Mushkin PC3500 Level II Dual Pack Testing Troubles

I'm new to this forum and to overclocking the P4C on i865 chipset boards using dual pack modules, so please bear with me on this memory testing question (especially if my question sounds stupid).

I've put together a P4-2.4C system and it's running fine with a pair of 512Mb Mushkin Level II PC3500 modules (bought as a dual pack) on a DFI PS83-BL board (i865 chipset) at standard speed of 200 with the recommended 2-3-2-6 memory settings suggested on the Mushkin website for i865/875 boards. Stable with Prime95 and no errors on the memory when tested both individually and together at 200 and 1:1 using memtest86 v3.0.

Also, the original bios for the DFI has a AGP/PCI/SATA frequency lock that I have set to 66/33/100, which I understand is necessary to overclock the i865/875 chipset boards ... Right?

Okay, now when I move up the FSB from 200 to 216 and leave the memory ratio at 1:1 the PC3500 memory should be at spec or DDR433 ... right?

Okay ... after I do that and run memtest86 on the Mushkin Level II PC3500 modules (again both individually and as a pair), I now get errors ..... random errors .... sometimes one or two and sometime lots of errors, regardless of the timing used (from 2-3-2-6 to 3-4-4-8). I also tried bumping the default voltage from 2.6 to 2.7 with no improvement.

However, when I use the 5:4 or 3:2 ratio to slow down the memory, it all tests fine while the FSB is still above 200 (I once set it as high as 250 for testing purposes).

So my question is am I using the correct technique to check my DDR433 memory and does it sound like it may be bad? Should it not test good using 1:1 at least up to 216 (DDR433)? Any help or suggestions would be much appreciated .... thanks, Rick

Sounds like you've done everything just perfect so far (except the DFI board..., no just joking ) so maybe the RAM is funky. Have you flashed the latest BIOS for this board? Can you try the RAM in someone elses board to see if it works?

use higher voltage, like 2.8v/2.9v to check if it's a memory problem.

Thanks for the info ... I stumbled amd fumbled around some more with the DFI board's settings and so far got one module to run memtest86 without errors for several hours at 1:1 with the CPU frequency set to 217 and memory at 2.5-3-3-7 @ 2.7 volts (i.e. the P4-2.4C is O/C'd at 2.6). Right now the system, with just that single module, has been running Prime95 in Windows XP Pro for 6 hours with no errors (the randon torture test). So far so good, although the memory timings are not as tight as the advertised spec of 2-3-2-6 for that module, but I'll play around with that later.

My next step is to try the same test settings on the other module and after that try it with both modules in dual channel mode. Once I've confirmed the memory is okay, I plan to try running at a higher CPU frequency like 225 or even 250 and back off the memory to 5:4.

Interestingly, the latest DFI bios update no longer includes the option to lock the AGP/PCI/SATA Bus and so I flashed back to the original. I wrote DFI tech support an email about that oversite.

Wish me luck .... thanks!

Well do.
Good Luck man.

eecon,

I'm running the same memory sticks as you are and I'd like to bring up an issue that seems to be affecting my performance. As you can see from my sig I've got 1G of Mushkin's PC3500 lvl2 memory. While I can post and work in WinXP normally @ FSB=246 (3.2GHz) and FSB:MEM=5:4, memtest86 will consistently report errors. Even though timings are tight at 2-2-2-6, the sticks are running at a "mere" 198x2=396 MHz, way below the rated PC3500=433 MHz.

At Mushkin's site there's this page where they report that 1G of RAM has a significantly worse performance at overclocking than 512M, at least on Abit boards. Has anybody else seen this...?

Vdimm has been at 2.8V (Speedfan reports 2.77V) all the time, these sticks are volt-hungry.

I'd appreciate any inputs.

Could not get to your link about the 1Gb vs 5121mb memory issue. Please cut and past the entire link.

Also read the fine print from this Mushkin link about Intel 865/875 chipsets not running at 2-2-2-6 (i.e. use 2-3-2-6).

http://www.mushkin.com/epages/Mushk...80bfcec86273fc0a801050673/Product/View/991098

Thanks eecon, I read that. I'm pretty sure I've also tested with 2-3-2-7 and it failed, but I'll retest tonite just to make sure. By the way, I tested the link and it worked for me... anyway here's the address

http://www.mushkin.com/epages/Mushkin.storefront/Product/View/990879#abitic7max3

I must be blind ... where on that page do I go to find that info?

sorry eecon... I'm just realizing there's a cookie associated with navigating in Mushkin's site... click on the link above, than on DDR SDRAM >> DDR MAINBOARD QUALIFICATION PAGE >> SOCKET 478 / ABIT BOARDS. The first two boards are the MAX3 and the IC7.

What are your GAT settings??

I read the info on the IC7 and the IS7 in the Mainboard Qualification area but still don't see where "they report that 1G of RAM has a significantly worse performance at overclocking than 512M"

How about pasting their statement here?

GAT? Is that my memory timing settings?

I Ultimately found one of my 512 L II twins to fail memtest86 at DDR433 speeds at 2-3-2-6 when I tested them individually .... just packed up the twin 512 LII's for RMAing back to Mushkin.

Sorry about your sticks, eecon.

GAT are the Graphics Accelerator Timings under Advanced Chipset Features in the BIOS. I've read some reports of people who had significant improvements by disabling the last two settings.

Here's Mushkin's statement on the IC7-Max3:

(...) Some other problems surfaced when running 2 x 512 MB DIMMs. In that case, the overclocking capabilities of the board dropped immediately to a high of 260 MHz external CPU clock, regardless of what the CPU / DRAM ratio was set to. Interestingly, in this case, the memory was running only at 177 MHz. The same DIMMs in the same system were running in 1:1 ratio up to 230 MHz bus speed which excludes the memory per se as the culprit but underscores the notion that extreme overclocking and high system memory configurations do not mix and match.

On the IC7:

The ABit IC7 is 100% compatible with all Mushkin PC3200 and PC3500 Black series. Because the chipset works in "dual channel" mode, it is necessary to populate the board with a minimum of two identical DIMMs in the complementary slots in order to use the performance features.
USER TIPS: enabling any of the different performance options will override the manual latency and frequency ratio settings and force a 1:1 CPU:RAM ratio at 2:x:x:x latency settings, with the "Ultra" (now "F1") setting forcing a 2:2:2:5 latency configuration at DDR400 frequency.

We also found that using the "Turbo" (or "Street Racer") setting at 2:2:2:7 yields overall better performance since a longer tRAS prevents the controller from inadvertently terminating an ongoing data burst. Enabling the CPC (Command Per Clock) adds another small notch to the overall performance but can cause stability problems as well. The Read Delay timer should be set to 5-6 cycles for best stability and performance or else left on Auto.

We found that DIMMs that require CAS latencies of 2.5 or higher at DDR400 will not function in any of the available performance modes. The workaround is to set the performance tab to "Disabled". Even though this appears counterintuitive at first, it makes sense in that there is no point in reducing chipset latencies when the memory itself would add those latencies again at the back-end of the "food chain". The HW monitor shows voltage values that are below the real voltages, so better not crank them up too much.

Thanks ... I see now were it was on their website. I'm not using an Abit board and my bios does not have GAT settings (it's an el-cheapo DFI PS83-BL). The latest bios does not even have an AGP/PCI lock for 66/33 ..... I had to flash back to the original March 2003 Bios which had that feature (although now come to think of it, the newer bios without the lock also ran as good as the original bios, so maybe they just locked it by default on the newer bios .... two weeks now and still no response from DFI to my email about that separate issue).

I'm just hoping to get a FSB at 217 with a 1:1 out of this system for a mild overclock to from a 2.4C to a 2.6C with 2-3-2-6 timing and the retail heatsink and thermal pad (although eventually I may use some goo-off and acetone to clean off the original Intel thermal pad and then properly apply some authentic Arctic Silver).

Had no problem running "one stick at a time" that way, but when quad pumping them together at 217 and 1:1, the one bad stick may have been contributing to the occasion instability or the occasional Memtest errors. The system ran and tested fine with both sticks together at 200 and even 209 (including Prime95's torture test). However, I firmly believe that each stick should be good for 217 per Mushkin's specs, so I'm hoping the replacements will both test out separately okay to 217 and also when working together at 217.

At that point I'll be happy to live with a 2.4C safely running at 2.6 for the next year. Then I'll just buy a replacement 3.2C when the price is right as my board will indeed support a stock 3.2C.

We shall see.