System fails RAM stress test but CPU stress test OK.

Oct 30, 2004
11,442
32
91

I'm working on overclocking a Socket 939 Athlon X2 4400 (2.2 Ghz stock, 1 MB cache each core) and when I run the Stress Prime Orthos test to stress the CPU itself it's fine, but when I do it to stress the RAM I get an error.

Now, I previously had a single core Opteron 148 in this computer and it seemed pretty stable at the same RAM settings and at a much higher overclock than the 4400. The computer has 4 x 512 RAM in it.

Is it possible that the RAM is now undervolted because the 4400 is sucking up more voltage? The voltage readings on my computer show it as being about 0.05 V lower than where I have it set in the BIOS.

Should I up the RAM voltage? It's currently set at 2.8 V. Should I start playing around with the RAM timings? Currently at 2.5-3-3-7 and 2T with a divider of 133/200. (Current CPU overclock is 11 x 225.)
 

BonzaiDuck

Lifer
Jun 30, 2004
16,054
1,683
126
You've done something that many of us haven't had (or taken) the opportunity to do: go from a single core to a dual core with the same mobo and memory.

I'm not "AMD-savvy" although I recognize the fact that the memory controller is built-in to the processor with AMD systems. I can only intuitively guess that the change in processor has something to do with your observations here.

Without entangling myself in the AMD over-clocking details, I would start all over with default settings for both the RAM and the (new X2) processor, and test the memory under those circumstances. Then, go through your over-clocking regimen of finding RAM and processor "limits," and tweaking your latencies. Try and unload yourself of assumptions, recollections and settings that were associated with the processor being replaced.

If you have been careful not to exceed the warranty voltage on your RAM modules, I would continue that practice.

I can tell you that moving from dual-core to quad-core INtel created an entirely different regime of acceptable RAM timings and speeds -- something I really noticed. The FSB thresholds for different sets of timings, dropped noticeably from an E6600 to replacement with a Q6600 processor. So where I might have used 3,4,4,8 while running the RAM (1:1) with the E6600 at [DDR] 740, I had to loosen the timings to 4,3,4,9 just to run them at 720 with the Q6600.
 
Oct 30, 2004
11,442
32
91

Thanks for posting your observations, Bonzai.

Maybe I should try it with relaxed timings. Runs fine on Stress Prime Orthos at 11 x 200 and 11 x 210 for 10 minutes with the same settings. Maybe the issue is the Memory Controllers on the new (to me) CPU?

It seems like the new (to me) dual core CPU needs more voltage in the BIOS to maintain a given voltage on my mobo stats/temp/voltage monitor than the single core Opteron did. Could it be possible that the RAM is getting undervolted now and that even though I have it set for 2.8 in the BIOS that it's lower in reality?

What about motherboard chipset voltage (currently at 1.6 V)? Could anything bad happen by trying the RAM voltage at 2.9?
 

BonzaiDuck

Lifer
Jun 30, 2004
16,054
1,683
126
That depends on the manufacture of the RAM modules and their "recommended warranty maximum." If these are DDR(1) modules, I've known some warranty-limited to 2.9. If they're OCZ's, their "EVP" warranty protection extends above 3.0 or 3.1.

I've got a set of Platinum EL DDR-400's, which were rated to either 2.8 or 2.85. I only push them to 2.75V, enabling me to run the stock latencies of 2,3,2,5 @ 452 Mhz.

After I've experimented with OC'ing a system to a reasonable maximum, I eventually reduce the over-clock to a point where the spec'd processor maximum voltage is only exceeded by a couple-hundredths of a volt. The setup with the Platinums was perfectly consistent with that scenario.

Some of us have been using Crucial Ballistix DDR2-1000's, rated to 2.2V. We've set these at 2.2, but more people than I report that they "burned out the light-bulb" after a while, even if the modules were set at the warranty maximum. And in fact, on any given motherboard, the "set" voltage differs from the "monitored, reported" voltage, and both observations often differ from the "measured" voltage taken with a multi-tester.

It pays to be careful.
 

BonzaiDuck

Lifer
Jun 30, 2004
16,054
1,683
126
As to how a change of equipment (like your CPU) might cause a different "real" voltage on the RAM modules, I've often speculated about that possibility. I might suggest that getting "monitored" readings from the BIOS under both processors might reveal something there.

My best sense of it is that changing any piece of equipment would be cause to re-evaluate over-clock and latency settings. At least when swapping a bad motherboard for the same make and model, you might find that stable over-clock settings for the old one work with the new one. But swapping processors? I'd say you need to do some testing like you did with the older processor before you certified "stable settings."

Same for adding memory. Using the same make and model memory, if you go from two slots filled to four, it's a cause to retest and recertify -- possibly at different settings.
 
Oct 30, 2004
11,442
32
91

Oh, I know all about the headaches associated with going from 2 x 512 to 4 x 512.

Anyway, could anyone recommend a good program for monitoring CPU temps? I have one that came with my motherboard but it only reports the overall temps (or maybe just the temps from <gasp> one core). I'd love a program that would tell me the measured voltages on each core and the RAM, etc.

Mobo is an Epox +9NPA- Ultra nForce 4
 
Oct 30, 2004
11,442
32
91

I just installed Coretemp. Seems to give reading that are a couple degrees lower than what my motherboard reports. Not sure what Speedfan does, but I don't think overheating is really a problem and I have a second fan on the back of my HSF, an Arctic Freezer 64 Pro, that I ziptied down to it.
 

stevty2889

Diamond Member
Dec 13, 2003
7,036
8
81
Originally posted by: WhipperSnapper

I just installed Coretemp. Seems to give reading that are a couple degrees lower than what my motherboard reports. Not sure what Speedfan does, but I don't think overheating is really a problem and I have a second fan on the back of my HSF, an Arctic Freezer 64 Pro, that I ziptied down to it.

Go with speedfan instead, Coretemp doesn't work properly with X2's.