GUIDE: Burning In Your Memory

[Jason]

Many people talk about burning in their system, specifically their memory, to ensure that everything is stable and will operate reliably. According to Webopedia, burn in is defined as, "To run a system or device for a period of time to ensure that all components are working properly." So this guide attempts to do just that for your memory. Good cooling is a must, especially when you are going to be stressing your components for a long time. I wouldn't add any "extra" cooling just for the purposes of burning in your system, you want it to be like you would run on a daily basis.

Some people say that Samsung memory in particular needs to be "burned in" to reach its maximum potential, others say that is a load of you know what, but whatever you view is on RAM and "burning in" hopefully you will find this information useful to check your stability at the very least.

It will take some tinkering to figure out your maximum overall performance vs. raw MHz. Sometimes you can run more aggressive memory timings, but at a slower speed, and they will yield better results than if you used less aggressive timings and a higer speed. So be sure to have a pencil and some paper handy to write down benchmark numbers. I like to see how high I can get with the most aggressive timings, then start to relax them and see if I can get any more speed out of the memory.

A few things you will need before starting:

Required - Mad Onion's 3DMark 2001 SE
Required - SiSoft Sandra 2002 SP1
Optional - Quake 3 Arena & Q3Bench


Burn Baby Burn:

1. First set your memory to stock speed, timings, and voltage and boot into windows.
2. Disable your screensaver.
3. Start SiSoft Sandra and run the Memory Bandwidth Benchmark once and record the score, this can be used as a baseline to compare against for later testing. Close Sandra.
4. Start 3DMark 2001. Run the benchmark once and record the number, again, this can be used as a baseline to compare against for later testing.
5. Click the "Change" button under the "Options" section. Check the "looping" box and set the "Test Repeat Count" to run 5-10 times (your choice really how much you want it to run). You can optionally increase your display settings under the "Display and CPU Settings" to help put more load on your system.
6. Reboot your system, go into the BIOS and increase your memory speed & timings some, nothing too crazy, the key is to do it in small steps. I like to start with the CAS Latency and work my way down to the Command Rate. Some memory absolutely refuses to run with certain timings, it is good to write down what works and what doesn't during your testing. Save the bios and boot into windows.
7. Follow Step 3, 4, and 5 over and over. If 3DMark locks or freezes on you, increase your DDR voltage some and try benchmarking again.
8. Once you feel you have reached the max stable speed and timings, load up SiSoft Sandra and run the Memory Bandwidth Benchmark once and record the score.
9. Close the Bandwidth Benchmark window and scroll down the main list and click on the Burn-In Wizard. Click Next, Then unselect all tests EXCEPT the Memory Bandwidth Benchmark, and optionally the Cache & Memory Benchmark. Click Next. Run the test at least 25 times, and under HIGH priority.
10. If the Sandra tests pass, then start back at Step 5 and work your way down the list again upping the speed more. If it fails, then go back to Step 5, but decrease your speed some.
11. Once you are sure you have reached the maximum speed and timings, be sure you have all the relative memory numbers written down from the BIOS (MHz, Timings, Voltage), and benchmark scores for both 3DMark & Sandra Memory Bandwidth Benchmark.
12. You can try to relax some of your timings and increase your memory speed, I suggest only changing one timing at a time. Once you make a change, then start back at step 3 and work your way down the list. Sometimes relaxing a few timings can yield a lot more MHz and give you better overall performance; sometimes the extra MHz doesn't make up the difference. Compare all your scores at different speeds / timings, it will become obvious if you start going in the wrong direction, and what will give you the best performance.
13. When you feel you have reached the speed & timings that give you maximum performance, load up SiSoft Sandra again. Repeat Step 8, EXCEPT set the burn in to loop instead of a fixed number of runs. Let the test run overnight. If it is still running in the morning, then go back to step 3 to test with 3DMarks again, if it passes you can try and increase your speed / timings once again, if it locked / rebooted then you will need to decrease your speed some and go back to Step 3 or 8.

--Helpful Images--

You don't necessarily have to follow the guide step for step, some people like to run more loops, some like to run less. Running 3DMark or Sandra in a continuous loop all night really puts the stress on the system and lets you know the next morning if your RAM is stable at that speed or not. Be sure to write down EVERYTHING so you can refer back to what worked and what didn't, it will save you a lot of rework & headaches in the long run. I included Q3A optionally because it is also an excellent stress tester. Q3Bench allows you to run at all sorts of resolutions, and it will run each resolution multiple times, so you can just let it run all the tests and come back after watching some TV to compile your results. Please read more on the Q3Bench for how to use that program. You can save and use those results too to compare for best overall performance.

Feel Free to PM me with any questions / comments / additions / corrections.

Very nice work there Jason This smokes you know who's Guide

[Jason]

Update:

Added some images to the bottom to help clarify...

[ZeroDread]

Great, That helps....

[phobix]

Very Helpfull 2 thumbs up!

[xMagusx]

very well done man!!!

[DoGMaN]

I'd say that this may be the most important thread I have seen in a long time. Personally I swear by burning in my components. Usually for at least 48 hrs.

An other good program to run is memtest86 from www.memtest86.com. Make the disk and boot from it. When it starts press C then 2,5,5 hit enter then 0. This will loop test 5 which is the most stressful on your memory. When you think you are at the limit of your ram, run test 5 looped for 3-4hours. If you can successfully make it through with no errors, you can try another Mhz or two. Be warned, this will heat your memory up a lot. If you don't have heat spreaders, be sure to watch your system very carefully.

[biohazard438]

very nice man. im gonna have to use that when i get my new ram. this should definatley be a sticky. ooops just saw that it already is.

[phobix]

Further to the memory guide...I would suggest also running prime95 / Torture as a means of stability testing. IMHO, This is the tried and tested method of finding instabilities in memory. I have seen instances where 3dmarks and the memory bandwidth benchmarks would pass under infinite loop conditions and only after running prime would the system fail.

[mikedick6969]

Wish I hadent overlooked this sticky! I t would have saved me lots of frustration. Sometimes thats the best way to learn though. Excellent sticky bro!!! If I saw this 2 weeks ago when the new Corsair came, I wouldnt have been so bitter up til a week ago. Also excellent wording and organization, very easy to follow! I rate this a 5! THANK YOU!!!

[Doctor]

Thanks Jason!

[phobix]

Add memtest86...Forget about my previous comment about Prime...It's not as good as Memtest86...Memtest86 rocks for memory stability testing and burning in

[Drozdov]

memtest86 and loop test #5

[Overthehill]

At first I thought it was a load of S**t, as an Electrician I'm somewhat trained in the goings on in semiconductors, but when I first put together my rig I couldn't get the fsb over 170. 166 was just stable enough. Everything (Motherboard, CPU,RAM), was new. Now it's been a month and a half, with the last two weeks folding and I zipped the fsb up to 180, at stock volts. My video card freaks if I go any higher, or I'm sure I could. I can only attribute this to the "burning in" theory. It had to evolve from something. I wonder if it can be quantised?

[mcmagostini]

Originally posted by phobix
Very Helpfull 2 thumbs up!

i reinforce that

[shaggy2003]

Originally posted by DoGMaN
I'd say that this may be the most important thread I have seen in a long time. Personally I swear by burning in my components. Usually for at least 48 hrs.

An other good program to run is memtest86 from www.memtest86.com. Make the disk and boot from it. When it starts press C then 2,5,5 hit enter then 0. This will loop test 5 which is the most stressful on your memory. When you think you are at the limit of your ram, run test 5 looped for 3-4hours. If you can successfully make it through with no errors, you can try another Mhz or two. Be warned, this will heat your memory up a lot. If you don't have heat spreaders, be sure to watch your system very carefully.

That is a good suggestion, but if I might add one thing. When you run Memtest-86 and press C to get to the configure screen, after entering 2,5,5,enter, press 3,3 then enter again. What this does is loops test 5 (like you mentioned) but it also addresses the entire range of your memory. This way you get a more robust test on the memory by using the entire range.

[XeroX]

wow dude good help!

[katana]

nice i guess illt ry to reburn my memory in

[Anjr00]

I'm gonna burn in my Samsung RAM (its labeled as Apacer, uses Samsung chips), cause it won't run stable at 200 MHz with even slightly tuned timings. Without tuning the timings, I get like 2500 MB/sec, but if I can run the the system stable enough, I can get 3200 MB/sec by increasing my timing a little more. Hopefully if I burn it in, I'll get better scores.

[ca2l3vin]

Excellent Guide Jason!