by itself at the full resolution.
And no, have not OC'ed the video card, and in fact given how robust the 7970 would be for a single screen... wasn't planning on it. However Crossfire is in my future once games can recognize it (or drivers for it vice versa), which will help future proofing it.

So next you recommend running the Furmark then?

BTW - the LinX is at 8 runs now... no errors.

Cheers.

EDIT: I think I see what you mean now. I think I was running the GPU test alone as that was the only tab I was on. I didn't see features however to turn on/off the CPU tests, there was just 1 big green on and one big red off button. Thought I'm guessing nothing else was stressed given my CPU during the entire time was hovering around 33C.

profJim
02-12-2012, 05:55 PM
Running oZone Furmark or MSI Kombuster should stress your video card in a "reasonable" manner.
Each program has some optional settings that will stress your card even more.
I think that you can run the OCCT gpu test in a window, which would let you monitor system temperatures and voltages during the test.
If's been a while since I've run OCCT, but as I recall it's the psu or torture test that stresses your complete system.
Make sure that you're running the latest non-beta version of OCCT.

ped
02-12-2012, 06:13 PM
Hey ProfJim, everyone.

Figured since this is becoming a larger tangent from the original post, I moved the discussion here:

http://forums.tweaktown.com/audio-visual/47526-sapphire-radeon-hd-7970-errors-occt-recommendations.html

Under Audio & Visual.

You'll see my reply ProfJim there.

Thanks all!