Frustrated Opteron 165

[Remeniz]

I'd like to take this opportunity to say hi and yes, this is my 1st thread on this fantastic site.

I've got an AMD Opteron 165 1.8 Ghz Dual Core CPU with CCBWE 0551UPMW Stepping.

Any ideas if this is a good stepping code?

I have PC4000 RAM which means I can up my FSB to 250Mhz without worrying. This will give me 2250MHz. I have had it running at 2600Mhz and tested for 7 hours with Orthos. The next morning when I spotted no errors I thought I was on to a winner with this chip. But when I rebooted it wouldn't POST. So i've settled on a modest OC of 2347MHz.

This was the highest ever OC i've had with this CPU. But it gave errors in Orthos and restricts my memory to 150MHz, or 300MHz DDR PC 2400. HT X 3.


This is next best. As above but with a slower FSB speed.


This is a 2600MHz OC after a 7 hour stress... and after a re-boot it wouldn't POST.


And this is what i've settled for. Although I will continue to try and get a stable OC @ 2600MHz.


So is the CPU stepping ok? And whats your conclusion on the OC's?

 

[Remeniz]

I'm running at 2712MHz right now. Started Orthos to stress the cores. It failed straight away on Small FFT's and Large FFT's. And the errors seem to be on one of the cores.

 

[jonspd]

You should be able to get atleast 2.6 stable from that set IMHO.

It's all a matter of playing with the settings.

What cooling and psu are you using having 2 psu's might not help.

 

[Wayward_Son]

If memory serves, CCBWE isn't a bad stepping but it's not the greatest. Should give you some decent clocks.

It really looks like you're trying for too much too quickly. You're playing with so many options at one time without isolating the variables: you're changing your vCore radically from one profile to the next, changing your HTT link multiplier and doing who knows what with your RAM divider. Slow down!

Also, a seven hour torture test isn't usually looked at as a good stability test. Most people here go for an absolute minimum of 12 hours with many recommending a full 24 hours of continuous testing with Prime or Orthos before they are confident in the overclock.

You've arrived at a stable clock of 2.35. Now it's time to tighten everything up at that speed. This means finding your best RAM timings at that speed. Looks like you're running two sticks of dual channel? Try dropping your command rate to 1T. You shouldn't have any problem and you'll get slightly better RAM performance, but if you're running four sticks then you must run 2T. Also, you should be manually setting your RAM divider, not leaving it on Auto. Same for RAM voltage. I would manually set your RAM divider to DDR400 and voltage to 2.7v to start with. Now, do a 10-12 hour torture test. If it's stable, then I'd move on to CPU.

If your CPU will run at that speed at 1.4v, then try dropping it down a notch at a time until you start getting errors in Orthos. Also, since you have the Premium version of that board (I only have the Basic) you should see an option for Enable CPU Overvolting or something to that effect. Enable that feature.

Leave your HTT link multi at x4 for now.

While you're at it, disable Cool-N-Quiet in your BIOS.

Give that all a try, run some good torture testing and get back to us. You've got a good board and a good processor so you should see some nice clocks, but don't get ahead of yourself. Baby steps.

 

[Wayward_Son]

Also, what are you using for cooling? While you're torture testing, what are your temps with Core Temp?

http://www.alcpu.com/CoreTemp/

 

[Remeniz]

Woah. Thanks for the replies peeps.

I should slow down I guess.

Ok well at the moment i'm at 2700MHz and my vCore is at 1.5. It was at 1.4 but it wouldn't POST so raising it to 1.45 got me into Windows and fail after 3 seconds in Orthos. I'm going to try Orthos now and see how I get on. My HT is X 3 to get 900MHz. The thing is my computer will run fine at these setting until I stress the cores, and one of them will fail.

Now my gripe is with the Memory. Im only running it at 300Mhz at the moment but if I move it up to the next divisor to get nearer 500Mhz it gets stuck on the POST screen. Do you reckon changing the command rate to 1T will help? I remember trying to run it at 1T and it didn't want to POST. I'm running dual channel.

For cooling i'm using the Akasa Evo 33 HSF which I think is doing very well. Because idle, at this current OC of 2700Mhz, is only 36'C and during a stress it doesn't go over 55'C, if the test runs long enough for the CPU to reach this tempearture that is.

With regards to the two PSU's I think they fine. I have the 450W PSU running just the Mobo so this should be better than running the Mobo and HD's etc. I have a 250W PSU running just the HD,s, Optical Drives, etc.

 

[Wayward_Son]

You shouldn't be worrying about 2.7 right now. Again, I advise that you stick to that 2.35 clock and then tighten up all your timings and voltages so that the the entire rig runs stable at that speed with the fastest memory and motherboard speeds with the least amount of voltage possible. Once you're stable there and everything is trimmed to perfection, THEN and ONLY THEN do you start going higher, incrementally.

Instant gratification doesn't work well with these Opterons. My 3700+ Sandy was a breeze to overclock, but figuring out how to make everthing work on my Opty 170 (a virtually identical chip to yours) took months. I am now satisfied that I am running this thing at it's absolutely fastest stable speed with the least amount of voltage, and that took a lot of patience on my part.

 

[Remeniz]

Ok. I'll revert back to my small OC, tighten things up and go from there.

What is it that I need to focus on when tightening the settings?

 

[Wayward_Son]

Faster RAM timings with less RAM voltage while maintaining stability. You're trying to trim back latencies and increase speed, all the while using the least amount of voltage to be 100% stable. You do this by tightening up the timings like going from 3-4-4-8 to 2.5-3-3-7. If that's stable, then maybe try 2.5-3-3-6. If that's stable, keep it and try 2.5-3-2-6. Or 2.5-3-2-7. You want the lowest lowest timings that are stable. Then lower the RAM voltage a bit at a time, like from 2.8 to 2.75, then 2.7. Every time you make a change, you must run stability tests! It's not an overnight process. Also, like I mentioned earlier, try 1T instead of 2T.

Keep your HTT at x4 for now.

For CPU, give it 1.425 vCore while you concentrate on your RAM timings. Once your RAM is tightened up and the system is stable, I mean REALLY stable, then you can start reducing vCore one notch at a time (1.4125, then 1.40, etc.) doing stablity tests after every incremental change until you find the minimum voltage where the system is stable in Orthos for at least 12 hours.

You can look at it like a sports car engine: you're trying to extract every last bit of horsepower from the engine, getting the car to go as fast as it possibly can, while using the minimum amount of fuel to accomplish that. You're maximizing power, speed and fuel efficiency all at the same time.

Same thing for your computer. You're trying to find a point where your system is as stable as possible, you're getting everything out of your RAM that you can get, all the while using the lowest amount of vCore and vDIMM to keep things running without crashing.

Once you have reached that point, use the computer or a week or two without touching anything in your BIOS. Surf, game, multitask, benchtest, do whatever it is you normally do with the rig. If it's performed flawlessly during that time with no crashes or anything, THEN you can start trying to find a faster overclock profile. The idea is to form a stable, efficient foundation and then move up bit by bit from there.


EDIT: To further illustrate the idea I'm trying to convey, take a look at my rig. If I try to go any faster at all, or reduce any voltages anywhere, then go from a 100% stable (and very fast) overclock to one that crashes after a few hours in Prime95. If I change a single RAM timing any lower then they are, then it starts failing Prime. I am currently at 285 x 10 for my processor. If I go just one more notch to 286 x 10 (a grand total increase of 10 mhz processor speed, from 2.850 to 2.860 ghz), then it starts failing Prime. If I lower my vCore from 1.4 to 1.3875v, it starts failing Prime. This this is going as fast as it possibly can with the fastest timings and the least amount of voltage, and it is rock-solid stable. I have pushed it to the absolute limit of what it will do. I did that by carefully adjusting one single variable at a time (a RAM timing here, a bit of CPU voltage there) with lots of stability testing every time. And as I said before, it literally took months to get it tuned.

 

[Parad0x420]

excellent write-up Wayward! i too am in the process of stabilizing my OC, on my fx60 that is.. since i switched M/Bs. sounds like i will have to follow a bit of that advice and test test test some more.

 

[chop_wood]

id rather slowly find the stable cpu max (in small increments and lots of testing low voltage as recommended)and keep my mem at low mhz loose timings until everything is cherry . id hate to find the ram sweetspot at ddr400 for example, but then having to find it all over again with new ram divider once i found the max cpu oc.

 

[Wayward_Son]

That's a valid point, Chop Wood. The other route is to keep the RAM and HTT link at low speeds and find the CPU's maximum clock with a good voltage, then go back through and tighten up the RAM and motherbaord.

I am giving my advice based on the idea that our fellow has found a stable clock at 2.35 for right now. Before going any higher, let's solidify a stable clock at this speed that we can then use as a jumping-off point for higher clocks.

If it were my computer, I'd certainly be comfortable shooting for some higher clocks beyond 2.35, but then I feel that I'm experienced enough with A64 overclocking on the A8N-SLI that I could do that without many worries. Judging by what I've seen so far, I don't feel that Remeniz has that same level of experience (no offense intended) so it's best to let us start at a tight, solid, stable overclock of 2.35 and then go up from there.

 

[Remeniz]

Ok. I taken some of your advice, thanks peeps.

I reduced the HT multiplier down to 1, (the lowest setting), and the RAM divider to 200MHz, the lowest setting.

The maximum FSB stable speed I achieved is 2712MHz @ 1.472V. The temps are very good at this speed and volts.

I got the system to boot into windows and have been running it for a few hours, doing what I usually do, and i've had no hang up's or BSOD's.

Do I work on the HT speed, which I think is straight forward at X 3 or do I work on the RAM timings which are set @ 2.5-4-4-8 ?

 

[Wayward_Son]

Sigh.

Look, you're not listening. I'm trying to give you really good advice, but you're not listening.

Lowering your RAM divider to the lowest setting isn't the answer. Neither is reducing your HTT to x1.

I've given you all the advice I can give you at this point. Good luck.

 

[Remeniz]

Thanks anyway.

 

[chop_wood]

lol sorry wayward if i may have undone some of that nice work you had going there.

as wayward said, that few hours isnt going to cut it. get a 12hr orthos of mid day summer heat (not just overnight), before you add more variables.. wayward knows whats up.. on my opty 165 before this current rig it took me a month or 2 before i had my golden 3.0ghz running for the next couple years. and of course i had to redo it everytime switching from 500 2225 winbond > platrev2 @ 600 > 2gb of tccc (esp saying dfi had a different bios for each ram)

 

[Remeniz]

Yea. I thought OC'ing was pretty easy.

I've been on it all day. I've also ran MEMTEST a few times and it's producing errors.

I'll carry on and maybe in a month or two i'll post again.