Stability Issues

[REDtoLINE]

Alright here's the deal, my overclocking experience has a little bit of a story to it, but if you can bear with me, I think it will help me achieve a much higher overclock than I originally intended. I have the following hardware:

Intel Core i7 920 C0
ASUS P6T6 WS Revolution
OCZ DDR3-1600 7-7-7-24 OCZ3P1600LV2G (3x2GB=6GB)
Corsair HX750W
Xigmatek Thor's Hammer (Push/Pull with 2x120mm 90CFM 2000RPM Fans)

I've been having a lot of problems with getting a stable overclock and I was hoping someone here could help me shed some light on my issues. Originally I was trying to get my CPU to 3.8 GHz, but I've been hitting all kinds of walls. First I couldn't even POST at 3.8 GHz without overvolting to 1.4v-1.42v which is well beyond my comfort zone. Secondly, my motherboard refuses to POST with a BCLK of 200 or higher, which is quite strange for a board of this quality, and sucks because a BCLK of 200 can give me a perfect 3.8 GHz using a 19x multiplier and a perfect 1600 MHz on my RAM using a 10x multiplier. Third, my system was giving me all kinds of problems when I tried to take my RAM to 1600 MHz, which is strange because that's what it's rated for. All said and done, I said screw it and decided to put everything back on AUTO and give up on overclocking. Well originally, I had thought my stock voltage was 1.25v, and I couldn't get 3.36 GHz relatively stable without overvolting to 1.25625v, which I figured wasn't much of an overvolt considering the performance increase I was getting. Well once I put everything back to stock settings, I noticed that the stock voltage wasn't 1.25v, but rather 1.225v and this confused me considering many people can get to 3.4 GHz without overvolting and I had to overvolt quite a considerable amount from stock just to get it stable for a few hours. So I said what the hell and tried 3.2 GHz using a BCLK of 160 MHz, and the stock voltage of 1.225v. Well after starting up a Prime95 small FFT torture test before taking a nap, I woke up 4.5 hours later surprised to find everything still running stable. I was happy to discover that I was severely overvolting the entire time, so I decided to run a Prime95 blend test just to check that the system was entirely stable. Well, it failed and I was puzzled as to why because I assumed it was my RAM causing the problem. My BCLK is currently 160 MHz, my RAM multiplier is 6x, putting the RAM at 960 MHz which is nothing since it's rated at 1600 MHz, the RAM voltage is 1.5v, and the QPI voltage is 1.2v all of which are stock voltages. JEDEC recommends timings of 6-6-6-14 for 914 MHz at 1.5v, and my RAM AUTO timings set my RAM to 6-6-6-18 at it's current 1.5v, which should be perfectly stable at 960 MHz. Well I've been on the OCZ Support Forums trying to figure out the issue, thinking there is something wrong with my RAM and trying all different kinds of settings. I ended up running MCI Memtest on the RAM, and after a few hours I came back and there were no errors, which makes much more sense since this RAM isn't being stressed at all and is running well within the JEDEC specifications. But now this raises the question, what is causing Prime95 large FFT tests to BSOD after a few minutes, and causing Prime95 blend tests to either report a rounding error after 5-30 minutes and me physically stopping the test, or BSOD after I let it run for a few hours? I don't feel it's my CPU or RAM, since these specific tests are showing that they are stable, so I want to point the finger at the motherboard, but I have no evidence.

I would greatly appreciate it if someone could help me with this issue, and I'm sorry for rambling but I find that I miss certain details if I don't tell everything as a short story. For reference, here are all the BIOS settings reported by ASUS TurboV:

BCLK = 160 MHz
CPU Voltage = 1.225v
DRAM Bus Voltage = 1.5v
QPI/DRAM Core Voltage = 1.2v

CPU PLL = 1.8v
ICH PCIE = 1.5v
IOH PCIE = 1.5v
ICH = 1.1v
IOH = 1.1v

All DRAM Reference Voltages for CHA, CHB, and CHC are 0.5x

Also worth noting, my CPU Multiplier is 20x, my RAM multiplier is 6x, and my Uncore and QPI multipliers are on AUTO. C1E is disabled, and so is SpeedStep. CPU-Z reports my QPI Link around 2880 MHz, my NB frequency around 2560 MHz, and my RAM timings as 6-6-6-18-44-1T.

Thanks again for taking the time to read this and I hope someone can help me solve this bottleneck problem which was probably holding me down the entire time I was trying to overclock.

 

[BTRY B 529th FA BN]

Memtest 86 I think its either test 5 or 6 that is the most stressful and will exploit errors first. The rest I wouldn't know about cause I am selling my D0 in favor of keeping my Q9650 & UD3R

 

[REDtoLINE]

I also forgot to mention that I did a few tests with LinX as well. I tried doing tests using the minimum amount of memory (16MB) and 500 runs, but it would fail between 10-30 seconds which is about test #75-#150. I also tried testing using all the available memory (5126MB) and 20 runs, but it seems to fail around 10 minutes into testing which is test #2.

I will download Memtest86+ and give that a try. I never really trusted it however because in the past I would have it pass my memory after hours upon hours of testing, but I would BSOD a Prime95 blend test within the first 5 minutes. I guess it couldn't hurt but I was told MCI Memtest is more reliable and I passed it after a few hours so I felt that my memory was stable, not to mention it's running pretty slow considering what it's rated for.

 

[REDtoLINE]

Okay I just ran Memtest86+ and decided to let all the tests run since I feel they are all important. After 40 minutes, all tests passed with no errors. I'll let it run some more but i'm pretty sure i'm not going to get any errors. This memory is running well within spec, and there is no reason for it to fail. Especially since Memtest86+ and MCI Memtest have both confirmed that it is stable at 6-6-6-18 at DDR3-960 with 1.5v, but obviously the question at hand is still what is causing Prime95 large FFT, Prime95 blend, and LinX tests to fail?

 

[aigomorla]

its probably either VTT or not enough QPI/DRAM.

try boosting QPI/DRAM to about 1.325 and then test out what vcore 3.8ghz requires.

And to answer your other question, 1.25 but it goes down to 1.22, in bios you need to enable load line calibration. That should fix it.

Or a Vdroop option. However if your gonna play with these make sure you dont push your cpu voltage very high. Ummm 1.475+ high.

 

[REDtoLINE]

Originally posted by: aigomorla
its probably either VTT or not enough QPI/DRAM.

try boosting QPI/DRAM to about 1.325 and then test out what vcore 3.8ghz requires.

And to answer your other question, 1.25 but it goes down to 1.22, in bios you need to enable load line calibration. That should fix it.

Or a Vdroop option. However if your gonna play with these make sure you dont push your cpu voltage very high. Ummm 1.475+ high.

Well when I was trying to get DDR3-1600 to work, I played around with the QPI Voltage a lot, and could not get anything to be stable. However, I'm actually running Memtest86+ right now with my RAM multiplier at 10x giving me DDR3-1600 7-7-7-24-1T (which are the specs for this RAM) and a QPI Voltage of 1.35v (I tried 1.3v but Memtest86+ wouldn't start) and the RAM seems to be stable. Although I'm only 25% into the test, it's looking quite promising. This would mean that all the times I tried to achieve DDR3-1600 in the past, it wasn't my RAM causing the BSODs like I had originally thought.

Now I assumed that because I was running the RAM at the stock Intel specs of 1.5v, and below the stock DDR3-1066, that I wouldn't need to increase the QPI Voltage at all. I would think anything over 1.2v would be too much, wouldn't you? I guess I'll give that a try after this DDR3-1600 test is finished, but I was thinking that there was a setting besides CPU Voltage, QPI Voltage, or DRAM Bus Voltage that was causing these tests to fail when the RAM was at DDR3-960.

 

[Rubycon]

Change your ram. There are known issues with OCZ modules and the P6T6 Revolution. Corsair Dominators 8-8-8-24 1T 12GB at 1600 work perfectly here.

This platform also prefers odd multi's. 200x19 should work perfectly. (C0) 21x200 should work perfectly for a D0.

 

[REDtoLINE]

Originally posted by: Rubycon
Change your ram. There are known issues with OCZ modules and the P6T6 Revolution. Corsair Dominators 8-8-8-24 1T 12GB at 1600 work perfectly here.

This platform also prefers odd multi's. 200x19 should work perfectly. (C0) 21x200 should work perfectly for a D0.

Where did you get this information from? I never heard anything about OCZ modules having issues with my motherboard.

 

[Rubycon]

Originally posted by: REDtoLINE


Where did you get this information from? I never heard anything about OCZ modules having issues with my motherboard.

I'm in a unique position where hardware is pretested for compatibility before getting sent to the docks and this was one thing that came up with this combination. You could have a bad board but I would be suspicious of that ram.

 

[REDtoLINE]

Originally posted by: Rubycon

Originally posted by: REDtoLINE


Where did you get this information from? I never heard anything about OCZ modules having issues with my motherboard.

I'm in a unique position where hardware is pretested for compatibility before getting sent to the docks and this was one thing that came up with this combination. You could have a bad board but I would be suspicious of that ram.

Well I really don't know what to do. I don't know anyone who has some DDR3 I could test with. I was suspicious that it was the motherboard when I had to use such high voltages for a small overclock but after I realized I was severely overvolting, it turned my attention to the RAM. It seems like it just doesn't want to cooperate with this board, and it's driving me crazy.

 

[BTRY B 529th FA BN]

Sounds like a motherboard problem RMA A.S.A.P.

 

[Rubycon]

Originally posted by: BTRY B 529th FA BN
Sounds like a motherboard problem RMA A.S.A.P.

It could be. But ram is a lot cheaper to ship than a mainboard. I'd tell OCZ to keep their sticks too.

 

[REDtoLINE]

Well first I'm going to try and find someone with DDR3 RAM that I can test with. If it ends up being the RAM, I really have no idea what I can do. It's not like I can just return the RAM, I bought all this stuff in May. I just recently got a heatsink, and that's what motivated me to do all this overclocking. So even if I got new Corsair, what would I do with this OCZ? Otherwise if it is the motherboard, then I'll have to send it back to ASUS because I it's too late to RMA to newegg, and ASUS is going to take forever to get me a new one if that's the case. This is such a pain, what a waste of 3 days. I'm just putting everything back to stock, screw this.

 

[ectx]

Originally posted by: REDtoLINE
However, I'm actually running Memtest86+ right now with my RAM multiplier at 10x giving me DDR3-1600 7-7-7-24-1T (which are the specs for this RAM) and a QPI Voltage of 1.35v (I tried 1.3v but Memtest86+ wouldn't start) and the RAM seems to be stable. Although I'm only 25% into the test, it's looking quite promising.

How did this test go? If I were you I'd use the setting and begin testing in Windows. I have the same memory and I passed the memtest86 after 24hr at ~1600 but cannot get it stable in Windows (prime 95 large in place, Linpak 5120mb test) at that speed at all.

I can get my memory stable at 7 7 7 20 1t at 1.64v (never even bother with a lower voltage) at 172x8=1,376 for a Gigabyte mb (EX58-UD4P) and an Intel mb (DX58SO). Can get it higher with looser timing/higher voltage (I'd say about 185x8 at 8 8 8 23 1t, but I am not 100% done with testing yet).

In my situation, I had to bump up VTT (QPI/DRAM on Asus, Gigabyte) before I can get the systems stable at 155-160 BCLK... Need to bump up different voltages to get higher Bclk (one chip of mine would not be stable at BCLK 175 and the other, a D0 stepping, could go higher but I have not done testing yet).

 

[BTRY B 529th FA BN]

^ if it fails large fft's wouldn't that have more to do with your motherboard subsystem than the memory alone!? that's what I always though...

 

[Rubycon]

Try 19x200, DDR3 1600 9-9-9-24 1T, CPU voltage 1.35 LLC=ON (same as vdroop OFF) and leave everything else on AUTO. That should not be an issue with this board at all.

 

[REDtoLINE]

Originally posted by: Rubycon
Try 19x200, DDR3 1600 9-9-9-24 1T, CPU voltage 1.35 LLC=ON (same as vdroop OFF) and leave everything else on AUTO. That should not be an issue with this board at all.

Well I tried those settings and it would freeze on the windows loading screen (I also tried 1.4v and it made no difference). However, I just moved the CPU Multiplier to 17x (putting me at 3.4 GHz) with a CPU Voltage of 1.25v, and I was able to boot into windows. I quickly turned off my computer though because when I opened TurboV, I saw that the AUTO setting on my DRAM Bus Voltage was a whopping 1.8v, which is way too high for my liking! So I manually put in 1.64v, and windows loaded. I'll run some torture tests and get back to you on how things are going. I just don't think my CPU is going to get to 3.8 GHz, which is really sad. I just replaced all the fans in my case with 120mm 90CFM 2000RPM fans, and recently bought a new high end heatsink.

I thought my NB was holding me back this entire time, but now that I have a perfect 200 MHz BCLK I can easily test the CPU at 3.4 GHz, 3.6 GHz, 3.8 GHz and so on. Once I get all that stable, I can then make sure my NB and RAM is stable (I don't see any reason why it wouldn't be, the timings are super loose) and start trying to tighten the timings. I might have to keep the RAM at worse timings than what it's rated for to get it stable (I could increase the DRAM Bus Voltage but I don't feel comfortable running it over 1.66v), or possibly get new RAM. I guess my CPU and RAM were both holding me back, and now I can start stablizing each one at a time.

Thanks for your help Rubycon, this is the spark I needed to start testing and hopefully I can get my CPU and NB to be stable, that would make me happy cause replacing RAM is nice and easy. Thanks again!

 

[REDtoLINE]

Alright I seriously believe now that this RAM is holding me back. I could not get 3.4 GHz stable and eventually couldn't even boot into windows using any voltage, even the 1.25v that originally worked. I tried 3.2 GHz at many different voltages, and 3.6 GHz and nothing was working. Using a BCLK of 200 MHz means the lowest speed I can run my memory is 1200 MHz, and finding timings for a speed that doesn't have JEDEC is a real pain. So I said screw it and decided to try 3.8 GHz using a CPU Multiplier of 21x and a BCLK of 181 MHz which allowed me to run the RAM at 1086 MHz with 1.5v and I kept the 9-9-9-24-1N timings. Using these settings I couldn't boot 3.8 GHz at 1.375v, but it did boot at 1.4v however it was unstable. I've BSOD a couple of times and am currently testing 1.425v just to see if the 3.8 GHz is stable at this voltage.

When I use a BCLK of 200 MHz, my CPU should still be able to run at 3.8 GHz since it can run at 3.8 GHz using a different set of BCLKs and Multipliers, correct? I have a really nice motherboard and doubt that it cannot handle the 200 MHz BCLK, especially when guys are getting theirs beyond 230 MHz on some forums. All signs seem to be pointing at the RAM. Any thoughts?

 

[BTRY B 529th FA BN]

I had issues with my voltages spiking. I RMA'd the PSU's. Sold the new ones off and bought 2 Enermax Revolution 950's and that solved my problem. That doesn't rule out a bad mobo though...

 

[Rubycon]

You should really try this with different RAM. The board has no issues with BCLK >200 at all.