New motherboard, new adventure

[ozzy983]

Alright, so the time has come for some results on the new motherboard. I've barely scratched the surface.

First off, let's take a look at the old one @ stock settings.



Now, let's compare to the new motherboard:



OK, so right off the bat, not much difference. Memory is slightly higher, but easily margin of error, or more to test, or some other strange thing.

Now, previous overclock was capped @ 3.1ghz @ 1.525V. I figured, I should be able to do better than that. Truthfully, at first, I was not. However, the whole point to getting this motherboard was for ACC, which has been known to help people with their overclocks.

3.1 @ 1.45v, Acc = +2%:



K, so I tried 1.375, then shot straight to 1.45v. Was freezing second pass in 3dmark, so tweaked acc to +2%, and now it passes a 5 loop of 3dmark. Might be some room for a slightly lower voltage, but will tweak at the very end.

How is the new cooler you might ask?

It's big! Much larger than I expected. Tops the graphics card by 2" or so. How cool you might ask? Idle right now is 26C, and load @ 1.45v is 39C. Says cores idle at 7C, which is impossible, since the room temp is not that low. Old system with tornado was about 15C for the cores, so they are cooler. I figure AMD over reports by about 10-15C, probably a selling thing.

CPU is out of the case atm to rule out air flow issues. As5 was a bit sticky, probably because it sat outside for a bit before I got home to put the package inside.

Their is a feature on the GB motherboard called "EC Firmware", which has no documentation for it, which means it's probably doing something that it shouldn't. From what I can tell online, it is the old code for ACC which enabled processors to unlock cores. It also has been reported to help overclocks require less voltage. I left it at default, but will look to tweak later.

Now, back to the tests.

Update 1:



Oh yeah! Now we are up to 3.2 ghz. Passed 5 way for 3dmark06, so time to move on. Same setting as 3.1, which means voltage was probably way too high initially.

Update 2:



3.3 ghz now, and stable enough. Moved on to 3.4 and that seems to be the spot that is going to take some effort to get through. Basically, back to where i was at 3.1 awhile ago. So, CPU voltage starting to get drained, crashes on 3dmark even at 1.55v, so thinking it may not be a cpu voltage issue. Increased the NB/HT to 2200 which according to an article i read is about where 3.4 should be, just incase the issue is caused by trottling.

I def think it's possible to stabilize 3.4, but past that, not so sure. Not atleast without doing something radical, like increase the vcore past it's threshold.

[Ratatat]

Oh no, Ozzy. I fear you have been afflicted with the overclocking bug. Unfortunately, there is no cure at this time.

Glad to see you got it up and running. Judging from the screens, especially the one of 3.1GHZ, I can see your memory is starting to be throttled by the CPU. Probably not noticeable, but you could benefit from a small increase on the HT/NB to compensate. I'm sure you'll push it at the end.

As for the EC option, that is the old code used by companies to "unlock" additional cores, but it also will help your overclocking, as it was targeted toward black edition processors. It will allow them to use far less voltage than previously, but at the cost of your sensors. They will not function while using the code.

I'm really looking forward to seeing the difference, if any, in the new motherboard. Seeing how your temps have decreased significantly, leads me to believe you'll have better results. This will be true if you can get another core or two enabled. Keep in mind, even if successful, most of the time, the cores are disabled for a reason.

Though, not to get too off-topic, but I believe, toward the end of the dual core cycle, which is the case for the 7550, 7750 and 7850, it is cheaper for AMD to disable two core, rather than actually manufacture a separate processor altogether.

Wow, just noticed your edits. 3.3ghz! that's phenomenal just for a motherboard upgrade.

That, honestly, leaves a few questions that need to be answered.

1) Was the original motherboard in some way defective.

2) If not 1, then is the 770 or 700 chipset(s) really that poor performing

3) If not 1 or 2, then is ACC really that big a deal, and why can it not be incorporated into a Bios.

It makes me want my Kuma back, so I can fiddle with it more.

3.4 should be attainable, but it will start to become very power hungry. You may get some better results trying to overclock the northbridge again.

[ozzy983]

Ah, time for an update.

Having a pretty big winter storm here in NJ, so I have time to mess around further with my overclock. Unfortunately, I haven't been able to get to 3.4 GHZ. Seems this very well maybe the wall for the processor. I had it crash after 6.5 hours of prime @ 3375mhz. So, I figure with a bit of tweaking, I can get that extra 75 mhz.

Before I try that though, I decided to try and go crazy with the NB and memory timings. I thought it might be possible the memory is causing the issue because it's just not fast enough.

Anyway, I did get 2400 mhz NB stable, and that's something I couldn't do on the previous motherboard.



Nice increase in 3dmark scores, and really trying to get it to 13k. Memory scores in sandra also show almost a 20% increase from stock!

On average, 3dmark shows about 6 fps higher on each game test. So, going to pop in RE5 and see if i can't get 60 fps steady with 16x aa. Was able to get about 56 fps avg @ 3ghz.

[jedihobbit]

You're so right Ratatat, there is no cure except for more equipment!! And yes he is making me want to pull my Kuma out of the forsale box to play with!

Just to compare (nice work with the 7750 BTW!!) apples to oranges, I've started playing with my new 790GXE / 945 combo and have found (as it has lock multi) that the more FSB I add the lower I've had to set the HT & NB frequency so as not to go too far above 2000 also it has be running my 1066 4GB at 800 to also limit the cap. Snow here in VA is also making for the good excuse to play more this weekend.......

Then again I do not go anywhere indepth with my oc'ing as you do ozzy!!

[ozzy983]

I just like to be thorough. One thing i hate about being disorganized is i'll try a setting and it won't work, and about 30 mins later, I will try it again, because I never remember which settings I've tried.

think I'm pretty much to the point where i have to trade one component for the next; lower the NB/HT to raise the fsb, or fiddle with this or that.

3350 seems to be the limit; now i need to max the memory out with the lowest latencies.

[jedihobbit]

What is the model number of you muchies??? You're talking about tightening up the settings and was just wondering it you might want to leave it at 5-5-5-18-24-2T but go to 533(1066) and give it a try? It took seveal of the gurus here (including ud) to make me realize that more is gained with a higher GHz than tighter timeings now a days.

[ozzy983]

Memory is only rated @ 800mhz. 1066 will sometimes not even post. Regardless of the voltage i throw at them. It seems in order to get some stability at the higher settings, i need to work the nB up to about 2200.

ht doesn't seen to make a difference one way or another.

even tried the memory running @ 200 mhz, lol, and the pc still dies out at 3.4. odd thing is, it actually seems to be more memory issues than processor. well, just based off the BSod with xx file, memory dump, and then restart.

so, now, going to 16.5x, and slowly walking the fsb up and seeing where the problem begins. acc is an odd setting, and I'm not exactly sure what the + and - do.

priming the system right now, blended test, at 3342, which prior to increasing nb would bsod about 2 minutes in. increased the nb to 2200 (2228) and it's working better. Very odd, usually they recommend to decrease it, which makes me think the IMC on the kuma is rather weak compared to the phenom 2.

[Ratatat]

Ah, time for an update.

Having a pretty big winter storm here in NJ, so I have time to mess around further with my overclock. Unfortunately, I haven't been able to get to 3.4 GHZ. Seems this very well maybe the wall for the processor. I had it crash after 6.5 hours of prime @ 3375mhz. So, I figure with a bit of tweaking, I can get that extra 75 mhz.

Before I try that though, I decided to try and go crazy with the NB and memory timings. I thought it might be possible the memory is causing the issue because it's just not fast enough.

Anyway, I did get 2400 mhz NB stable, and that's something I couldn't do on the previous motherboard.



Nice increase in 3dmark scores, and really trying to get it to 13k. Memory scores in sandra also show almost a 20% increase from stock!

On average, 3dmark shows about 6 fps higher on each game test. So, going to pop in RE5 and see if i can't get 60 fps steady with 16x aa. Was able to get about 56 fps avg @ 3ghz.

That's good. Amazing what a little $60 chip can really do with the right components. 600+ mhz over stock is considered amazing. Anything more you get past 3.3 is just icing on the cake.

What is the model number of you muchies??? You're talking about tightening up the settings and was just wondering it you might want to leave it at 5-5-5-18-24-2T but go to 533(1066) and give it a try? It took seveal of the gurus here (including ud) to make me realize that more is gained with a higher GHz than tighter timeings now a days.

Hmm...this is one of those debates that has raged on ever since the A64 platform has really taken off. To a certain extent, I'm going to agree that memory speed is less important than processor speed.

I'm also going to say that polishing the entire system is more beneficial than a few measly mhz on the CPU. I'd also say, that increasing the NB 200 or 400 mhz is more beneficial than even increasing the CPU, as well.

I think, in the end, the debates continues to rage on because individual use, and application specific instances continue to "prove" neither is better, but each is best in their own right.

[jedihobbit]

I think, in the end, the debates continues to rage on because individual use, and application specific instances continue to "prove" neither is better, but each is best in their own right.

[ozzy983]

ok, tiny bit of theorycrafing here, and these are the results with a slightly lower NB/HT, higher memory freq, and higher processor speed.



As you can see, memory performance is cut in half, 3dmark score is slightly lower, but overall negligible and processor is half and half in sandra.



Little more theorycraft here, tried increasing the FSB, which I got to 240 without having to drive the memory through the roof with increased voltage, and the NB/HT only 1920. Memory is a single stick incase I blew it out. CPU performance is much better this way.

Very interesting indeed.

[Ratatat]

ok, tiny bit of theorycrafing here, and these are the results with a slightly lower NB/HT, higher memory freq, and higher processor speed.



As you can see, memory performance is cut in half, 3dmark score is slightly lower, but overall negligible and processor is half and half in sandra.



Little more theorycraft here, tried increasing the FSB, which I got to 240 without having to drive the memory through the roof with increased voltage, and the NB/HT only 1920. Memory is a single stick incase I blew it out. CPU performance is much better this way.

Very interesting indeed.

This is normal, Oz. I think a lot of people have a few misunderstandings about how all the technology fits together. This is one of the biggest draw backs to the internet, in general. Their is just as much misinformation as there is information.

In any case, increasing the FSB (HTT on A64) systems will give you the best possible numbers. Consider it this way: the 200mhz standard HTT is your base multiple. It's the equivalent of having a road with a 25 mph speed limit and your car can only do 100 mph.

Now, you've went and increased that BASE speed by ~20% or so. Remember, all your numbers, memory, CPU, NB, and HTL are all derived from your HTT.

This is basically the equivalent of having the speed limit increase by 20%, which is 30 mph, but the car still only does 100 mph. Your data now moves along the bus slightly faster.

This is why I said before processor speed isn't everything. There are dozens of factors that can impact performance. Keep in mind, different technology works different ways; this is even true between different A64 platforms.

I, will however, continue to stand by my early point, which is never cripple one component for another.

Example, if you can get 100mhz off the CPU by lowering your ram from 800 to 533 or 400, then don't do it; the overall system is going to be slower. However, if the system is only used for say, video editing, than raw processor power is more important than memory. In most day to day situations, though, the entire system is more important than one component.

[ozzy983]

Well, i haven't fiddled too much today; been working on getting the system rock solid @ 3335. ran prime95 before, and 1 core died out about 3 hours in. Raised the CpuNB a tad because it's running @ 2300 mhz, and it wasn't very high to begin with.

Back to the prime game, running 4 hours straight now, so hopefully problem is solved.

[Poci]

Try OCCT its better stability test then prime.

[jedihobbit]

ozzy I'll let ya be the 7750BE expert and have my fingers crossed for you!

Can't remember if I brought this over to you other tread, but my “for real” OC was 13.5 x 239 for 3.228GHz

And then my usual S & Gs best “bootable” setting………… 13.5 x 247 for 3.3336GHz

Interestingly enough cpu-z is picking up the 1066s as 800s.

Also reading your oc'ing threads has helped me work with my Biostar TA790GXE / PII X4 945 combo.

[ozzy983]

I find that a wide variety of applicatons are needed to test stability; but, in the end, the most important thing is it does what i need it to do, when i need it to do it.

I think 6+ hours of prime is good enough, and looping 3dmark for another 2 hours is fine for stability.

I'd really like to top 3.4 ghz, but I'm thinking its just not gonna happen. i can actually boot windows at 3.5, but seems to need about 1.625V to get the kuma somewhat stable at 3.4. HUGE increase over 1.45 needed at 3.3. Though, I suppose, there could be some setting i'm missing which could stabilize 3.4 at a lower voltage.

ACC doesn't seem to help much at these settings and the processor doesn't seem to like doing minus numbers, and only does up to +4%

jedi,

glad your overclock is going good, wonderful world it is when that happens.

Update: system passed the 3 hour 3dmark loop, so i consider it stable. moving on to test 3375 now with same voltage. i'll run a shorter prime, basically, i set it for custom and change the time to 7 minutes instead of 15 per size.