Need some memory settings help

[MaxBurn]

I have a fairly new system here and I added some memory to it recently and had a couple crashes. Some testing revealed that it failed memtest86 on test 6 with 20 or so errors. I noticed that when I added the second two memory modules the BIOS automatically selected DDR333, I overrode that setting to DDR400 but left all other timing settings to auto.

Motherboard: Gigabyte K8N PRO SLI
Memory: Corsair Value Select VS512MB400 DDR400 512MB. ZPU-Z reports it's timing as 2.5, 3, 3, 8 (CAS, RAS To CAS, RAS Precharge, TRas). I Bought the memory in pairs of two (kits), one kit is maybe two years old and the other is brand new. Same part number on all memory modules and CPU-Z reports each stick has the same timing stored in it's SPD. I did check the BIOS auto configure, it has these settings stored.


Testing Results:
-All four memory modules installed, BIOS reports dual channel at DDR400: Fail memtest86 test 6.
Test 6 is: [Moving inversions, 32 bit pat]
This is a variation of the moving inversions algorithm that shifts the data pattern left one bit for each successive address. The starting bit position is shifted left for each pass. To use all possible data patterns 32 passes are required. This test is quite effective at detecting data sensitive errors but the execution time is long.

-All four memory modules installed, BIOS reports dual channel at DDR333, pass memtest86.

-Two old memory modules in position 1 and 2, BIOS reports dual channel at DDR400, pass memtest86.

-Two new memory modules in position 3 and 4, BIOS reports dual channel at DDR400, pass memtest86.



Looks like I don't have any defective parts, so I need to increase timing somewhere to make all four sticks work happily together at their rated speeds? I have never really played with memory timings and there is a lot there to change with trial and error. My first thought was to go to CAS3, looking for suggestions.

 

[CougTek]

You must have an Athlon 64 with an old core, like a Winchester. The older Athlon 64 automatically set back the frequency to 333MHz when all four banks are populated. The newer San Diego, Venice, Manchester and Toledo core don't do that. Handruin had the same problem with his Athlon 64 3200+.

The problem isn't with your motherboard IMO.

 

[time]

Like Coug said, the CPU only supports 4 banks at 333MHz.

Stop worrying, it's dual channel RAM and you'll never notice the difference.

 

[Handruin]

Unless folding at home is involved. I think CougTek speculated a 7-10% increase in productivity with the faster bus. Otherwise you're right, time...I'd probably never notice it. I am out of luck unless I ditch half my RAM.

 

[MaxBurn]

Actually the CPU is an A64 San Diego core 3700+ (single core), sorry I left that bit out. It is fairly new, like within a couple months. Motherboard is also running the F7 (latest) BIOS.

It's a new day now, I put it to CAS3 and DDR400 and am retesting now.

 

[MaxBurn]

Well it failed again, this time with only 11 errors. I then noticed that in manual mode the BIOS set the TRas to 7 instead of the rated 8. I changed that to 8 and retested, now came back with 26 errors in test #6. I guess I need to go read the BIOS guide to learn what all these mean and where to go.

I thought I was ok going with four memory modules as I believed they had solved this downclocking problem, but I guess not. Now thinking about returning the memory, selling my old stuff somewhere and getting two bigger modules.

The PC is used for gaming among other things, video card is a evga 7800 GT CO (factory overclock) and I don't really want to step down in performance. I doubled the memory from 1gig as I tried out the fake factory cinematic HL2 mod which recommends 1.5gig minimum (2gig helped performance a lot coming from 1gig), and looking around that's where games are going I guess.

 

[MaxBurn]

I have tried all sorts of timing settings and still can't get it stable with all four at DDR400. I am done messing around, going to get some Corsair TWINX2048-3200C2 2GB Kit DDR400 XMS3200 (2-3-3-6), $196 after a Corsair rebate at zipzoomfly. Pretty good buy IMO.

Lesson learned: AMD has problems running at DDR400 with four memory modules. Not even the later San Diego core fixes that.

 

[MaxBurn]

It would have been nice to have this in the motherboard manual. Shame on Gigabyte, and props to Kingston:

Maximum Memory:
2GB using 400MHz DDR Modules
4GB using 333MHz DDR modules

http://www.ec.kingston.com/ecom/con...-SLI+Motherboard&distributor=0&submit1=Search

Corsair's memory configurator didn't come out and specifically say this, but it didn't recommend four modules at 400 either.

 

[P5-133XL]

It's not as likely to be an AMD issue but rather a MB/RAM quality issue!

The RAM signals need to be high quality and very syncronized. When you put in 4 modules rather than two, the signals have more distance to travel and that means more opportunity to degarde because of line capacitance, race conditions, or simple noise. No matter how good the memory controller, it can't compensate for motherboard or RAM quality issues that limit the timings.

Sorry Gigabye and Sorry Corsair but I would much more likely be placing the blame on them rather than AMD.

 

[Tannin]

Blame? What blame? If I read the thread right, the problems only arise when you override the BIOS settings and pretend thay you know better than the experts at Kingston and Gigabyte. Set it to the factory defaults and rest easy.

In any case, how much faster would it be at 400MHz than at 333, if any? It's not a Pentium 4 that craves RAM performance like a addict craves another taste of scag, it's an Athlon, which doesn't particularly care about RAM speed anyway.

 

[MaxBurn]

Tanin: Using this argument method is like my saying it's OK to steal a couple of spark plugs out of your (insert favorite fast car here) because you aren't using the power on legal driving anyway. I am not buying into that. You are correct the automatic settings work perfectly, if not up to what it should be. This gives me the feeling they are covering up for something.

P5-133XL: You are probably correct. It used to be easy to assign blame on this sort of issue but now with things split up it's hard to say. I guess now that I think about it it's more likely to be the motherboard as both memory manufacturers I looked at steer you away from four modules in their recommendations.

The memory slots are VERRY close together on this motherboard:
http://www.gigabyte.co.jp/nippon/k8n-pro-sli/photo_k8n_pro_sli_big.jpg
But I don't see how they could have possibly gotten them closer to the CPU.

I take issue with Gigabyte as they completely leave this out of their documentation, but they clearly know what is going on in their BIOS. This type of misinformation from manufacturers I could do without, I would much rather know about this up front even if I have to dig it out of the manual or support section of their website.

Now my thoughts are is there ANY Nforce 4 SLI motherboards that will run all four DIMM slots at DDR400, or is this just something that I "should have known" about the platform in general?

 

[LiamC]

Actually the CPU is an A64 San Diego core 3700+ (single core), sorry I left that bit out. It is fairly new, like within a couple months. Motherboard is also running the F7 (latest) BIOS.

It's a new day now, I put it to CAS3 and DDR400 and am retesting now.

The only way to know this for sure is you post/check the OPN from the heat spreader. Are you sure it's a San Diego? Can you post the OPN?

 

[time]

Have you checked your Command Rate setting? The BIOS normally looks after this, but you are using manual settings. It should be 2T for 4 banks.

Some resources:

http://www.lostcircuits.com/memory/ddr2/4.shtml
http://www.lostcircuits.com/advice/bios2/7b.shtml

Reading the above, the most likely culprit seems to be the RAM. This is not surprising considering it's Corsair's budget range.

Your sparkplug analogy is wildly misplaced. Dropping the memory clock from 200 to 166MHz (probably with faster timings) is not going to have a significant impact on your system. If you doubt this, try to measure a difference in application performance when you change the clock.

 

[MaxBurn]

The only way to know this for sure is you post/check the OPN from the heat spreader. Are you sure it's a San Diego? Can you post the OPN?

I sincerely hope that's what I have as this is what I ordered:

80719-5 AMD Athlon 64 3700+ Processor (San Diego) Socket 939 Retail ***Free 2nd Day***

Is the OPN on the retail box for the CPU? I don't see it... It does detect as a 3700+ and I only see one of those available and it's a San Diego.


I also don't want to speak to soon but I may have this solved. My friend found some references to a "1T / 2T setting" that needs to be changed to 2T for this sort of situation. The trick here is that this setting is in a HIDDEN menu on this motherboard. On the main BIOS screen you have to hit CTRL+F1 which will add an menu selection which has a couple of features plus this one in it. Basically I overode the memory divisor to result in DDR400, kept the timing settigns to AUTO and set that to 2T in the hidden menu.

It just passed memtest86 test number 6 three times (always failed before), I'm going to rerun the full battery for a couple cycles.

 

[Tannin]

No Max, it's like buying a car that can do 6000 RPM and has a tall overdrive 5th gear (which most cars do these days) and then complaining that you can't use 6000 RPM and 5th gear at the same time. If you want a car that can do 6000 RPM in 5th gear, then you need a full-on sports car, not an everyday family sedan.

In any case, as Time has sort of said already, 4 memory modules is problematic at the best of times. This isn't about how much RAM you have, it's about how many modules are fitted at the same time. 4 X 256MB is the same as 4 X 1MB from this perspective. The more individual sticks of RAM you have, the longer the traces, and much more importantly, the greater the inevitable parasitic inductance generated by both the traces and the RAM itself. As you know, the higher the inductance of a circut, the lower the AC frequency it can run at. Indeed, if you put enough L and C in there, you have just reinvented the traditional passive DC rectifier. In short, a lower frequency for more sticks of RAM is plain good sense.

Now, if you want to bend the rules, I'd start with an overclocker's motherboard, such as A-Open, Albatron, maybe MSI (or whatever the currently fashionable brands are, I don't keep track of these things anymore since I got a life) and overclocking-spec RAM. You expect a family sedan to be solid, reasonably priced, rather boring, and very reliable. Gigabyte do that perfectly most of the time. But if you want a sports car, you probably need to take your chances with an A-Open or something of that ilk.

 

[mubs]

My hardware is different, but I do want to point out that I've been running 4x256MB (maxed out) on my Tyan Tiger 100 BX based board for the last 5 years. Flawlessly. The modules are PC-133 and my bus speed is 100 MHz overclocked to 112 MHz (!). Am I just lucky, or were the old BX northbridges able to do this or what?

 

[P5-133XL]

I'm sorry, but historicly most MB's with 4 RAM slots have had problems filling all four. This is not an issue that limits specific chipsets of AMD's. Some seem to have relatively done well at dealing with it, but others have not. The quality of RAM you use can have a massive difference in its ability to capture a signal.

There's a valid reason why servers use registered RAM or buffered RAM with ECC. It is because both of these RAM forms capture signals far better and correct errors to allow for more RAM slots and for these RAM slots to be fully filled. Yes, the RAM costs more and yes there are latency slowdowns but that is what it takes to have those slots fully filled and guarenteed.

At this point, the only real solutions to the problem is to live with the slowdown and keep your RAM; Ues less RAM and only fill two slots and use the remaining on another machine; Replace your RAM with larger value oriented RAM and only fill two slots using your current RAM elsewhere; Replace your current RAM with high quality over-clocking RAM that is better able to capture signals.

 

[Tannin]

You already answered your own question, Mubs: you are running PC-133 RAM on a PC-100 system. This works just fine, and you can even overclock it a bit, but note that your RAM is running well within spec at 112MHz. And, in reality, the BX was a 133MHz chipset. Intel just decided no badge it as a 100MHz chipset for marketing reasons - i.e., in a failed attempt to promote Rambus.

Secondly, the BX and MVP-3 generation of main boards were the last ones at the end of the era where multiple RAM modules wasn't really a problem because the frequencies were just so low that the very small extra inductance of an extra pair of RAM sticks hardly ever mattered.

Notice that the first chipset to suffer huge problems witth 4 sticks arrived only a short while after the BX - the infamous Intel i820. With that one, they had to recall all the motherboards and "fix" them by snipping off the extra RAM slots! More recent designs do it a bit better: they just adjust themselves to the highest RAM speed that is practicable.

 

[mubs]

Thanks, Tannin. 'Cause I tend to build systems only sporadically, there are pretty big holes in my knowledge/experience. I've just been very lucky a lot (knocks on head).

 

[Tannin]

No worries. I still have an MVP-3 overclocked to 112MHz with a K6-III+ 450 running at .. er ... 500 and something. Last of the line. In fact, I might retire it today, I've stopped using it. It's going to be 41C outside today and my little house gets hotter than that, so I'm off to the office for the afternoon where it is cooler. Who knows, I might even get some proper work done.

 

[MaxBurn]

Tannin and P5-133XL: You are making sense to me but I just don't see why this can't be addressed in good hardware design or perhaps the BIOS can relax some timings instead of taking the easy way out and dropping the overall speed. This isn't a server with 4gig or more memory in it. We are however talking about four modules of dual sided memory, probably the worst case situation this board could encounter.

An update: It just passed the full battery of memtest86 and adding in the other three passes of test #6 (where it used to fail) I would say the problem is solved. I will run memtest for a couple more complete loops later on too just to be sure.

mubs: Are your 256MB PC133 chips on that system dual sided?

 

[Tannin]

Oh, my point was that my K6-III and your P-III are directly equivalent, and we have both overclocked them the same amount in the same way. Stands to reason, really.

 

[Tannin]

No worries Max. I'm guessing here, but my take on that would be that the relaxing-timings method (a) is likely to be faster but less stable, where the drop-the-bus-speed method (b) is likely to be more stable at the expense of ultimate performance. I'd expect Gigabyte to go for method (b), as that is their forte (where is Sechs, I need an accented character!) and is what their market expects.

BTW, what is the overclocker's motherboard of choice these days?

Also BTW, where is Sol, he's usually pretty full botttle on this stuff.

 

[P5-133XL]

Just as a side note, the quality of your power supply can directly correlate to the number of RAM errors experianced. Here we are talking about one error in a few months to several errors per day on an otherwise totally stable system. In this case, I'm not accusing the PS. Don't expect MB based voltage regulators to compensate for a poor quality PS. I'm just noting this for the sake of throughness.

 

[mubs]

mubs: Are your 256MB PC133 chips on that system dual sided?

No idea. I cannot look at it now as I'm using the system. I checked my purchases from Crucial, and this is what I have:

Qty & Item        Date Purchased   Price

2 x CT32M64S4D8E    12-09-2000     $157.49 each  (superseded by D7E below)

1 x CT32M64S4D7E    03-07-2001     $89.24 each

3 x CT32M64S4D7E    08-02-2001     $39.09 each

1 x CT32M64S4D7E    09-22-2002     $42.29 each

current price: $66

There are more than 4 sticks because I built two identical systems, and the other sticks went into the second system. Lifting from Crucial's web site:

Module Details:
* Part Number: CT32M64S4D7E
* Module Size: 256MB
* Package: 168-pin DIMM
* Feature: SDRAM, PC133
* Configuration: 32Meg x 64
* Error Checking: Unbuffered, Non-parity
* Speed: 133MHz
* Voltage: 3.3V
* Memory Timings: CL=2

I'm sure somebody here can look at the config (32Meg x 64) and immediately tell you whether it's one-sided or two. I vaguely remember that it is two sided, and wondering if it will all work. That's why I bought a few at a time; if more than 2 didn't work initially, I would have stuck the extra in the second system and not bought any more sticks.

 

[mubs]

BTW, what is the overclocker's motherboard of choice these days?

AFAIk, it's DFI, more specifically the LanParty UT NF4 SLI-DR Expert. That's what I'm planning to buy. Not because I'm a rabid overclocker, but because
a) it has the expansion slot config that most closely matches what I want, and
b) I will be doing a modest amount of overclocking.

There was a time when they were considered a flaky mfr., but I've read that with the Athlons, and especially A64s, their design/manufacture is in the top tier.

 

[LiamC]

The only way to know this for sure is you post/check the OPN from the heat spreader. Are you sure it's a San Diego? Can you post the OPN?

I sincerely hope that's what I have as this is what I ordered:

80719-5 AMD Athlon 64 3700+ Processor (San Diego) Socket 939 Retail ***Free 2nd Day***

Is the OPN on the retail box for the CPU? I don't see it... It does detect as a 3700+ and I only see one of those available and it's a San Diego.

No the OPN is only on the heatspreader, you'd have to take the heatsink off to read it.

3700+ chips were originally FX-51's that were old stock, hence "Clawhammer" core. The SD core is relatively new. You wouldn't be the first one caught out this way. But it all seems to be irrelevant as your problem looks to be solved.

 

[LiamC]

THis might provide more background

http://www.storageforum.net/forum/viewtopic.php?t=5383

 

[LiamC]

And some more

http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=2418&p=3

 

[MaxBurn]

3700+ chips were originally FX-51's that were old stock, hence "Clawhammer" core. The SD core is relatively new. You wouldn't be the first one caught out this way. But it all seems to be irrelevant as your problem looks to be solved.

Well now I am curious, this is what's on the CPU:

AMD Athlon 64
ADA3700DAA5BN
KABQE 0547DPAW
1374429L50028

Google results seem to suggest this is a: 1MB 90nm Rev. E San Diego


Thank you, yes good reading.

 

[CougTek]

As I wrote before the forum corruption occured, it is a Rev. E3 core. The OPN guide for AMD processors is here .

 

[MaxBurn]

Just an update: Using the Value Ram at DDR400 on this board was a no go. It was close enough to pass memtest86 for over nine hours, but corrupted data in windows caused lockups etc. I thought that was a corrupted install from running the memory at the wrong timings etc but it did it immediately again after installing windows and loading drivers.

So for grins and giggles I threw that memory into the computer I am using for a file server. Supermicro P4SCT mobo and a P4 3.0 Northwood with all four DIMM's of that Corsair value ram. It doesn't tell you much about what it's doing in the BIOS at all but CPUz and CPU ID say it's running DDR400 with timings of 2.5-3-3-8 like it should. I don't know if that's 1 or 2 T, I am assuming 2T. It passed two loops of memtest and has been running for a day and a half. In that time I ran cinema craft encoder for over five hours and at the same time had a winrar creation that took about six and a half hours. All is well with that machine as I have come to expect from Supermicro.

So I guess I will be getting new memory for the desktop, I selected Corsair TWINX2048-3200C2 2GB Kit DDR400 XMS3200 as it was recommended by the corsair memory configurator. Also like to point out that they did not recommend the four DIMM's of Value Ram like I had done above, they limit that recommendation to two DIMM's.

 

[Bozo]

It's this kind of agravation and nonsence with AMD motherboards that keeps me from switching to AMD.
I wish AMD would produce their own Quality board that you install and it just works.
Maybe Supermicro will step up to the plate with a Quality board.

Bozo :mrgrn:

 

[MaxBurn]

It's this kind of agravation and nonsence with AMD motherboards that keeps me from switching to AMD.
I wish AMD would produce their own Quality board that you install and it just works. Maybe Supermicro will step up to the plate with a Quality board.

Bozo :mrgrn:

They do, just not something I would want in a desktop computer to game on and stuff.
I don't get why they hide it on a separate site though, looks like OEM only?

http://www.supermicro.com/Aplus/motherboard/

 

[Bozo]

They look like they were designed for a definite purpose. Maybe for SUN??

Besides the price is probably astonomical.

Bozo :mrgrn:

 

[Mercutio]

It's this kind of agravation and nonsence with AMD motherboards that keeps me from switching to AMD.

News update:
Most motherboards - Gigabyte, Intel, Asus, Epox, DFI, whoever - that have four memory banks - have at least some issues with all four banks being full. Checking manufacturer compatibility in that case becomes much more important.
This isn't an AMD issue. It's a "consumer PC Parts are made for consumers" issue.

 

[CougTek]

Besides the price is probably astonomical.

Actually, last time I checked, their prices were lower than Tyan's equivalent. I don't know if that has changed though. I compared prices of boards featuring the NVidia Pro chipset. The boards with the ServerWorks chipset are certainly overpriced and their performance must not be that impressive either.

 

[Bozo]

Never had a memory problem with any Intel brand motherboard. Crucial, Kingston, Corsair all work first time everytime no matter how many slots are used.
If the memory sticks aren't matched, you don't get dual channel mode, but all the RAM is usable without causing problems or having to 'tune' the BIOS.
Even my shitty ASUS MB (Intel 875 chipset) has all 4 slots filled.

Try this: Buy an Intel brand 'consumer' MB (D915GEVL), hit your favorite memory web site and buy 4 sicks of RAM ( I use Crucial, their low priced 'consumer' stuff), attach a SATA hard drive, and finish populating the MB with a CD, floppy or whatever. Put an XP CD in the CD drive, and turn the power on. I'll bet it comes right up and boots from the CD and starts the XP install. No driver to load for the SATA HD, no worrying about what processor core you have; no dumbing down the memory timmings or removing a stick or two; no BIOS updates to get it to see all the RAM or the CPU; no VooDoo and praying to the computer Gods that the damn thing stays up long enough to get the right drivers installed. It just works.

AMD makes a great CPU. Better than Intels. BUT, they need to start making a Quality motherboard to support it.

Bozo :mrgrn:

 

[Mercutio]

The last Intel-brand Intel motherboard I used (granted, that was December 2004) couldn't stand having 4x512MB Crucial PC3200 DIMMs in it. Wouldn't even boot. I think I ended up buying two 1GB sticks for it.

 

[Tannin]

Bozo, you can put 4 sticks of RAM in any mainboard for AMD CPUs I've tried it in in the last ... oh .. quite a few years, and not have to do anything. This entire thread is about getting 4 sticks going at overclocked RAM speeds.