Overclocking Opteron 165 - unusual results

[twivel]

Hello,

I've been playing with my opteron 165 for a couple weeks. Here's my setup:
* Opteron 165 - 0551 stepping
* 2x1GB OCZ Platinum EL PC CAS2 memory
* 550 Watt power supply
* DFI Lan Party SLI-DR Expert
* Thermaltake Big Water CPU cooling

I'm not that new to overclocking but I hope to get some help from some of the more experienced Opteron 165 overclockers out there.

I stopped testing the board FSB at 340 because I would never go that high. RAM runs at 206mhz 2-3-2 1T. 225@CAS2.5. Infineon Memory Chips.

Prior to water cooling, I tested Opteron 165 to run stable at 2556 @ 1.475v 2-3-2 1T (195mhz RAM) for 14 hours dual prime.

After water cooling It ran stable for 12 hours in dual prime at 300x9 (2.7ghz) with RAM at 150mhz with 2-3-2 1T @ 1.525v

Thinking I had plenty of room to spare on RAM, I increased the divider up one step (resulting in roughly a 166mhz RAM speed). It wouldn't finish the POST process.
I dropped it back to 1/2 and reduced CAS to 2.5. Again it wouldn't post.

Ultimately, my Opteron is very picky about CAS and multipliers when running at 2.7.
2.7 - 150mhz RAM - 2-3-2 1T = 12 hours stable @ 1.525v
2.7 - 166mhz RAM - 2-3-2 1T = Fails in POST when verifying DMI data
2.7 - 166mhz RAM -2.5-3-2 1T = Fails in POST when verifying DMI data
2.7 - 210mhz RAM - 2-3-2 1T = Boots fine, fails in dual prime95 (1.525v)
-> Yea, what in the world? Won't POST at 166mhz but WILL post at 210mhz with all other factors the same?!?!

Other results:
2.7 - 210mhz RAM - 2-3-2 2T = Boots fine, fails in dual prime95 (1.525v)
2.7 - 210mhz RAM - 3-3-3 1T = Boots fine, fails in dual prime95 (1.525v)
2.7 - 210mhz RAM - 3-3-3 2T = Boots fine, fails in dual prime95 (1.525v)
2.65 - 210mz RAM - 3-3-3-2T = dual prime95 stable (so far) (1.525v)

-> My opteron seems to be picky about CAS2.5 and some dividers, but not all of the highest ones.
-> Through all of this, my temps are: CPU: <= 42c, PWM: <=45c, CHIPSET: <=45c.

Anyone have any suggestions? I think my memory controller is being pushed to it's limits at 2.7ghz.

NOTE: I use memory timings reccomended by my memory maker for this board[/b] - I only tweak the values in: 2-3-2 1T/2T.
http://www.bleedinedge.com/forum/showpost.php?p=145181&postcount=2
Thanks
Twivel

 

[twivel]

(Previous Results):
2.7 - 208mhz RAM - 3-3-3-8 1T = Boots fine, fails in dual prime95 (1.525v)
2.7 - 208mhz RAM - 3-3-3-8 2T = Boots fine, fails in dual prime95 (1.525v)
2.65 - 204mz RAM - 3-3-3-8-2T = dual prime95 stable (1.525v)
(New results):
2.65 - 204mhz RAM - 2-3-2-8 1T = dual prime95 stable (so far) (1.475v)

So at 295x9@1.475v(2650mhz), I'm able to do really good memory timings. At 300x9@1.525v(2700mhz), it's 12 hours stable @150mhz 2-3-2-8 1T, but upping memory divider at all gets it to where it's either unstable or won't POST regardless of the RAM timings. Again, this RAM has been tested to run fine on it's own at 206mhz 2-3-2-8 1T.

I want 2700, though I'm not sure I want to up the voltage enough to get it. Any other tricks or suggestions on max voltage?

 

[TabrisDarkPeace]

Lock your PCI bus at 33.333 MHz.

Also try testing at 200 MHz even, because the divider is for PCI/FSB ratio is clean there.

May also be clean at 233, 266, and 300.

The Opteron 170 would've been a better choice if those are the speed you want btw. Because you cut costs in overclocking you've limited your max gain.

Using a Opteron 170 (266 x 10) would be damn close clock speed wise to what you have now (2.667 GHz), but you would'nt have the memory problems your having now. Also 280 x 10 (2.8 GHz water cooled or better only) may be possible as other buses would be +/-5% within spec (usually works fine), especially if PCI was running sync, but at 35 MHz (vs 33.333 MHz). This is all assuming 266 knocks the divider up though, and it should.

The tricks to overclocking are usually:
- Keeping PCI bus in spec
- Keeping PCI bus in sync (obviously sync works better than async locking it, but you may be able to push more in async, or you may not)
- Running the slowest timings you can, then working towards faster ones
- Modifying the memory controller timings (not RAM, the one on the CPU), if your board supports this (Tyan boards often do, but they lack overclocking options, and nTune isn't good enough). There may be custom software to do this at a chipset / integrated memory controller level, but I have yet to see any.

Can you increase the voltage to your RAM (hopefully some passive cooling on it) to try and overcome the problems at hardware init ?

================================================

You can run Prime95 from two different folders (or more, up to 32 should be OK) at once btw, so it hits both CPUs and memory at once.

It will cut your testing time in half if you have a dual-core system.... if you are not already doing it. Even on single-core with HyperThreading you'll still gain a fair bit of time. On quad-core it helps heaps

Tabris - Recommended Prime95 settings.txt:

Min FFT size (in K): 128
Max FFT size (in K): 4096 (or larger if supported)
Run FFTs in place: Disabled / Off
Memory to use (in MB): 1/n *
Time to run each FFT size (in minutes): As low as 15 and as high as 60 minutes to personal taste. (15 min recommended though, even for todays systems)

* - Where n is the number of cores/threads the system can run in parallel, and how many Prime95 instances are being run, each from their own folder at the same time.

eg: 4 threads at once = 1/4 memory = eg: 4 GB total, so 1024 MB each

Prime95 can detect errors in calculations (usually caused by faulty memory, or overclocking CPU to far), I don't know if SuperPI can though.

================================================

Of course when testing CPU, also run 2 x P95 at once, but use the inbuilt Prime95 profile for CPU testing.

12 hours is ample testing btw, if it passes the above for 2 hours it should be pretty darn rock solid. Then when selecting the final settings run it in for 12 hours or so.

What RAM voltage are you using anyway ?, those OCZ sticks can go pretty high.

 

[BillW]

It seems like I've tried every possible tweak I can think of. I ordered a new motherboard to test it out just in case that will make a difference. I will soon post everything I've tried. But my girlfriend is b****in about me being on the computer so I'll give a quick rundown.

First of all it runs well at 2.35ghz(2350mhz)at 1.35v with ram running at 218mhz. Anything above that requires an intense increase in voltage.

Overclocking above 2.4ghz requires a significance drop in ram frequency. Also It wouldn't even go above 2.4ghz with 2x512 patriot + 2x512 Hyndai. once I took out the Hyndai ram it went more smooth. Although Patriot ram is still cheap and I'm thinking the ram itself may be why I'm having all these problems.

So with just 2x512 patriot(and i will be receiving another 2x512 soon to add) I was able to get it to boot at 300x9 + 4xHTT + Ram at less than 180(I'm not near that computer so I can't say for sure, It was low though, between 150-180mhz) although I had to increase the voltage of the CPU to 1.55v! and the Dimm to 2.8v. It failed in less than a minute repeatedly on Prime95. But anyways WHAT DOES ROUNDING ERROR EVEN MEAN? It froze after messing around for a while in windows. I have an EPoX EP-9NPA3ULTRA Socket 939 by the way. I am able to clock a 3700+ San Diego to 2.8 at stock voltage(1.4v) which is confusing to me that I have to increase the voltage on the opteron so furiously. Of course the 9x multiplier on the Opteron is probably to blame compared to the 11x on the San Diego. Oh ya and when overclocked to 2.7/1.55v in the bios it shows the processor at 60degree's! What is some good motherboard monitoring software for windows?

The new motherboard coming(which I need even if it doesn't make a difference) is an ASUS A8N-E Socket 939. I have my fingers crossed.

Someone help! I want to get this processor to 2.7-2.8! I will post more about what I have tried and what I will soon. She's b*****in again! Later Guys. Thanks.