RAM: 2 DIMMs okay, 4 DIMMs not

[jesterb84]

Recently, I ran into errors with my computer where my system would blue-screen or games would crash randomly, although not often. In addition, extracting large archives (i.e. RAR files) would error out with CRC fails sometimes and other times the entire archive would extract fine. Originally, I attributed this to my overclocked CPU failing, but another member had advised to check the RAM, which is what I proceeded to do:

I restored my CPU to stock clock/vcore and aimed to test my 2 DDR2 RAM kits and the mobo's 4 DIMM slots separately. First, I removed all the RAM and placed my first 2GB kit (2 x 1GB) in the 1st and 3rd DIMM slots of my motherboard (channel 0) and ran Memtest HCI to 2000% with no errors. Then, I removed all the RAM and placed my second 2GB kit (2 x 1GB) in the 2nd and 4th DIMM slots of my motherboard (channel 1) and ran Memtest HCI to 2000% again with no errors.

Things started to get a little ugly when I placed all 4 sticks back together. Once, I left Memtest HCI on overnight with all sticks in and woke up to find that my computer had restarted itself. Another time, 1 error was detected at about 1300% testing. Yet another time, Memtest HCI suddenly complained that the OS (Windows 7 64-bit) was unable to test all portions of used RAM and to run multiple Memtest sessions simultaneously. This was ludicrous but I followed the instructions anyways. When I ran 2 copies of Memtest HCI, multiple errors popped up during the pre-test stages!

Therefore, I concluded there was something wrong in placing all 4 sticks together, even in a system with absolutely no overclocking. I have a few questions:

1. How reliable is Memtest HCI? I heard that it was likely to detect errors that memtest86+ would miss out on, but the fact it runs in Windows has me doubting whether it checks the entire RAM or just the portions which happen to be free during runtime.

2. Assuming Memtest HCI is 100% reliable, would this mean that at stock settings, my system cannot run 4 DIMMs reliably? Would adding more voltage to the DDR2/FSB/MCH settings in the motherboard BIOS increase stability?

I should also note that I have been using the first 2GB DDR2 Kingston kit since I built this system 2 years ago without issue. The second 2GB DDR2 Kingston kit is exactly identical and I got that about 1 year ago and for all the times up until recently, there appeared no issues with that as well. The errors just suddenly happened a few months ago...

 

[Elixer]

No idea about Memtest HCI, I think most people use memtest86+ or memtest86.

It is pretty much impossible to get exactly identical DIMMs in that timespan.

If you get errors in memtest86(+) then usually, you can either pump more juice, or relax the timings, or both.

 

[jesterb84]

No idea about Memtest HCI, I think most people use memtest86+ or memtest86.

It is pretty much impossible to get exactly identical DIMMs in that timespan.

If you get errors in memtest86(+) then usually, you can either pump more juice, or relax the timings, or both.

Yes, well no 2 sticks of RAM are absolutely 100% identical, even if they are in the same kit. I'd like to then rephrase that part by saying the manufacturer, model, part numbers, and specifications are identical. 🙂

Thanks for bringing up memtest86+. I will probably give that a try by making a bootable USB thumb drive version. Additionally, what setting would I need to adjust in the BIOS - MCH or DDR2? All 4 sticks are supposedly 1.8V.

 

[n7]

If two DIMMs = stable & four = unstable, you generally have a NB issue.

Usually the "solution" is to increase NB voltage (MCH Voltage in your BIOS).

You might also need to bump up vDIMM a bit also to run 4 DIMMs.
Or lower RAM speed.
Or loosen timings.
Just depends on mobo/RAM/expertise/etc.

I'd suggest running some Memtest86+ overnight.

If it's error-free, that doesn't necessarily mean your system is stable though.
It just means your RAM isn't erroring.

Memtest86 only stresses portions of the RAM at a time, which puts very little stress on the NB.

This is where HCI Memtest comes in (or LinX).

HCI is best run in as many instances as you have logical cores IMHO, so two instances in your case.

It's very important to make sure you don't use too much RAM, so i generally make sure to calculate it with some headroom.

You could do something like 2 instances each set to stress, say, 1250 MB.

As HCI is stressing all the RAM simultaneously, that is not only harder on the RAM, it's also much harder on the NB.

You could also run LinX @ max RAM, which will very heavily stress your CPU, RAM, & pretty much entire mobo...

Hope this gives some ideas.

 

[RebateMonger]

For non-buffered memory (read: most non-server motherboards), the more memory modules installed, the more chance of memory timing issues. As n7 notes, higher memory voltage, lower clock speeds, or looser memory timing can help a marginal situation.

Servers typically use buffers to ensure that voltage wave forms don't become distorted as more memory modules are added. These reduce performance and cost more, so aren't typically used on desktop boards.

 

[jesterb84]

Thanks for the great replies, and the comprehensive explanations on memtest86+ vs. HCI! Seems memtest86+ it is tonight. If that goes through in the morning, then I will up the power to the MCH a bit and proceed again with HCI before I leave for work.

 

[jesterb84]

Just to update everyone, I finished performing some tests the past couple of days and memtest86+ is showing that the older of the 2GB memory kits has issues. I'll proceed to call Kingston on Monday and see if I can RMA and get them replaced. Would not want to spend more money on DDR2 as it's definitely not worth it at this point.