Overclock.net › Forums › AMD › AMD CPUs › Best way to test CPU and RAM stability?
New Posts  All Forums:Forum Nav:

Best way to test CPU and RAM stability?

post #1 of 8
Thread Starter 
Dual-Core CPU and two RAM sticks running in Dual-Channel mode. I underclocked my RAM and ran the Prime95 Small FFTs test with a single instance of Prime95 and 2 threads and determined the top stable speed with a 27+ hours run. I also tested the RAM for 12 hours with Memtest86 at it's specified configuration and with the CPU at it's stock configuration. Then I set up the CPU with the configuration I had determined was stable, and used the Memclock to CPU Limiter to avoid overclocking the RAM. The resulting, final RAM clock was about 1.5% below it's specified, (and tested), stable configuration.
However, my system freezes, every time. After about 1 hour.
I found that by increasing the voltage to my CPU, or by decreasing the CPU overclock, I could avoid the freezing, or at least it seems that way. So it seems that the Prime95 Small FFTs test was not sufficient to determine the CPU overclock. Perhaps I need to be testing the memory controller as well. Perhaps the memory controller requires additional voltage, and that explains the problem.
So what's the best way to determine the actual maximum stable overclock? Should I run two instances of Prime95 with their respective affinities set to each CPU core, and have one do a single threaded Blend test while the other does a single threaded Small FFTs test? Two single threaded Blend tests? What is best?
post #2 of 8
Blend tests everything cpu n ram related so use that. Set your oc on the cpu at the voltage you got stable and the ram and run blend. Any errors up the cpu voltage a tad

Just a side note your ht or nb arnt oc arw they
post #3 of 8
Thread Starter 
Quote:
Originally Posted by Caribou007 View Post

Dual-Core CPU and two RAM sticks running in Dual-Channel mode. I underclocked my RAM and ran the Prime95 Small FFTs test with a single instance of Prime95 and 2 threads and determined the top stable speed with a 27+ hours run. I also tested the RAM for 12 hours with Memtest86 at it's specified configuration and with the CPU at it's stock configuration. Then I set up the CPU with the configuration I had determined was stable, and used the Memclock to CPU Limiter to avoid overclocking the RAM. The resulting, final RAM clock was about 1.5% below it's specified, (and tested), stable configuration.
However, my system freezes, every time. After about 1 hour.
I found that by increasing the voltage to my CPU, or by decreasing the CPU overclock, I could avoid the freezing, or at least it seems that way. So it seems that the Prime95 Small FFTs test was not sufficient to determine the CPU overclock. Perhaps I need to be testing the memory controller as well. Perhaps the memory controller requires additional voltage, and that explains the problem.
So what's the best way to determine the actual maximum stable overclock? Should I run two instances of Prime95 with their respective affinities set to each CPU core, and have one do a single threaded Blend test while the other does a single threaded Small FFTs test? Two single threaded Blend tests? What is best?

Quote:
Originally Posted by richie_2010 View Post

Blend tests everything cpu n ram related so use that. Set your oc on the cpu at the voltage you got stable and the ram and run blend. Any errors up the cpu voltage a tad

Just a side note your ht or nb arnt oc arw they

My HT is, in fact, underclocked. I read that there's no practical speed benefit from running it at 4x instead of 3x, so I have it at 3x to consume less power and produce less heat. I did test it at 4x as part of my troubleshooting, however I didn't notice any benefit to stability. Is North Bridge just another name for Hypertransport, or is that something else I need to look into?

I would like clarification on what tests to run. You said run the Blend test. Run the Blend test on two threads, or run two Blend tests on one thread each with two instances of Prime with their affinities set to one CPU each, or run one Blend test and one Small FFTs test on one thread each with two instances of Prime with their affinities set to one CPU each? Stability testing is time and power consuming and deteriorates the system components. I don't want to have to do this again. I need a definitive test.
post #4 of 8
From my experience, IntelBurnTest showed the need for more voltage going to my uncore/C.P.U.-N.B., but Prime95 Blend might have been been adequate for my R.A.M. as I ran ganged until an error occurred about an hour later, so I ran I unganged, and unganged had no errors until I stopped it after nine hours.

Linpack (IntelBurnTest, O.C.C.T, or etc.) might produce more heat than Prime95, so I bet that Linpack stresses the C.P.U. more than Prime95.

Also, using less than all available R.A.M. might help because reports via other users were suggesting it, but I cannot remember how much for how much would be good.
post #5 of 8
Thread Starter 
Quote:
Originally Posted by L4dd View Post

From my experience, IntelBurnTest showed the need for more voltage going to my uncore/C.P.U.-N.B., but Prime95 Blend might have been been adequate for my R.A.M. as I ran ganged until an error occurred about an hour later, so I ran I unganged, and unganged had no errors until I stopped it after nine hours.

Linpack (IntelBurnTest, O.C.C.T, or etc.) might produce more heat than Prime95, so I bet that Linpack stresses the C.P.U. more than Prime95.

Also, using less than all available R.A.M. might help because reports via other users were suggesting it, but I cannot remember how much for how much would be good.

Just to make sure I understand the lingo you used. Ganged is achieved by running one instance of Prime95 with 2 threads, and unganged by running 2 instances with one thread each with their affinities set to one core each?
post #6 of 8
Quote:
Originally Posted by Caribou007 View Post

Just to make sure I understand the lingo you used. Ganged is achieved by running one instance of Prime95 with 2 threads, and unganged by running 2 instances with one thread each with their affinities set to one core each?
Ganged could use two or more threads but performance might decrease, so I suppose that Ganged might need but one thread for stability testing purposes, but I was not thinking that initially...

However, testing with two threads on Ganged mode might be necessary too; I use Unganged because my applications are usually multi-threaded.
post #7 of 8
Thread Starter 
Quote:
Originally Posted by L4dd View Post

Ganged could use two or more threads but performance might decrease, so I suppose that Ganged might need but one thread for stability testing purposes, but I was not thinking that initially...

However, testing with two threads on Ganged mode might be necessary too; I use Unganged because my applications are usually multi-threaded.

I misunderstood what ganged-unganged means. I don't think my motherboard supports this feature. I am using two memory sticks in dual-channel mode, but it seems that this has more to do with how the CPU is allowed to utilize them. Anyway, there is no mention of ganged-unganged modes in my BIOS.

I ran IntelBurnTest at Maximum Stress Level, and it successfully completed a 20 repetition run with no errors in 1h40m. I guess I need a different test.

What would you recommend?

I'm going on two clues. One is that I tried using HCI Memtest before and my system failed. Then I bumped up my CPU voltage and it ran without errors for close to four hours. So that HCI Memtest might be able to detect my system fault. However, I just don't know how much I should expect to run it for me to be able to say I've found a stable configuration. The second clue is that I lessened my overclock by 7.2%, (and kept the voltage the same), and didn't experience any freezes after using my computer for the rest of the day. It was much more stable, but just because it didn't freeze for the rest of the day doesn't mean disk drive data corruption and other errors aren't happening beneath the surface. So I need a definitive test.
Edited by Caribou007 - 7/11/13 at 8:40pm
post #8 of 8
Quote:
Originally Posted by Caribou007 View Post

I misunderstood what ganged-unganged means. I don't think my motherboard supports this feature. I am using two memory sticks in dual-channel mode, but it seems that this has more to do with how the CPU is allowed to utilize them. Anyway, there is no mention of ganged-unganged modes in my BIOS.

I ran IntelBurnTest at Maximum Stress Level, and it successfully completed a 20 repetition run with no errors in 1h40m. I guess I need a different test.

What would you recommend?

I'm going on two clues. One is that I tried using HCI Memtest before and my system failed. Then I bumped up my CPU voltage and it ran without errors for close to four hours. So that HCI Memtest might be able to detect my system fault. However, I just don't know how much I should expect to run it for me to be able to say I've found a stable configuration. The second clue is that I lessened my overclock by 7.2%, (and kept the voltage the same), and didn't experience any freezes after using my computer for the rest of the day. It was much more stable, but just because it didn't freeze for the rest of the day doesn't mean disk drive data corruption and other errors aren't happening beneath the surface. So I need a definitive test.
Your R.A.M. is stable with Prime95 Blend?
Test length depends on the usage. How long would you use your overclocked P.C. doing what?

Yeah, the C.P.U. and motherboard could be limiting your R.A.M. capabilities if your motherboard is AM2 and/or if your C.P.U. is not a Phenom+. What are your C.P.U. and motherboard?
Edited by L4dd - 7/12/13 at 12:10am
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: AMD CPUs
Overclock.net › Forums › AMD › AMD CPUs › Best way to test CPU and RAM stability?