Noobie OC attempts on Opty 165 & Ultra-D

[Quinn]

As was originally suggested, I went back to play with RAM timings. The vast majority previously had the default BIOS setting of "Auto". I found what looked to be a reasonable match in the OCDB, and changed my settings to match. I'm currently using:

 

DRAM Frequency Set............................ - 166=RAM/FSB: 05/06

Command Per Clock (CPC)....................... - Disabled

CAS Latency Control (Tcl)..................... - 3

RAS# to CAS# delay (Trcd)..................... - 4

Min RAS# active time (Tras)................... - 8

Row precharge time (Trp)...................... - 3

Row Cycle time (Trc).......................... - 13 Bus Clocks

Row refresh cyc time (Trfc)................... - 16 Bus Clocks

Row to Row delay (Trrd)....................... - 03 Bus Clocks

Write recovery time (Twr)..................... - 03 Bus Clocks

Write to Read delay (Twtr).................... - 02 Bus Clocks

Read to Write delay (Trwt).................... - 03 Bus Clocks

Refresh Period (Tref)......................... - 2560

DRAM Bank Interleave.......................... - Enabled

DQS Skew Control.............................. - Increase

DQS Skew Value................................ - 220

DRAM Drive Strength........................... - 8

DRAM Data Drive Strength...................... - 1 (Reduce 00%)

Max Async Latency............................. - Auto

DRAM Response Time............................ - Fast

Read Preamble Time............................ - Auto

IdleCycle Limit............................... - 4 Cycles

Dynamic Counter............................... - Disable

R/W Queue Bypass.............................. - 4x

Bypass Max.................................... - 2x

32 Byte Granularity........................... - Disable(4 Bursts)

 

With these settings the strange problems on various divider settings seems to have gone away. I can't say I completely understand them all and believe some of these can be tweak this way or that for better optimization. The RAM is running at 245Mhz @ 300 FSB, with DRAM voltage at 2.80v. I booted windows, and started OCCT. The system froze as OCCT was loaded. Since I'm pretty sure my CPU can handle this speed and higher, I decided to bump the RAM voltage to 2.9v. At this setting, I was able to start OCCT, but it failed during the "idle" test stating a computational error. So I bumped the RAM voltage upto 3.0v, and now OCCT loads and executes w/o problems. Still need to dual prime, tho. And I might still be able to raise the FSB another 20 or so, at least from a CPU standpoint. For now tho, 2.7Ghz is fine.

 

From looking throught the OCDB, it would appear that requiring RAM voltage of 3.0v is a little high for most rigs. It seems many get their memory running at this or higher speeds with around 2.8v. Do this mean I have a problem? Or is 3.0v really ok and should I not worry about it? Will fine tuning the memory settings more be likely to allow me to reducing the voltage i'm using? My gut feel is the answer is NO, but I'm a rookie, and am hoping it can be. Thoughts?

 

Some basic details on the OC at this point:

300 FSB, stock cpu voltages: about 1.32v

ambient temp is about 80F,

at idle: 26-27C cpu, 42 PWMIC, 36 chipset

at load: 38C cpu, 55 PWMIC, 43 chipset

 

Appreciate your help.

[Wild Thing]

Try these, let me know how it goes.

 

DRAM Drive Strength........................... - 7

DRAM Data Drive Strength...................... - 3

[Praz]

Also try DRAM Response Time set to Normal. That will relax the MAL/RP values which should help at 300 FSB.

[Wild Thing]

Using 4 memory sticks on nf4 boards can cause problems. Very few ppl have successfully used 4 sticks. Try 2 sticks in the orange slots and various combinations of 2 sticks in the roange slots. Change memory voltage to 2.8 and up the cpu voltage to 1.4. Djrop mem divider to 133. you should be able to clock that 165 to 2.8 on stock voltages. The 4 sticks are probably holding you're oc back.

[Quinn]

Ran a few more stability tests.

 

SuperPI 32M failed after 12 mins: "Not Exact Round" error.

Dual Windows Memtest: 2 failures after 800% coverage.

 

Try these, let me know how it goes.

 

DRAM Drive Strength........................... - 7

DRAM Data Drive Strength...................... - 3

Will try.

 

Praz gives out better advice as time goes on!

Also try DRAM Response Time set to Normal. That will relax the MAL/RP values which should help at 300 FSB.

Will try this too.

 

Using 4 memory sticks on nf4 boards can cause problems. Very few ppl have successfully used 4 sticks. Try 2 sticks in the orange slots and various combinations of 2 sticks in the roange slots. Change memory voltage to 2.8 and up the cpu voltage to 1.4. Djrop mem divider to 133. you should be able to clock that 165 to 2.8 on stock voltages. The 4 sticks are probably holding you're oc back.

My bad, didn't update my sig (have corrected this). I purchased 2 sticks of OCZ PC4000 Gold GX XTC. This is the memory I've been using since yesterday and one of the reasons I believe I got something settings way off because of the amount of vdimm I seem to need with this quality memory.

 

I'll try the settings changes both of you suggested and report back.

 

If stability doesn't improve, I might try dropping the memory divider a notch and see if it will pass more stress tests. Since dual prime keeps failing on core 0, I'm not sure my failures are all from memory. Think I might need to up vcore a tad?

 

Thanks for your help.

[Quinn]

Made those three changes.

 

SuperPI 32M ran successfully

Dual windows memtest failed :eek:

Dual prime failed on core 0 :eek:

 

Going to try drop the divder now.

[Wild Thing]

Glad you're getting some success. Up vcore to 1.400v

 

Do you have dual prime setup properly? Here: http://www.dfi-street.com/forum/showthread...ht=prime95+dual

[Quinn]

Glad you're getting some success. Up vcore to 1.400v

 

Do you have dual prime setup properly? Here: http://www.dfi-street.com/forum/showthread...ht=prime95+dual

Yes, I used that thread when I had set it up, so I believe its configured properly.

 

I'm running the dual windows memtest with the divider set to 150 3:4, and at about 500% coverage and no errors *fingers crossed*. If this completes successfully, I'm up vcore and see if I can get prime to run on core 0.

 

Will post an update as I get more results.

 

Thanks

[Praz]

If core 0 is consistently failing by itself and not core 1 also, you can be pretty sure it's cpu related.

[Quinn]

If core 0 is consistently failing by itself and not core 1 also, you can be pretty sure it's cpu related.

Thats my feeling as well. I'm pretty sure I'm dealing with two issues ATM. One being the memory. Dual memtest have run longer now, with divider at 150 3:4, than ever before. But its still climbing its way to 2000%. The second being the CPU itself. After memtest completes (hopefully successfully), I'm going to play with vcore to see if I can core 0 to prime. After that, I'll go back to attempting to get the memory to run well with less vdimm (its still at 3.0v).

[Quinn]

Good news and not so good news.

 

Good news: At 3/4 mem divider, dual windows memtest completed 2300% coverage w/o failure. Additionally, I managed to get dual prime stable. At stock vcore, core 0 would fail within seconds, with vcore of 1.4v, core 0 failed in 4 mins, With vcore at 1.45v @ 2.7Ghz, dual prime ran for 10+ hrs w/o failure. (I need to go back and see if vcore of 1.425v might be ok.) I'm a little disappointed that due to the "weak" core 0 in my cpu, that I wasn't able to manage 2.7Ghz at stock vcore. I've also noticed that even small changes in vcore result in rather large temp increases, even on WC. For me +0.1v vcore up my max tcore from 38C to 47-48C range. I'm actually surprised that cpu got that warm. I expected my WC to be able to disappate this heat better. Perhaps I'll remount the block to see if anything improves.

 

BTW: at these settings, ambient 80F, idle is 30C, dual prime load is 45-47C. PMWIC is pushing into the low 60s at full load. Gotta concoct an additional fan mount to cover this "dead spot" on the board. Chipset temps around 40C, setting SG to max chipset fan RPMs drop this from about 50C.

 

Things that still need investigation:

- Remount waterblock to see if better heat removal possible

- See if other RAM settings changes will allow me to lower my vdimm below 3.0v and hopefully back at a divider setting of 5/6, closer the rated the speed of this ram.

- See where the dual prime failure occurs at stock vcore. Curious to how much extra CPU speed I'm getting the for apparant extra 8-10C the added vcore is creating.

 

I'm curious to know of others experiences on Opty max OC abilities at the various testing stages. At stock vcore, I found I could successful run BIOS memtest at FSB 320+ and 2.8v vdimm. To get windows to boot, I had to drop FSB to 300 and increase vdimm to 3.0v. To get dual prime stable I had to up vcore by 0.1v. Is it typical to have such a large swing (20+ FSB), to go from BIOS memtest max OC to windows boot max OC? I was thrilled when BIOS memtest passed at 320. But now a little disappointed that I can't even seem to run at 2.7Ghz without bumping the vcore and generating a significant increase in heat. Is my difference in "max OC" stages normal? and its just lack the experience to realize it? For those, like you Praz, can your system that run at FSB 330+, run BIOS memtest w/o failure at 350 or higher?

 

As always, your thoughts and comments are welcomed.

[Praz]

Everyone has a different opinion on the value of Memtest. I only use it for two things when testing memory. One is to validate that the memory will work at the manufacturer's stated specs. The other is as an indication that the timings won't completely hose a Windows installation. The secondary timings are much more critical under Windows then when running Memtest.