DDR DIMM Slots = Russian Roulette

DIMM Slot Positions effect memory stability on various motherboards – Joe

UPDATE 5/11/00:

Since I wrote this up, I have been attempting to find chipset specsheets to document what some of the chipsets support in the way of memory. Intel as always provides a lot of specs freely available on their website:

“The MCH supports a maximum of 2 double sided DIMMs (4 rows populated) with unbuffered DDR200/266 (with or without ECC). Note that in mixed mode, populating ECC and Non-ECC memories simultaneously is not supported.”

Source: Intel 845 Chipset: Memory Controller Hub (MCH) for DDR, Datasheet (Jan 2002)

Say anything about Intel, but at least they are very clear about what their chipsets can do.

Thinking VIA would be as clear, I plowed through their site looking for a similar document – could not find one. They have a nice support forum, so I spent some time there looking for an answer and was told:

“So, for your answer – the KT266a can have up to 4 DIMM slots (8 banks), for a total of 4gb of RAM (4x1gb)”

Source: FORUM Thread

I was referred to this on ABIT’s site:

“According to the datasheet of VIA KT266A chipset (Revision 2.1, September 14, 2001), x4 bit memory chip is not supported.

Page 3, Advanced High-Performance SDR/DDR DRAM Controller

Supports 8 banks up to 4 GB DRAMs (512Mb x8/x16 DRAM technology) for registered SDR/DDR modules
Supports 6 banks up to 3 GB DRAMs (512Mb x8/x16 DRAM technology) for unbuffered SDR/DDR modules”

Source: ABIT

I am still attempting to get this document referenced from VIA – so far, no luck.

Finally, Marton Balog over at ProHardware (in Hungarian) sent me this:

“SiS 645 boards only support 4 banks of PC2700 DDR SDRAM, so being able to run only 2 double sided DIMMs is exactly what the specs say! And naturally, this is true for all SiS 645 and 645DX boards out there, as this is a chipset limitation.”

Unfortunately, my attempts to locate the specs from SIS were not successful, but I have no reason to doubt Martin’s accuracy.

The upshot of all this is that having four DIMM slots DOES NOT mean that you can stuff them with as much RAM as you’d like. Motherboard manufacturers are a little coy when it comes to fessing up to exactly what their boards support – their website spec sheets don’t usually get into these specifics, and some manuals gloss over what is possible.

Even if the CHIPSET supports 8 banks, if DOES NOT necessarily mean that the board will run 8 banks flawlessly. It’s up to the manufacturer to design a board will support the chipset’s capabilities, and if you think that happens all the time, I have a bridge to sell you.

The way motherboard’s address memory slots will effect how DIMM slots should be populated. Not all motherboards use a 1 2 3 4 scheme; some use 1 3 2 4. So if you’re having some stability problems, try different DIMM slot positions before RMAing your RAM.

Finally, for general use, I think it prudent to use no more than two DIMM slots. Anything above that may get you very frustrated.

SUMMARY 4/4/02: Interference patterns with multiple DDR modules may appear in just about any motherboard.

After I wrote an article about DIMM slot positions Abit’s KR7A-R, I’ve been getting emails from folks who tell me either they had the same problems or similar problems with another motherboard. Curious as I am, I decided to broaden the inquiry into DIMM slot positioning and RAM performance for a number of other boards:

• Abit KR7A-R, Athlon XP (VIA KT266A)
• Gigabyte PIV GA-8SRX, Northwood 1.8 (SIS 645)
• Gigabyte PIV GA-8IRXP, Northwood 1.8 (Intel 82845)
• Iwill XP333 2.1, Athlon XP (ALi M1647)
• Shuttle AK3.1, Athlon XP (VIA KT266A)

I tested the following double sided 256 MB DDR modules:

• Crucial DDR 266, CAS 2.5
• Kingmax DDR 266, CAS 2
• OCZ DDR 333
• Princeton DDR 266, CAS 2

I purposely wanted to test a mix of modules to see how these boards would cope with such diversity. I used RAM Stress Test Professional (R.S.T. Pro) as the test tool. This is a stand-alone diagnostic card that plugs into a PCI slot. The card is self-booting – there is no OS to get in the way of results. According to UXD, the RST Pro has

“…the capability of testing and validating RAM modules in a system. On many occasions, memory modules that pass standalone testers will fail using R.S.T. Pro. This is caused by the system chipset, noise, and several other factors. R.S.T. Pro in this environment, is the best, easiest, and fastest way to validate, and test system RAM.”

I have used RST Pro before with great results – it tests stability and high FSBs more rigorously than running benchmarks in Windows. With this card, not only can you test system RAM, it will tell you which chip on a memory module fails.

In the tests that follow, conservative memory settings were used in all cases.

¹With this configuration, the board would only recognize one DIMM.
²With this configuration, the board would only recognize two DIMMs.
The Gigabyte manual clearly indicates these limitations per Intel’s 845D spec.

Hmmm…a LOT more RED that I expected.

I found that I could run just about any SINGLE stick of RAM³ for hours without any errors, so individually each DDR module was OK. However, as soon as another module was added, DIMM positioning became the controlling factor. When errors occurred, they showed up within the first 10 minutes of testing (note: each test cycle takes about 30 minutes).

When RAM failed in multiple installations, usually only ONE RAM module showed errors; errors were massive – usually every chip on the module in the first DIMM showed errors. This strongly suggested that DIMM positioning was determining stability.

Frankly, as I was testing, I was dismayed at the results. Boards that are populated with 3 or 4 DIMM slots should work, but some boards are not able to accommodate RAM in every slot. This is limited testing to be sure, but it’s interesting to find that the PIV boards do a better job than the AMD boards.

The upshot of all this testing is to seriously question the viability of many motherboards on the market to fully populate DDR DIMM slots with mutliple RAM sticks. Are three and four slot boards more for advertising copy rather than serious use?

This admittedly limited testing opens up a host of issues, not the least of which is the failure of the hardware sites to adequately report these problems (I include myself in this group). These issues are routinely reported in Forums around the web, so I think I can safely say these findings are not exclusive to the boards I tested.

One piece of advice that comes out of this effort is that using one stick of quality, JEDEC standard 256 MB or 512 MB RAM will yield more stability than using multiple sticks.{mospagebreak}

³Some boards did not like either the Princeton or OCZ in any slot, and in those instances these were not used.

Email Joe

Ed Note: These are some of the responses we received on DIMM experiences; some are excerpts from larger messages. Thanks to all who took the time to respond.

Jeff Haole
If they tell you where to put the RAM, is it really a problem? Is the added cost and complexity of designing the board to work with any and all possible DIMM configurations practical? I wonder if the other mobos with 4 DIMM slots have similar issues.

I run 2 DIMMS in mine, and I never had any problems because I read the manual. And as for RAM stability: If the thing runs at 188FSB for weeks on end crunching G@H while going about my business WITHOUT a reboot, would you call that unstable? Calling the KR7A a beta product is just absurd.

Falceron
I thought it was proper procedure to populate RAM slots from slot 0 (or 1) onwards. I believe every mainboard manual I’ve read has mentioned this. I just fail to see how this is really a problem. It would be interesting to see how well some other boards perform in this test, though.

I haven’t experimented with more than two ram modules in a long time, but I know that there’s less of a margin of stability as you add more DIMMs.

guru dude
Anyone in their right mind would NOT buy an Abit board and use different memory… I love criticizing you guys.. you need more validity so I hoped that helped…

“Me”
Please indulge my questions for a moment also. I do understand how you
feel about companies making beta stuff just for us to spend our money
testing, but I do have an honest question or two.

1. Why would somebody use the RAM slots “out of order” anyway? I’ve
   never had a problem with other stuff in the way of the slots.

2. Although everything should work perfectly on a purchased product,
   this seems to me to be a “quirk” rather than a defect since the simple
   solution seems to be to use you RAM slots in order and then everything
   works fine. Is there something that maybe I’m not understanding that is
   more serious here?

ce|cho
Old computers, like the old Pentiums (i’s i mean) used to have to have memory installed in a certain configuration. it is a very common thing for computers to do, and the only reason most of us run into problems is that we don’t read the manual until we are having problems (everyone does that) and then we expect a high performance motherboard to run with any old ram in any old configuration.

Roger W Turner
It seems to me that the Gigabyte GA-8IRXP footnote gives the game away.

Your articles could be largely rewritten from a different perspective:
How the manufacturers are letting themselves and their customers down by
not documenting the correct slot-filling rules.

Of course, this doesn’t excuse the “advertising copy rather than serious
use” point about having one slot more than is usable (where the
GA-8IRXP appears also to be guilty).

Some MB manuals certainly do document the rules: Order of filling slots;
modules tested; and even rules that effectively acknowledge that there’s
one slot too many.

• Abit KR7A:
  3 GB Unbuffered, 4 GB Registered
  “…we suggest that you be populate DIMM1 to DIMM4 in order.”

• Iwill XP333
  No restrictions – “Memory capacity up to 3 GB”

• Gigabyte GA-8IRXP:
  “The motherboard has 3 dual inline memory module (DIMM) sockets, but it can only support a maximum of 4 banks of DDR memory.”

• Gigabyte GA-8SRX:
  “Supports up to 2 un-buffer DIMM DDR333 or up to 3 un-buffer double-sided DIMM DDR266/200…3 GB max…supports ECC”

• Shuttle AK31:
  “…up to 4 GB system memory…PC1600/2100 compliant…non-ECC DDR.”

Kristjan
I suggest that you should try these memory location tests with registered
DDR memory, because

1. Motherboard manual [Abit KR7A] says that supported maximum amount unbuffered memory is 3GB and registered 4GB;
2. I had similar problems with Asus A7M266-D board (also 4 DDR dimm slots) and found out that particular ASUS supports 2 unbuffered dimms with maximum memory size 2GB or 4 registered dimms with maximum memory size 3.5GB (not 4GB).

So it seems that electrical loading is still a big issue with DDR and
registered memory seems to help. And that’s logical because with registered memory, memory bus gets only the load of input registers, not all the memory chip’s, and each register on each dimm gets the load of only that dimm’s memory chips.

mike
Just wanted to let you know in order to use all 4 slots, you have to have ECC RAM – at least for the Shuttle it’s that way. I would imagine it is probably the same for the other boards, since they’re all the same chipset. You didn’t mention if you used ecc ram [NO] when you used all four slots, so that is why I brought it up.

DaveA
Thanks for the info. I have a Shuttle AK31A (KT266A) with the four DIMM slots. I have only run it with all four slots populated with 256mb Crucial DDR266 2.5t. (1 gig total) I built it with all four filled and have had no memory issues in “normal” (what ever that is) use. I have never run a memory test but I haven’t had any crashes either. Have you tried filling all four slots with matched memory? Thanks again for the effort.

Michael VanLoon
Most motherboards say that you need to use registered memory if you’re going to fill all the slots. Testing with registered memory might yield some interesting results. I’d be very interested in seeing the results of that.

I know even my SDRAM board (MSI 694D Pro2 [Via 694 dual PIII]) wouldn’t run stable with all four DIMM slots filled unless you trimmed back the memory timings to really conservative values, if you used unbuffered memory. But once I threw four registered DIMMs in there, I could clock it very aggressively with no problems at all (both sets were Crucial memory).

Railer39
WOW! Very interesting topic indeed. It is funny that it has been such a long time since computers are made and we are still suffering problems of RAM incompatibility. I remember I had to play around with my memory locations just so both chips would work on my P2. Isn’t it about time to set a standard and fix these problems?

But one question, is it Abit to blame? Via? Crucial? Or the whole industry is not working together?

Steve
Just wanted to drop you a note. I really liked your article on memory slots and stability/MB quality. This has been an issue for me for years as I generally fill up all slots in all the systems I’ve ever used. Plus
the size of the memory I use (my ‘small’ systems usually have 1GB of RAM, and I have others up to 3GB).

I’ve played this game a lot and you are right – there is literally zero info for a scientific test about this on any site. All hardware sites seem to just go after the gamer market and use one stick of ram.

Chris
We have known for years that populating multiple DIMM slots causes stability issues – even back with PC100 and PC133 SDR modules. As the performance delta of new hardware increases, so does the risk and the prevalence of such stability issues due to a variety of interference. Just imagine what those GF3 and GF4 boards are doing to interference, let alone their fans and heat.

I even recall problems with two PC66 SDR sticks in an ASUS P2L97, and later with a pair of PC133 sticks in an ABIT KT7-RAID (KT133 not ‘a’) running at 133… though they ran fine at 100, another ridiculous compromise for a hardware implemented feature that should just work…

I agree with you completely – if they populate a board with three or four slots, they should all work (let alone two!). They are cutting corners and blatantly misleading consumers. This equipment is not cheap, and should work as designed.

David Goddard
I’m using a Shuttle AK3.1 with 3 sticks of 256MB Crucial ram in slots 1,2,3 with no problems at all.

Colin O’Neill
My old KT7A-R had the same problem I sold it to a friend he had the same problem. Until the latest BIOS came out, the board would work just fine with any three slots fill the fourth…

PG
This issue seems to apply to the Iwill XP333 as well. I had two identical PC2700 256MB Samsung chips. Upped voltage to 2.70 on ram; runs fine at 200 MHz (400 MHz DDR) if using one chip in first slot. But both chips in various slots failed to boot at 200 MHz. I didn’t test default of 166 MHz. Why bother? 😉

Michael Niesslein
I am very glad that people are analyzing this. [Abit KR7A] I know that I have DIMMS 1 and 2 populated with Corsair XMS2400 (2 x 256) and I have instability issues due to the RAM. I get thousands of errors in mem test86 tests 5 and 8 with DIMMS 1 and 2 populated. I purchased both DIMMS at the exact same time and the instabilities are pretty bad.

cybercfo
There are lots of reports in the Soyo usenet newsgroup about stability
problems when all three DIMM slots are in use on the Dragon+ (KT266a),
particularly when the sticks are 512MB. Same kinds of symptoms.

Michael Snyir
I want to say thank you for a good read on “DDR DIMM Slots = Russian Roulette “. Over at amdmb.com Soyo motherboard forums there have been numerous posts regarding this issue with the Soyo Dragon+. Soyo even put out an updated BIOS that supposedly fixes this issue but it does NOT in many cases. I wish you had tested this MB because of the rampant boot failures that have been seen.

Soyo has been taking a pasting over this issue for many months now, glad to see it’s a lot more generalized then first believed. I never had an issue with my Dragon+ on this, so it’s hit or miss…don’t know!? It’s a shame many have blamed one company.

KrK
Abit KG7, ran great all day with 256K DDR. Add in the second, and system stability went SOUTH in a big hurry (It would even sometimes freeze solid during POST.) Played with the DDR’s and finally got it working well.

Frank
If you happen to be interested, I originally purchased an Abit KT7A for the computer I’m currently using and discovered the hard way that it was horribly unstable when I filled all three memory slots. I was, needless to say, extremely annoyed. The only suggested “solutions” involved “not using all the slots.”

What the heck was the point of HAVING three memory slots if I couldn’t use them? When I replaced the Abit motherboard with an MSI board, all of the instability problems went away (same memory sticks, of course).

Robert & Melissa
I have seen a lot of concurring evidence on your memory position report. I would like to add that my board was completely unstable with my two sticks of Crucial memory until I removed the northbridge hsf and redid ABIT’s sloppy application of the thermal paste. Before, I had hard locks with certain memory settings, especially any interleaving turned on and would not run at all in slots 1 and 2, but 3 and 4 would work – but not 100%. Now I am at 100% with memtest86 all timings at max at 145 MHz.

Daniel
That almost sounds like a problem I had with an 815ep (Insmell) Northbridge – even if the stick of RAM only has the ram chips on once side, it still might not be a true single sided/ single banked stick of RAM. It all has to do with the way that the SPD of the RAM is programmed.

I had to find this out the hard way; after bitching out some of the people at Asus, they finally found the problem. It has to do with the SPD. Check the limits of the Northbridge and the check the SPD programming on the Ram and I am sure you will find the problem.

More or less what happened with mine:

Could boot Windows, would see all the RAM but would BSOD all the time and end up eating the files on my hard drive. I lost 6 OS installs within a two week time frame. I was told the best thing I could do was buy Corsair or Mushkin because they tested their RAM on Asus boards.

Jim Thurber
I have two Crucial PC2100, Cas2.5, 256 mb sticks. On my KR7A-Raid board, (Win98se, Raid 0) with one stick of memory installed in slot 1, complete stability. With the second stick installed in slot 2, all kinds of weird problems, freeze up, FS2002 pro in mid flight going to Win98 desktop, and other unusual happenings.

Placing the second stick in slot 3 (1 stick of 256 mb in slot 1, the second in slot 3, total 512mb) I have complete stability. Works for me!

Denis Lomax
I am suffering from the same memory problems with this board as you report. When using two memory sticks, I get weird “available memory” figures in System information; however when using either stick, the system reports the correct amount.

Mike Barrett
I have an Abit KR7A133 board (XP1800+, 256MB Crucial) and have experienced numerous BSODS, system resets and lock-ups. I have followed many threads and tried most of the suggested fixes, but the only thing that works for my system is to back-off DRAM frequency to 100MHz in BIOS. At this speed, the system is totally stable. I would like to try another memory module from a different source, but I am loathe to throw more money at the problem; for a similar sum, I could bin the Abit and buy a mobo that does work reliably!

I feel sorry for the other people struggling with this problem. Some of them have wasted large sums of money purchasing exotic high power PSU’s (a favourite, but erroneous, cure-all), other makes of memory and finally in desperation, other brands of mobo. I strongly agree with your comments that Abit should not foist beta products on a gullible public; as you say, the manual implies that they are aware that a problem exists. A little more honesty and much more development work would be greatly appreciated!

Philip Cooke
I am running an Iwill XP333R with 2 x 256MB 2100 DDR modules Running Windows XP. Quake 3 Arena hangs the machine after about 10 minutes, but this can be VERY random! Removing 1 x 256MB memory module cures the problem – no hang-ups at all !

Sheng
I have same problem with SDRAMs.
I just bought a HIGH DENSITY 256M SDRAM,
it is recognized by INTEL 815E MB as 256M,
but it could not boot WIN2000. On the 4th
slot of ASUS P3B MB, it failed to start — beep,
and on the 2nd slot, it starts and boots fine
and works smoothly and I suddenly found that
it says 128M in my system. Funny, 🙂

Hallvard
Thank you for a great article. Inspired by your work, I did my own
testing on my Abit KR7A-RAID. I use 2x TwinMos 256MB PC2700 CL
2.5@166Mhz. I found that slot 2 and 4 was the best solution on my board;
actually it was as night and day. All other slot configurations failed
at high FSB and agressive timings, but with slot 2 and 4, I could run the
most agressive timings at high FSB, up to 158 FSB, with my XP 1800+ max
outed.

Clay
Thanks for the info, I thought I’d contribute my experience with my KR7A -133R.

• 1 512 Mushkin pc2100 222
• 1 256 OCZ PC2400 233 (both running at 150 MHz CAS2)

The ONLY positions these will work in is OCZ in slot 1 and Mushkin in slot 4. I was about ready to send the OCZ stick back thinking it was bad, but I decided to try re-arranging the sticks. No other combination or order will work.

Richard M. Koscielski
I read with great interest your article on DDR Dimm slots & Russian Roulette. Having just completed a new system about a week ago, I was ready to try some overclocking. It has the above motherboard, XP2000+ locked, 2 sticks of OCZ 2400 DDR memory, two ATA 133 hard drives, Zip, CD, and a 64 MB DDR video card. OS is Win XP.

Obviously, my only option was to raise the FSB, no go. Using the SPD memory settings, it would not boot at 134 FSB, but completely stable at 133 for days on end. Then I remembered reading another article about Dimm slots.

I removed one stick, works fine at 144 FSB. Put it back in slot 3 (1 & 3) works fine at 144 x 12.5 or 1800 MHz; been running stable now for almost 12 hours, no crashes. Then I changed the Bank interleave from default to 4, and the CAS from 2.5, to 2 – all is still well. Each set of components sets up different variables. Sure is nice to know there are people out there like you who publish articles like the one you wrote, that help people like me.

Pete
For the past ten months or so, we have used two-DDR-slot motherboards and paid the premium for high-capacity DIMMS, rather than spend the time to work out which combination (if any) of which DIMMS would give reliable performance on a particular 3- or 4-DDR-slot motherboard.

I think too many motherboard manufacturers downplay this. It looks like a generic issue affecting most, if not all, motherboards to varying degrees. Once a board goes over two slots, things can get dicey very quickly.

The pressure on motherboard manufacturers to get product into the market quickly, with eye-popping features (“Supports up to 4 GB RAM!”) that are more marketing hype than practical, cheats consumers. For many of our readers, a motherboard costing, on average, $150 is NOT a trivial investment – for this money, we expect everything to work.

If you buy a car that advertises it can go 120 mph, then find out at 65 mph the wheels fall off, I think you would get upset. While a motherboard is not potentially a life-threatening product, the principle is the same:

I want stuff to work as advertised!

Email Joe