Alright OC pros, let's get this thing screaming

[dman918]

Alright, I am far from being OC savy, so please give me a hand here.

Here is a short list of my parts:
Opteron 165
MSI Neo4 SLI with latest bios
OCZ Platinum 2GB (2x1GB) rated at 2-3-2-5
Arctic Silver Ceramique and XP90 with Panaflo fan at 5v
Zalman passive northbridge cooler

This is what I have done thus far:

I set the processor to be at 270*9, set the memory at 166mhz, and set the latency to 2.5. The memory is at 1T as well. Stock voltage on everything. I want to keep going but I am a little concerned about how to go about doing it.

I have considered a beta bios for the mobo as they have more memory divider options. I am fine with increasing voltages, but I am not sure how I should go about with the memory. I can go higher, but the memory would have to be set at 133 right now to boot correctly. Is that ok? I didn't think being under ddr400 speeds would necessarily be a good move. Should I increase voltages on the northbridge chipset? Should I loosen my memory timings to a certain level?

Please give me some advice on how to proceed.

Thanks,
Mike

 

[Noubourne]

Not positive on your MSI BIOS, but in mine, setting the RAM to "166" is essentially using a divider, and not actually a manual setting of DDR333. So, at stock HTT speed of 200 Mhz, your RAM is running DDR333, but as you increase your HTT, your RAM speed also increases, until you reach DDR400 again, except with a higher HTT than you started with (and thus an overclocked CPU, and regular-clocked RAM).

So I would use a program like CPU-Z to see what speed your RAM is ACTUALLY running at. I think you'll find that you're higher than 166Mhz at an HTT of 270, and probably well over 200Mhz (or DDR400) as well. If you kick down your "divider" to 133, then it should give you more room to see just how far your CPU can go. At 270HTT though, "133" is probably going to be well over DDR266 - which is what it would be IF your HTT was at the default of 200.

Hope that's not too confusing.

 

[imported_Vanden]

Get cpu-z to see the speeds.

Find your max htt. Drop cpu multi and mem divider so that they dont run above stock speeds, and start raising the htt.
Do the same for ram speed en cpu speed. Find the max values when the other two are below or at stock speeds.

Now you'll know what limits your overclock and what you need to adjust.
Start at the stock voltages. Bump the cpu voltage when you hit a max and see how much you need to get higher. At some point you'll need big voltage increases for a small gain in speed, time to stop. Keep an eye on your temps (speedfan).

Loosen the mem timings if you want to run your mem at higher speeds then stock.

Test for stability. I use superpi 32M for quick stability test. Prime95 or sp2004 for long tests, 24h+.

 

[n7]

Your Ram is presently running @ DDR440 (220), since dividers don't lock the speeds; they merely set the RAM to a ratio of the HTT & CPU speed.

I'd use the 133 divider & keep timing low (2-3-2) & try going higher.

Or, do things properly.

Follow Zebo's guide here, which still uses the best concept.

Find HTT max
Find CPU max
Find RAM max.

Here: thread - http://forums.anandtech.com/messageview.aspx?catid=28&threadid=1497607

 

[dman918]

That definitely makes sense. I am at 166 which makes me at 215mhz once the htt is considered. The problem is that I have to change my cas latency to 2.5 in order to run at the oc'ed 215. If I go to 133, the actual speed was ~180mhz, so I was thinking that may be hurting me a little in overall speed since that is not even the stock ddr400 speed. I was hoping a middle ground (~150 option) would give me a real speed of about 200, the stock speed. Yeah, it does get confusing when using all the different numbers. I will have to try some of these recommended techniques when I get home. I guess I am wondering how much the memory timings and speeds can be changed without too much of a performance hit. Also, is there a reason to raise the voltage of the northbridge?
Thanks!

 

[n7]

How to calculate RAM dividers:

(CPU multi x 200) / RAM divider (i.e. 166.66) = ___(always round this up if not whole number)

CPU multi x HTT / result of initial calculation = actual RAM speed

e.g.
OCed CPU clockspeed / result of the first calculation = actual RAM speed

one full example:

12 x 200 / 166.66 = 14.4 rounded up -> 15

12 x 220 / 15 = 176 aka DDR352



With A64 architechture (yes, same tech as your Opty), RAM bandwidth isn't a big issue.

You'll get far better overall performance by running a faster CPU with slower RAM than trying to base your CPU speed around your RAM.

Also, there are a number of excellent alternative bioses that will give you additional dividers as you mentioned that will work well.

 

[dman918]

Thanks for the response.

Here is what I got out of my initial OC testing.

Max HTT = 319 (335*.95)

Max CPU =
1.35v -> 2437 (285*.95=271)
1.465v -> 2565 (300*.95=285)
1.538v -> 2610 (305*.95=290)

I was very suprised to see such a small difference when increasing voltages. Is this normal? Also, if you take the 1.538v setting, speedfan views it as only 1.47v, so should I get the beta bios and crank up the voltage a little more? I heard you can get 1.5v-1.55v safely with this chip.

I will test memory next. Would it be better to be at the higher speed with looser timings, or at a lower speed with tighter timings?

Thanks,
Mike

 

[dman918]

OK, so after testing memory a little bit (with memtest86), I found that I could get 2-3-3-6 at 193mhz. This is with the processor at 2610mhz (290*9).
I tried running it at ~240mhz at 3-3-3-8 and it gave a couple errors, and I didn't know if I should loosen the timings more than that.
What do you all think? Does this seem like the best bet, or should I try a beta bios, up the voltage on the cpu a little more, and try a memory divider in between the two I currently have?

 

[imported_Vanden]

Just to be sure, but you dropped your ht multi right? If not, you are at 290*5=1450. You should keep it below 1000, so use the 3 multi.

 

[n7]

You should be able to run 2-3-2 @ 193, no?

What i'd do to decide whether to run 193 2-3-2 or 240 3-?-? is run some benches & see what giving you better scores for the applications you wish to run.

If 3-3-3 @ 240 isn't stable, maybe try 3-4-3 or 3-4-4.

I'm guessing 2-3-2 @ 193 will actually give you better performance, if you can find any difference at all...

 

[dman918]

Thanks for your responses.
Well, I ran into some issues. I do have my ht multi at 3, so that is fine. For good measure, I tried dropping my htt to 285, which makes the memory run at 236.55. No go. Even loosened my timings to 3-4-4-10 and it still got errors, so I said ok, I will run at 133mhz with 2-3-2-5 and NOPE. It seems there is a glitch or something with my mobo and it will not post at the 133mhz divider unless I have my HTT back to stock 200. Even tried 3-3-3-8 and 3-4-4-10 and it still wouldn't post. So know I really don't know what to do. Is there a beta bios for the Neo4 SLI that I should try? Maybe there is a fix in it for the memory dividers.
Oh, all errors happened toward the end of test 5 in memtest86. There would also be some errors in test 6.
Advice please!

 

[imported_Vanden]

When the proc is at its limits the memcontroller can start acting up. That not all dividers work isnt that strange for msi. Also the latest bios isnt always the best one for overclocking. Have you tried bumping the mem voltage? Be sure that you know what your mem can handle though.

What are your load temps at now? I've seen opterons 165 get higher with lower or even stock voltages. But its possible that this is your limit, 2.6 isnt a bad oc.

 

[dman918]

Yeah, I don't mind the 2.6 overclock, though I would love to see 2.7 to 2.8 . I can certainly handle 2.6 though. However, the memory issue is certainly a problem. I have raised the voltage from auto to 2.7 to 2.8 to 2.85. No difference. I think the 133mhz divider is just screwy. So... I guess I will need to try a different bios and see what happens. Any thoughts on best one to use, or any other thoughts in general?
Thanks!

 

[Noubourne]

Have you tried upping the RAM voltage to see if it will run at higher speeds? The OCZ Plat is under a Lifetime Warranty guaranteed up to 3.0v, so you can definitely try pushing it a little harder. I do not see that you've tried upping RAM voltage - only loosening timings.

FWIW, my DFI Ultra-D and Venice would not do the lower dividers either; I had to find a satisfactorily overclocked HTT and loosen my RAM timings to get to 2.6Ghz stable, even though I could boot at least partially into Windows at 2.8Ghz using 1.6v. On air, 1.6v took me past 50C quickly, and that was the magic temp for my Venice. My chip hates temps over 50C.

Anyway, you have some room for voltage increases on the RAM. Kick it up a notch and see if that doesn't buy you any more stability. My EL Plat Rev 2 2x512 kit did DDR600 with loose timings and 2.9v, but I didn't like the heat so I decided to knock it down and shoot for latency and quiet fans instead.

 

[dman918]

Thanks for the reply noubourne. I have tried to raise the voltages on my memory and it still didn't work. I could go to a max of 2.85 on the stock bios. That is why I am considering a beta bios that has extra voltages and dividers unlocked. But of course, there is also the question of which beta then. Also, how loose of timings would you recommend? I have gone as far as 3-4-4-10...
Thanks!

 

[dman918]

So what should I do?
It just seems like the 133 divider simply will not work...

 

[dman918]

As an update:
I tried 2 beta bioses and they both had the same problems. It seems only the 166mhz divider works. With the 133, 140, and 150 dividers, the computer will not post if the HTT is set over ~245-250. It will post at the stock HTT 200 level though.

 

[n7]

Are you checking your RAM timings each time you change the divider?

Often switching divider could also mess with timings on the Neo mobos.

Try running 2T maybe?

I mean, worst case scenario, maybe you use the 100 divider with the HTT @ 300, & you get your RAM @ 150, which is so low you could likely go with really tight timings like 2-2-2 possibly.

But honestly, i am betting using the dividers you say don't work is changing timings for RAM, & that's why they won't work.
I bet it's changing more than just CAS - TRCD - TRP - TRAS too, it's likely other settings you're gonna have to mess with...

 

[dman918]

What other settings should I try? I just tried 2t and it didn't change anything. I also tried auto settings and fixed settings for the timings, as far as 3-4-4-10 and it still wouldn't boot at 285 HTT and 133 divider.

NOW: I did try bringing the HTT down to 280 and it did post in the 133 divider, so I am not sure what is going on here...

 

[dman918]

bump

 

[dman918]

Anyone else?

 

[dman918]

Well, I contacted MSI tech support so we are talking via email. I am not sure what else to try here. Does anyone have any idea why the computer would not post at the 133mhz divider even though the effective speed of the memory would be LESS than even the stock DDR400 level?

 

[robertk2012]

will it post using the 100 divider?

 

[bennny]

I have that same board, but with the stuff in my sig. I raised the nf4 voltage to 1.65, and set my ram to 2-3-2-7-11 at 2.85 volts, cpu voltage at 1.4, and using the 133 mem divider for exactly 200mhz on the ram. Have you tried clearing the CMOS and then going in and setting everything up after that. Because the 133 ram speed for me works very well.

 

[dman918]

I am not in front of this computer and won't be again for about a week, so I will let you know for sure, but if I remember correctly, it worked perfectly fine at 100mhz memory divider. I did try resetting everything by pulling the cmos battery, but maybe I will try again for a little longer.

BTW, MSI's customer support people are HORRIBLE. Nice people, but they don't even understand what I am doing, and are trying to tell me there is no memory divider setting. Really a waste of time with them.