Also, it's flashed to latest BIOS already.

Have you tried cleaning the edge contacts on the DIMMs with a rubber pencil eraser?

Poor connection from oxidation will do funny things.

Make sure it is white vinyl, not pink #2.

The white ones on eversharp pencils are fine, but white vinyl art eraser is better.

I consider that MB one of the best MB ever, I don't remember any special RAM problem with it , not with DDR nor with DDR2 .

But failed RAM were a thing (and still are), so aside the connector cleaning already suggeste by other users the next obvious step is to try a different couple of RAM sticks and to try to swap the position and combination of memory slots (regardless what the manual suggests, is not uncommon to have wrong info there)

I did a full cleanup with erasers + ispropanol + brush, also brushed and cleaned the slots themselves. The sticks didnt look to be that dirty. Felt a small eureka when the test passed one time but after 45min in memtest86+ I recieved 5 errors again like a slap in the face.
This is my first almost finished retrobuild after multiple obstacles and this is the last hassle 🤣. Thinking of giving up and throwing away the other stick. I just want to know why!

The Solutor wrote on 2026-09-11, 16:18:

I consider that MB one of the best MB ever, I don't remember any special RAM problem with it , not with DDR nor with DDR2 .

But failed RAM were a thing (and still are), so aside the connector cleaning already suggeste by other users the next obvious step is to try a different couple of RAM sticks and to try to swap the position and combination of memory slots (regardless what the manual suggests, is not uncommon to have wrong info there)

Yeah it's just weird when they pass multiple passes when slotted solo.

Does Northbridge chip have heatsink or heatsink and fan? Seems like your memory controller is overheating or overstressed.

wadstroem wrote on 2026-09-11, 16:38:

Yeah it's just weird when they pass multiple passes when slotted solo.

Well I have in front of me an Intel mobo for (sandy/ivy bridge CPUs) that works perfectly with its original ram sticks, but refuses to work with any brand new stick i tried (and I tried a lot of them). In short RAM compatibility isn't an exact science.

BinaryDemon wrote on 2026-09-11, 16:46:

Does Northbridge chip have heatsink or heatsink and fan? Seems like your memory controller is overheating or overstressed.

It's an AMD board, the CPU is the memory controller.

Install a 939 CPU and you control DDR, Install an AM2 CPU and you control DDR2.

wadstroem wrote on 2026-09-11, 16:38:

The Solutor wrote on 2026-09-11, 16:18:

I consider that MB one of the best MB ever, I don't remember any special RAM problem with it , not with DDR nor with DDR2 .

But failed RAM were a thing (and still are), so aside the connector cleaning already suggeste by other users the next obvious step is to try a different couple of RAM sticks and to try to swap the position and combination of memory slots (regardless what the manual suggests, is not uncommon to have wrong info there)

Yeah it's just weird when they pass multiple passes when slotted solo.

The memory controller and the RAM itself is going to have to deal with different timings when in dual channel vs single channel.

Dual channel basically switches back and forth between both sticks when reading and writing, which is how you get much faster throughput and lower latencies in multi vs single channel.

It could very well be the CPU.

How you tried underclocking the CPU and/or increasing the memory controller and/or CPU voltage.

Also, what PSU are you using? How old is it?

The Solutor wrote on 2026-09-11, 16:51:

BinaryDemon wrote on 2026-09-11, 16:46:

Does Northbridge chip have heatsink or heatsink and fan? Seems like your memory controller is overheating or overstressed.

It's an AMD board, the CPU is the memory controller.

Install a 939 CPU and you control DDR, Install an AM2 CPU and you control DDR2.

Ah I knew some generations of AMD cpu's had the memory controller onboard, but wasnt sure where the cutoff was.

cyclone3d wrote on 2026-09-11, 17:04:

The memory controller and the RAM itself is going to have to deal with different timings when in dual channel vs single channel. […]
Show full quote

wadstroem wrote on 2026-09-11, 16:38:

The Solutor wrote on 2026-09-11, 16:18:

I consider that MB one of the best MB ever, I don't remember any special RAM problem with it , not with DDR nor with DDR2 .

But failed RAM were a thing (and still are), so aside the connector cleaning already suggeste by other users the next obvious step is to try a different couple of RAM sticks and to try to swap the position and combination of memory slots (regardless what the manual suggests, is not uncommon to have wrong info there)

Yeah it's just weird when they pass multiple passes when slotted solo.

The memory controller and the RAM itself is going to have to deal with different timings when in dual channel vs single channel.

Dual channel basically switches back and forth between both sticks when reading and writing, which is how you get much faster throughput and lower latencies in multi vs single channel.

It could very well be the CPU.

How you tried underclocking the CPU and/or increasing the memory controller and/or CPU voltage.

Also, what PSU are you using? How old is it?

Yeah my thought was also the CPU.
I have not tried underclocking or increasing CPU voltage, only DRAM voltage without success. How would one go forward doing it?
The PSU is Corsair HX650W from 2009, I changed the fan so it shouldnt be heat issues.

Funny that it still crashes (bluescreen, but not BSOD) when I launch some games, but only sometimes. That is with one stick only.
I guess I'm gonna keep a lookout for new sticks, hopefully some NOS.

wadstroem wrote on 2026-09-11, 18:34:

It could very well be the CPU.

Yes it could, obviously.

But my personal statistic (hundreds of machines per year in the golden years) tell's me that the RAM has more chances to be the culprit.

To put in numbers I faced hundreds of dead or unstable RAM sticks, while in three decades I collected the huge number of *five* malfunctioning CPUs obviously not counting the burnt Athlon XPs because a poorly installed heatsink, in that case the CPU was gone in seconds, often before reaching the BIOS summary.

wadstroem wrote on 2026-09-11, 18:34:

Yeah my thought was also the CPU.
I have not tried underclocking or increasing CPU voltage, only DRAM voltage without success. How would one go forward doing it?
The PSU is Corsair HX650W from 2009, I changed the fan so it shouldnt be heat issues.

To rule out the CPU as a culprit, downclock it to 1.8 GHz (9x200). If that doesn't help, try Prime95 under Windows and check if the CPU temps skyrocket in a couple of minutes (that would be a sign of a dried paste under the heatspreader - socket 939 CPUs are not soldered!).

Heatsinks on the northbridge and southbridge are passive and paste beneath them is today certainly dry. Replace the paste, especially northbridge can run very hot on this motherboard.

Occasional blue screen in games could indicate GPU or driver problem as well, but first things first - you need to get rid of the errors in memtest.
The PSU is also very old and could be the reason of blue screens in games.

Important options in BIOS:

Spread Spectrum - [Disabled]
Cool'n'Quiet - [Disabled]

Memory Clock - [400MHz]
Flexibility Option - [Disabled]
MA Timing - [2T]
Burst Length - [4 Beats]
other memory options - [Auto]

I'm a fellow 939Dual-SATA2 owner here. Actually, I got two of them (both obtained used / 2nd hand). My relationship with these boards is a love-hate one... but more on that for another post.

There are multiple things that can cause this board to be unstable. @Mister_No hit the nail on the head with many of the suggestions he made. But I will add/elaborate a little further.

1st order of business: go into BIOS and DISABLE anything relating to SPREAD SPECTRUM.

2nd: try another CPU if you have a spare. A lower-power one would be highly advisable - something like the 3200+ with the Venice core.

#3: right under the CPU, there are many rows of jumper blocks. These are for the AsRock "CPU Upgrade Card". On my 2nd board, these were in very poor condition and causing all sorts of issues. As such, it is a good idea to remove and reinstall these jumper block a few times to clean up their contacts. If you have the nerves to go through each pair of jumper headers on the back of the motherboard to verify that they are connected by the jumper (as they should be), this would be even better.... though it is time consuming and tedious.

#4: point a large fan towards the board's Northbridge and Southbridge. These run very hot. In very hot climates, the chipsets could get permanently damaged. In my case, my 1st board has now a failing Southbridge, which was actually the main cause for the board BSOD/blue-screening on me after I used it extensively one hot summer. First, the SATA II port died. Then followed the onboard sound. Then then the onboard LAN went out only to come back "alive" about a month later... though it still crashes the system with a BSOD with heavy traffic going through it. The USB ports also did something similar, but rarely crash the system. What's interesting about my case is that CPU and memory-intensive tasks NEVER crash the system, or anything relating to the Northbridge (i.e. the PCI-E bus). In contrast, PCI bus experiences regular corruption (most notable on PCI video cards, where I get random artifacts... that is, if I can even get the PC to work.) And no, it's not a PSU issue, because I've tried several recapped and known-good PSUs in that system. It's just a dying Southbridge and nothing I can do about it. But if I don't use the LAN or SATA/SATA II ports or PCI slots, the system is rock-stable. So it's become my GPU test system, since the PCI-E and AGP slot are OK. In any case, I now only run this motherboard with a large fan pointed towards the chipsets... and with that, I've gotten a good 10 years of use out of it as a GPU test system.

#5: these boards use OST RLP or RLS caps for many of the small 6.3V 1000 uF caps scattered around the board. My 1st board was an earlier revision (based on the BIOS sticker) and used RLS caps. From these, not a single one went bad and I still have the caps in my spare parts bins for reusing/testing purposes. However, my 2nd 939Dual-SATA2 came with OST RLP, some of which were bulging when I got the board (and presumably why I got it for cheap, since it wasn't stable with those bad caps.) Therefore, just because the caps look "OK" on your board doesn't mean that they are. The CPU VRM caps could also be in the same boat, depending on which manufacturer AsRock used on your board. For example, my 1st board uses United Chemicon KZE for the CPU VRM high side (12V rail) and OST RLX for the VRM low side (CPU V_core). The latter are known to be problematic on motherboards (though in the case of the 939Dual-SATA2, probably a little less often, since AsRock seem to have done a pretty decent job with designing the CPU VRM to be pretty low-stress for the caps.) On my 2nd board, the CPU VRM high side caps are the same, but the CPU VRM low side caps are all Rubycon MBZ, if I remember correctly - a very reliable cap brand and series. That said, I do remember seeing a few of these boards with United Chemicon KZG, and those are also known to be problematic, even when they are not bulging.
So better check what's on your board and make not of it here, just in case.

momaka wrote on 2026-09-15, 22:13:

#3: right under the CPU, there are many rows of jumper blocks. These are for the AsRock "CPU Upgrade Card". On my 2nd board, these were in very poor condition and causing all sorts of issues. As such, it is a good idea to remove and reinstall these jumper block a few times to clean up their contacts. If you have the nerves to go through each pair of jumper headers on the back of the motherboard to verify that they are connected by the jumper (as they should be), this would be even better.... though it is time consuming and tedious.

Hehe, I think I still have the tools to remove the jumpers all at once. For sure I still have the CPU Upgrade board.

Perhaps there was the OverclockerWorkBench forum, that was, more or less the official place were users of that legendary board used too meet, and there was a long string of unofficial updated firmware's with real meaningful improvements. It's obviusly gone like most of the useful part of internet, not sure about its status on archive.org, anyway the thread on HWupgrade (in italian) still survives. and it's 1860 pages long!!!.
https://www.hwupgrade.it/forum/showthread.php?t=998913

That data alone tells something about how versatile and long lasting this board was (and still is)

I'll never forgive Nvidia for buying ULI just with the purpose of killing the best chipset of the era, the same thing that Creative did with Aureal (another legend).

939 CPU's have popular memory controller bugs that include single and double sided sticks of ram also how they are populated.
My 939 won't allow me to use more than 1.5gb ram to even post.

zapbuzz wrote on 2026-09-16, 07:31:

939 CPU's have popular memory controller bugs that include single and double sided sticks of ram also how they are populated.
My 939 won't allow me to use more than 1.5gb ram to even post.

OK, let's not exaggerate the issue. E6 revision Socket 939 CPUs can perfectly handle 4x1 GB of double rank DDR400 RAM at 2T at 200 MHz (DDR400). Heck, my Opteron 185 does it even at 233 MHz (DDR466), with memory sticks being at standard 2.5V. Older revisions (E4, C0 and CG) need to be set to DDR333 when using 4x1 GB of double rank memory modules, but your mileage may vary.

What CPU do you have? Refusing to POST could have (and probably has) some other cause, rather than some mysterious memory controller bug. First thing that comes to mind as a probable cause of no POST is that one of the memory slots on your motherboard is flaky.