deblin
Dark Side of the Moon
Premium,MVM
join:2001-09-01
Middletown, DE
 ·Verizon FIOS
 ·Comcast Workplace
 ·DSL EXTREME
| [OC] Hitting a wall: CPU or RAM?
My P5B started to die today (the ICH8 SATA ports wouldn't work at SATA300, only SATA150), so I bought an Asus P5K-E.
A little history on my box:
- Q6600 CPU (G0 stepping)
- G.SKILL DDR2-800 RAM (supposedly the "good" ones with the Micron D9 chips)
- Asus P5B
I was never able to get it stable at 400x8. At 377x9, I'd get 1 core failing on mprime repeatedly, so I dropped it to 356x9 (3.2 GHz), and it was rock solid at Vcore=1.375.
So, figuring it might have been a limitation of the P5B, when I bought the P5K-E to replace the dying P5B, I figured I'd give it another shot at a better OC.
It STILL will no go past 9x356. At 8x400, it boots and I can run mprime for a while without problems, but as soon as I start compiling world in FreeBSD, I get a pagefault panic.
The RAM is rated at DDR2-800 4-4-4-10, so I don't know why it wouldn't run at 800 in the 4x800 scenario. The voltage for the RAM is supposed to be 2.1v (for DDR2-800 operation), so that's what I have it set at.
Am I hitting the wall with the CPU or the RAM or both? Should I try bumping the RAM voltage up with the 400x8 scenario?
What are your guys' opinions - RAM can't cut it, or did I get a dud G0 Q6600?
Thanks!
--
Hello...is there anybody in there? |
|
deblin
Dark Side of the Moon
Premium,MVM
join:2001-09-01
Middletown, DE
·Verizon FIOS
·Comcast Workplace
·DSL EXTREME
| Also, mprime (which I was finally able to compile and get running in 64-bit FreeBSD) is definitely useful in finding whether the CPU is stable.
What do you recommend for testing the RAM? Would memtest86 find RAM that's failing timing?
--
Hello...is there anybody in there? |
|
deblin
Dark Side of the Moon
Premium,MVM
join:2001-09-01
Middletown, DE | reply to deblin
Ok some more info.
It's not stable at 400x7 or even 400x6!!
So either it's the ram or the CPU simply doesn't like the 400 fsb.
Which is the more likely culprit? :/
--
Hello...is there anybody in there? |
|
deblin
Dark Side of the Moon
Premium,MVM
join:2001-09-01
Middletown, DE
·Verizon FIOS
·Comcast Workplace
·DSL EXTREME
| I forgot to mention that the temps are fine (low 40s under load).
I just booted at 377x9 (3.4 GHz), and I didn't crash the box with buildworld (compiling), but mprime crashes almost immediately.
--
Hello...is there anybody in there? |
|
kidsafe
join:2003-09-21
Burlingame, CA
| If you're getting a rounding error, it's not your CPU, not your RAM. However, while you're testing a CPU overclock you should really bump your RAM down to a 1:1 memory divider and loose timings...like 5-5-5-18.
Also in order to hit 3.4 GHz on my Q6600, I had to bump up my Vcore to 1.4075V in the BIOS (1.33 reported by CPU-Z under load). |
|
deblin
Dark Side of the Moon
Premium,MVM
join:2001-09-01
Middletown, DE | Well I took the Vcore up to 1.45 without luck, but that was at 400x8.
I can't do 1:1 divider since the max multiplier is 9 and 266x9 (stock) is 2.4.
--
Hello...is there anybody in there? |
|
kidsafe
join:2003-09-21
Burlingame, CA
edit:
January 24th, @09:32AM
| Sure you can. A 1:1 memory divider means your memory clockrate equals your FSB clockrate. At 400MHz FSB, your RAM would be running at 400MHz (DDR2-800). However, we're trying to test your max attainable CPU clockrate, so we'd start working in increments.
For example, try 356x9=3.2GHz. Your RAM would then be running at DDR2-712 speeds. Normally a Q6600 requires about 1.35-1.37V set in the BIOS to achieve that kind of overclock. Yours might need a little more. I don't think we've actually found the upper limit of your CPU by any means.
If you can run 3.2GHz stable for >24 hours, bump it up to 378x9=3.4GHz. As I said, I run mine at 1.4V in the BIOS. Even then, your RAM will still be running below DDR2-800 speeds. All the while you should be manually setting your RAM to loose timings just to remove any doubt that it's limiting your OC.
Also you might need to bump up both your MCH and VTT voltage up a couple of notches if you haven't already...or your Vdimm to 2.2V. I really think it's your RAM holding you back...especially if you're getting "pagefaults" at 400x8, but not 356x9. |
|
deblin
Dark Side of the Moon
Premium,MVM
join:2001-09-01
Middletown, DE
·Verizon FIOS
·Comcast Workplace
·DSL EXTREME
| Thanks, now I gotcha! I am actually at the sweet spot of 356x9 right now to attain 3.2 GHz, and you're right about the Vcore, I'm at 1.375 
So should I jump right to 378x9 or try it maybe 4 at a time? Should I set Vcore to auto for 378x9 just to see if it's stable? Or would that not necessarily mean it's getting a high enough Vcore? I am pretty sure I tried 9x378 and it failed mprime with a rounding error on one of the cores, even with a Vcore much higher than 1.40.
The P5K-E has a slew of voltage options, way more than my P5B did, I'm guessing I don't need to touch most of those? I don't know what they are off hand, but I can check.
--
Hello...is there anybody in there? |
|
deblin
Dark Side of the Moon
Premium,MVM
join:2001-09-01
Middletown, DE
·Verizon FIOS
·Comcast Workplace
·DSL EXTREME
| reply to kidsafe
So if I'm getting a rounding error at 3.4 GHz (9x378) even with a 1.4625 Vcore, that just means my CPU can't cut it past 3.2 GHz, huh? And at 400x8 it's not stable either, which points to my RAM. Maybe I'll try 400x8 and manually set the RAM timings (they were set to auto/SPD before).
--
Hello...is there anybody in there? |
|
deblin
Dark Side of the Moon
Premium,MVM
join:2001-09-01
Middletown, DE
·Verizon FIOS
·Comcast Workplace
·DSL EXTREME
|  reply to kidsafe
It's the CPU :(
Ok, it's definitely the CPU. I just ran a bunch of configurations to rule out the RAM and see how high I could go with the CPU:
400x8 @ 1.4v Vcore / DDR2-800 @ 5-5-5-15: fail (page fault)
400x8 @ 1.4v Vcore / DDR2-800 @ 6-10-10-25: fail (reboot)
266x9 @ Auto Vcore / DDR2-800 @ 5-5-5-15: stable
266x9 @ Auto Vcore / DDR2-889 @ 5-5-5-15: stable
367x9 @ Auto Vcore / DDR2-734 @ 4-4-4-12: stable
371x9 @ Auto Vcore / DDR2-774 @ 4-4-4-12: stable
375x9 @ Auto Vcore / DDR2-750 @ 4-4-4-12: fail (mprime rounding error)
378x9 @ Auto Vcore / DDR2-756 @ 4-4-4-12: fail (mprime rounding error)
So even with REALLY loose timings on the RAM at 400x8 it's not stable, and I can run it at 889 MHz with the CPU running stock 266x9 without issue.
So I took 371x9 and set the Vcore to 1.375 and it's stable (so far). Guess I'll stick with that.
My CPU just hits a wall at 3333 MHz. 
--
Hello...is there anybody in there? |
|
Dentist
join:2001-08-11
Waterloo, ON | Try jumping to 400x9? Maybe a FSB hole that you're hitting? Just watch the temps, probably need around 1.45-1.5 volts in that range.
I had issues with mine, easy up to around 3.4, couldn't stick it at 3.6, now running happily at 3.83. |
|
deblin
Dark Side of the Moon
Premium,MVM
join:2001-09-01
Middletown, DE
·Verizon FIOS
·Comcast Workplace
·DSL EXTREME
| It didn't like 425x8 (3.4) either. I'll try again, though, and 400x9 for shits and giggles
What Vcore do you think I'll need for 3.4 GHz?
--
Hello...is there anybody in there? |
|
Dentist
join:2001-08-11
Waterloo, ON
| Are you using a stock cooler or something stronger?
I'm not taking the blame for melting chips :P
At 3.83 (425x9) I use 1.48ish for everyday use (rock solid for gaming, no bsods, crashes, etc), but if I need to crunch a prime or something to prove that it's stable (which I don't care about whatsoever), it's gotta be 1.5+. |
|
deblin
Dark Side of the Moon
Premium,MVM
join:2001-09-01
Middletown, DE
·Verizon FIOS
·Comcast Workplace
·DSL EXTREME
| I'm using a lapped Thermalright Ultra 120 Extreme, so cooling shouldn't be a problem.
I'm back at 3.2 (356x9) at Vcore=1.35, and under full load my core temps are:
dev.cpu.0.temperature: 47
dev.cpu.1.temperature: 47
dev.cpu.2.temperature: 37
dev.cpu.3.temperature: 40
I tried your suggestion, hoping it was a hole in the FSB, but it didn't like 425x8 or 425x9 or even 410x8. :/ Even with Vcore all the way up to 1.55v.
So I guess I'll just stick with 3.2 for now. Not worth frying something over another 200 MHz And the temps are nice and low, which I like.
Guess I just got a dud Q6600
--
Hello...is there anybody in there? |
|
kidsafe
join:2003-09-21
Burlingame, CA | Maybe the northbridge needs more juice? I've never seen a G0 Q6600 that could do 356x9, but not 400x8. |
|
deblin
Dark Side of the Moon
Premium,MVM
join:2001-09-01
Middletown, DE | That's where my OC experience ends. What are the normal FSB and NB voltages, so I know what to bump them to? (This is on the Asus P5K-E if you're familiar with Asus BIOSes).
--
Hello...is there anybody in there? |
|
kidsafe
join:2003-09-21
Burlingame, CA | On my IP35 Pro, I have my CPU VTT set to 1.27V (1.20V is default) and MCH set to 1.33V (1.25V default). |
|
deblin
Dark Side of the Moon
Premium,MVM
join:2001-09-01
Middletown, DE
·Verizon FIOS
·Comcast Workplace
·DSL EXTREME
| I tried 1.4v for both and it was still unstable. 
I really think my CPU is just not capable. I couldn't run at 3.4 GHz on the P5B either, and at the time I assumed it was due to a limitation with the P5B.
--
Hello...is there anybody in there? |
|
deblin
Dark Side of the Moon
Premium,MVM
join:2001-09-01
Middletown, DE
·Verizon FIOS
·Comcast Workplace
·DSL EXTREME
| reply to kidsafe
Do you guys think the power supply could be at fault here? Or is the CPU more likely?
If the PSU were bad, I would think I wouldn't be able to run 3.2GHz (9x356) @ Vcore=1.325v without crashes, but it's 100% stable like this.
Just wondering if maybe the PSU is at fault. Thoughts?
--
Hello...is there anybody in there? |
|
deblin
Dark Side of the Moon
Premium,MVM
join:2001-09-01
Middletown, DE
·Verizon FIOS
·Comcast Workplace
·DSL EXTREME
| I'm thinking about getting one of those PSU testers for like $15. Are they worthwhile? What would I look for in the rail voltages to determine if the PSU is potentially a problem? Just voltages that are > 10% off the rail voltage (e.g. if the rail was 12v, if the voltage is less than 10.8v or greater than 13.2v? Or would fluctuations be the bigger problem? Or both?
Thanks!
--
Hello...is there anybody in there? |
|
