Pedantic
03-17-08, 09:02 AM
I'd like to share with you my recent experience with "burning-in" my RAM, and the resultant failure of doing so.
I had finally achieved what would appear to be a stable overclock, as shown here. (http://img120.imageshack.us/img120/865/finallycd5.jpg)
I ran two tests on across all four cores: one blend and one in-place FFT, as well as a small FFT test on each individual core.
The RAM was slightly overclocked to 815mhz DDR2, at the time of that screenshot. The divider is 6:7; stock voltage (2.1) and basic timings (4-4-4-12) were set in the manually in the BIOS.
Never being satisfied, I felt the need to push the system further. Since I knew the CPU was pushed about as far as it would go (in my motherboard) within reasonable voltage increases, I decided to have a go at the memory.
I wanted to see for myself if running highly-clocked RAM asynchronous with the CPU's FSB would hinder or augment performance, through various benchmarking. At the same time, I would burn-in my RAM. I was shooting for a 3:4 divider, which I actually achieved.
The method was as follows:
1. Raise the memory voltage by .05v
2. Increase the clock by 5-20mhz DDR2 until the system would fail to boot (into Memtest86)
3. Slightly lower the clock until it would pass a standard test once
4. Repeat 1-3, unless step 1 fails to be effective (go to step 5)
5. Raise the northbridge voltage by .05v, repeat 1-4
The final settings I used were a 925mhz memory clock (3938 Mb/s) at 2.35v to the DIMMs, and 1.45v to the actively-cooled northbridge (for a total increase of only .05v). Everything else was left as it was - including timings - when my system was stable.
I left Memtest86 running at these settings for about nine hours, or fourteen passes of the standard test.
By means of a few buttresses, I pointed a small house fan on my RAM to keep it cool during the testing. The DIMMs' heatspreaders would get warm, but never hot. I checked this periodically every few hours.
After benchmarking, I realized the increase in RAM clock did little to affect performance. In fact, it slightly hindered some more than it helped any. This is true to what I've read.
However, after returning all BIOS settings to the original stable configuration, the system is now unstable. The Prime95 blend test will fail after anywhere from ten minutes to ten hours. The in-place FFT test will fail much quicker than that.
I subsequently lowered the CPU FSB very slightly. Also, I returned the RAM to its original clock of 800mhz DDR2, currently at a divider of 13:15. This seems to have allowed Prime95 blend to run a bit longer, but still eventually fail. It seems to have had no effect on the in-place FFT test.
This leads me to conclude a few (mostly uneducated) theories. One, would be the sustained increase of the northbridge voltage actually damaged something on the motherboard. Or, the sustained increased voltage actually damaged the RAM, which could have been unsuitable to burn-in in the first place. All of my components were recently bought within the past two months.
I am more inclined to believe it is a problem of the northbridge. The reasoning being because of the resulting overall system instability. Though, since I have not re-tested the RAM, it could be the issue as well. I will conduct more tests in the near-future to get to the root of the problem.
I had finally achieved what would appear to be a stable overclock, as shown here. (http://img120.imageshack.us/img120/865/finallycd5.jpg)
I ran two tests on across all four cores: one blend and one in-place FFT, as well as a small FFT test on each individual core.
The RAM was slightly overclocked to 815mhz DDR2, at the time of that screenshot. The divider is 6:7; stock voltage (2.1) and basic timings (4-4-4-12) were set in the manually in the BIOS.
Never being satisfied, I felt the need to push the system further. Since I knew the CPU was pushed about as far as it would go (in my motherboard) within reasonable voltage increases, I decided to have a go at the memory.
I wanted to see for myself if running highly-clocked RAM asynchronous with the CPU's FSB would hinder or augment performance, through various benchmarking. At the same time, I would burn-in my RAM. I was shooting for a 3:4 divider, which I actually achieved.
The method was as follows:
1. Raise the memory voltage by .05v
2. Increase the clock by 5-20mhz DDR2 until the system would fail to boot (into Memtest86)
3. Slightly lower the clock until it would pass a standard test once
4. Repeat 1-3, unless step 1 fails to be effective (go to step 5)
5. Raise the northbridge voltage by .05v, repeat 1-4
The final settings I used were a 925mhz memory clock (3938 Mb/s) at 2.35v to the DIMMs, and 1.45v to the actively-cooled northbridge (for a total increase of only .05v). Everything else was left as it was - including timings - when my system was stable.
I left Memtest86 running at these settings for about nine hours, or fourteen passes of the standard test.
By means of a few buttresses, I pointed a small house fan on my RAM to keep it cool during the testing. The DIMMs' heatspreaders would get warm, but never hot. I checked this periodically every few hours.
After benchmarking, I realized the increase in RAM clock did little to affect performance. In fact, it slightly hindered some more than it helped any. This is true to what I've read.
However, after returning all BIOS settings to the original stable configuration, the system is now unstable. The Prime95 blend test will fail after anywhere from ten minutes to ten hours. The in-place FFT test will fail much quicker than that.
I subsequently lowered the CPU FSB very slightly. Also, I returned the RAM to its original clock of 800mhz DDR2, currently at a divider of 13:15. This seems to have allowed Prime95 blend to run a bit longer, but still eventually fail. It seems to have had no effect on the in-place FFT test.
This leads me to conclude a few (mostly uneducated) theories. One, would be the sustained increase of the northbridge voltage actually damaged something on the motherboard. Or, the sustained increased voltage actually damaged the RAM, which could have been unsuitable to burn-in in the first place. All of my components were recently bought within the past two months.
I am more inclined to believe it is a problem of the northbridge. The reasoning being because of the resulting overall system instability. Though, since I have not re-tested the RAM, it could be the issue as well. I will conduct more tests in the near-future to get to the root of the problem.