DDR400 memory running at DDR333 speed

hartzhartz
edited January 2006 in Hardware
Hi people.

My question is "Why is my PC3200 memory suddenly running at DDR333 speed?"

I do not overclock or play arround with memory timing settings. I have re-confirmed in my BIOS, and everything is set to Auto. I have removed the Dimms and checked them - it is indeed PC3200 modules.

And I am convinced that I checked this just after I bought my new PC (About 8 months old) and the memory was then running at DDR400 speed.

:nudge:

I am in the process of testing the memory with memtest86. When all four modules are installed and I change the memory clock speed in the BIOS from AUTO to 200MHz, I start getting errors in memtest86 (The test is clean when run at DDR333 / AUTO speed settings). The errors are always on the same bit and it is always test #5 which fails, and the errors always occurs at an address below 1 MB. I have not completed all the tests yet, but it seems that with any two modules (out of the four) installed, I do not get any errors.

Right now I am running the OS with 2 of the memory modules installed and the memory clock speed set to 200 MHz, and the system is stable (And just a few minutes ago had passed 5 full itterations of memtest86 with zero errors).

If anybody wants to see digital camera photos of the BIOS pages or memtest86 screen showing the errors, let me know please - I can upload them somewhere or email them to you.

Now I am going to re-install all 4 modules and start the memtest86 again. Could this just have been a seating problem?

For the record, here is my system specifications:
Memory: 4 x 512 MB Apacer UNB PC3200 CL2.5
CPU: AMD Athlon 64 3000+
Motherboard: Gigabyte GA-K8NXP-SLI (May be due a firmware upgrade)
Graphics: Gigabyte GV-3D1 (dual 6600GT)

I doubt that any of the drives or connected peripherals are relevant to this problem.

Any recommendations!? Thoughts? Ideas?

Thanx,
_Hartz

Comments

  • LeonardoLeonardo Wake up and smell the glaciers Eagle River, Alaska Icrontian
    edited January 2006
    It could be that one of the memory modules has a defect/has gone bad and that your motherboard is defaulting to the performance level of the least capable module.
  • csimoncsimon Acadiana Icrontian
    edited January 2006
    Have you cleared the cmos and set the bios to default?
  • hartzhartz
    edited January 2006
    Leonardo wrote:
    It could be that one of the memory modules has a defect/has gone bad and that your motherboard is defaulting to the performance level of the least capable module.

    This is a thought that occured to me as well. However, considder this:

    The way I did my testing is as follow:

    First I just checked the BIOS settings to confirm that everything is set to AUTO as it should be, and ran memtest86 which came up clean but reported that the memory was running at DDR333 speed (approximately).

    Then I forced the Memory clock setting to 200 MHz and re-ran memtest86. The memory speed was as expected (DDR400 or close enough - why can't it be exact in any case), but errors were reported on Addresses below 1 MB for Test #5.

    The third test was to set the Memory clock back to AUTO and to set the "Enable 2T" option to "disabled". This had the result of running the memory at DDR200 speed and the memory timings was reported as being 2-2-2-5 in this instance. The test did not find any errors after several hours.

    The fourth test was to remove two dimms and run with all BIOS settings set to AUTO. memtest86 reported that the memory was running at DDR333 speed, and there were no errors.

    The fifth test was to run these two dimms at DDR400 speed by forcing the memory clock to 200 MHz in the BIOS. This test was clean, so I thought I could eliminate two of the dimms!!!

    The sixth test was to remove these two dimms and run the OTHER two dimms at DDR400 speed (Forced to 200 MHz in the BIOS). The test was also clean, which basically voided the previous test's conclusion of eliminateing 2 of the dimms.

    At this point I was realy stumped, so I put back the other two dimms, and re-ran the test at Auto (test nr 7) and at 200 MHz / DDR400 speed (test #8). With Auto settings the memory ran at DDR333 speed and the test came out clean. At DDR400 speed, the expected errors were repotred, agains failing only during test #5, as usual on various addresses in the 1st MB of RAM.

    I am currently running a 9th test: All I did was to remove all the Dimms and re-insert them in the OPPOSITE ORDER!!! I wanted to see whether the address where the errors occur changed at all. So far, after 2 full passes of the memtest86 standard test, there have been no errors !!! This is running at a forced 200 MHz clock speed.

    I will post an update later when I've had about 10 full passes, but the fact that I don't get any errors when running with fewer than 4 dimms is seriously confusing me. This means I cannot properly test each dimm seperately.

    csimon wrote:
    Have you cleared the cmos and set the bios to default?

    Not purposely. I did do this a while back. Would the hold-F7-during-power-up method be good enough, or do I have to use the BIOS reset jumper (A real schlepp as it is in a hard to reach spot below the video card)

    My actions from here:
    a) assuming this test comes out clean, I will just restore the dimms to the previous order and re-run a test to see if any errors return.
    b) I will clear the CMOS, reset defaults, and test.
    c) I will update the BIOS to the latest revision, and test.

    If the memory continues to step down to DDR333 speed with AUTO settings, I am going to return it to the shop for a replacement.

    Again any thoughts or recommendations welcome. And I've been taking pictures of the whole process (helps me to remember what I've seen or to re-check details) of results later, so let me know if you would like any more details.

    Thanx!
    _H
  • lemonlimelemonlime Canada Member
    edited January 2006
    The reason is actually very simple.. On the 754/939 Platform, using 3 or more dimms causes the memory controller to default to DDR333 speeds, and 2T timing. It does this to reduce the stress to the integrated memory controller. Some processors can run them just fine at DDR400, however many can not. If your processor is a 3000+, I'm guessing that it may be a Winchester based processor, in which this theory would jive. Only the 'E' revision (aka Venice/San Diego) A64 processors allow 4xDIMMs at DDR400. If you increased it, and get errors in Memtest86+ or instability in windows, it is likely due to your on-die memory controller and not the memory itself. My guess is that you'd be able to run memtest successfully at DDR400 on any combination of two sticks in your system.

    You may be able to get your system stable at DDR400, however you'll likely need to loosen up tRCD, TRP and leave CPC disabled (2T).
  • TheLostSwedeTheLostSwede Trondheim, Norway Icrontian
    edited January 2006
    Lemon is right. All A64 prior to Venice only supports DDR333 with 4 sticks/8 banks of ram installed and you have to use CPC Off/2T command rate.

    However, 2GB of ram at DDR333 is a lot more useful than 1GB at DDR400.
  • Sledgehammer70Sledgehammer70 California Icrontian
    edited January 2006
    damn I want to answer that :)

    B ut yes Lemonlime is right! I heard AMD fixed this in the M2 socket set mobo's due to come out this quarter!
  • danball1976danball1976 Wichita Falls, TX
    edited January 2006
    MackanzOCZ wrote:
    Lemon is right. All A64 prior to Venice only supports DDR333 with 4 sticks/8 banks of ram installed and you have to use CPC Off/2T command rate.

    However, 2GB of ram at DDR333 is a lot more useful than 1GB at DDR400.

    So this doesn't affect the ASUS A8N-E/AMD Athlon 64 3200+ Venice?
  • hartzhartz
    edited January 2006
    Thank you, it makes a lot more sense now. I suspect that even with only two dimms installed, the system defaults to DDR333 speed, but I need to go re-test that.

    I have tested 4 dimms at DDR400 with timings of 2.5-5-5-9 and it appears stable. If I decrease any one of the items, I get a small number of errors (1 error every 10 passes).

    I must conclude that I must have been mistaken about previously checking that the memory was running at DDR400 speed.

    However I have one question: Could someone please clarify the term "stress" as used by lemonline in that context.
  • madmatmadmat
    edited January 2006
    Stress, I.E. the amount of work-load placed upon a circuit.
  • hartzhartz
    edited January 2006
    Thanx madmat, but it is still a little too fuzzy for my likings.

    I can only assume you mean that with 4 DIMMs, the memory controller draws more power, and subsequently outputs more heat. I'm not at all comfortable with the concept of a processor (or memory controller) doing usefull "work" in the joule sense (Eg all joules of work goes into wasted heat, eg it isn't usefull, unless you are running a processor to heat up your room)

    But that is another long discussion. My next question is "Does the memory controller actually drive the current on the bus to the memory dimms?" Doesn't the CPU have dedicated driver cuircuitry for bus IO? If it does, is this what is inferred above? If so, does the (extra) DIMMs draw their power from the bus? That would seem stupid to me (And complicate matters when it comes to controlling memmory supply voltages, something the infamous overclockers are so interested in)

    If this is not a power issue, I must assume that the memory controller activates additional circuitry to manage address related work. Or again, if this is a driving issue, and since the data bus throughput doesn't change when going from 2 to 4 dimms, then maybe there is some kind of a change on the address bus load?

    All in all, it doesn't make sense to me that there should be more electric current driven through the memory controller itself when having more physical dimms installed - the extra power to drive the extra dimms should come from the dimm's dedicated power / driving cuircuitry, not via any bus!

    disclaimer: I am no expert on this topic, just realy curious!
  • lemonlimelemonlime Canada Member
    edited January 2006
    hartz wrote:
    Thanx madmat, but it is still a little too fuzzy for my likings.

    I can only assume you mean that with 4 DIMMs, the memory controller draws more power, and subsequently outputs more heat. I'm not at all comfortable with the concept of a processor (or memory controller) doing usefull "work" in the joule sense (Eg all joules of work goes into wasted heat, eg it isn't usefull, unless you are running a processor to heat up your room)

    But that is another long discussion. My next question is "Does the memory controller actually drive the current on the bus to the memory dimms?" Doesn't the CPU have dedicated driver cuircuitry for bus IO? If it does, is this what is inferred above? If so, does the (extra) DIMMs draw their power from the bus? That would seem stupid to me (And complicate matters when it comes to controlling memmory supply voltages, something the infamous overclockers are so interested in)

    If this is not a power issue, I must assume that the memory controller activates additional circuitry to manage address related work. Or again, if this is a driving issue, and since the data bus throughput doesn't change when going from 2 to 4 dimms, then maybe there is some kind of a change on the address bus load?

    All in all, it doesn't make sense to me that there should be more electric current driven through the memory controller itself when having more physical dimms installed - the extra power to drive the extra dimms should come from the dimm's dedicated power / driving cuircuitry, not via any bus!

    disclaimer: I am no expert on this topic, just realy curious!

    I must warn ahead of time that I'm not much of an expert in this topic either, but I'm happy to share what I think I know :)

    I don't believe the issue is a heat or electrical limitation. Extreme cooling does not improve the IMC's ability to run out of spec, but in most cases makes it worse. The DIMMs receive their power via mainboard regulation, and the 2.5+ volts that the dimms use are drawn from the PSU, not the memory controller. The memory controller does 'drive' the dimms from an IO perspective, and varying degrees of 'drive strength' can be defined. Increasing drive strength has proven to increase stability when using 4xDIMMs. I recall someone from OCZ getting 4x512MB dimms to operate at DDR400 and 1T timing using very high drive strength values of 15+ on the DFI Expert boards. Using sub-zero cooling sometimes requires 'weaker' drive strength to combat overdriving of the DIMMs (The silicon simply conducts better when cold, causing high efficiency driving)

    As far as a technical explaination for the IMC's limitations, I'm afraid I'm not much of an expert in this subject. It may be a question of drive strength, but I am not sure. In more layman terms, I know that it was simply not designed to support four (double sided) DIMMs at full speed DDR400 operation. Forcing your A64 to run at DDR400 is essentially forcing it to run out of spec. Improvements in the 'E' revision A64 processors support DDR400 at 2T timing, and remove the need to use the 5/6 divider. Getting 4xDIMMs to run at 1T however (out of spec) is another challenge all together.

    I'm hoping Mackanz has some more info to share, because you've sparked my curiousity as well :D
Sign In or Register to comment.