Tuning ACC

[Daveburt714]

I've mentioned this before, over 10 Months ago on OCX...

Alot of people I really respect kind of shot it down, but I haven't given up on this idea, this technique has helped me stabilize my OC on a number of boards...

I don't run extreme cooling so this is based on Highend AIR/Water.

No Guarantees, but I've seen alot of people asking about ACC settings and although this may not be the perfect answer it's helped me....

This involves using "wPrime"... Here's what I've been doing.

1) Push your system pretty close to the edge (but fairly stable)
2) Run wPrime and watch the sequence of core completions (should be 1234 x 4)
3) If the cores get out of sequence adjust ACC to keep them straight
4) A lower ACC will make a core finish faster while a higher ACC will slow them down.

If wPrime crashes it doesn't really shutdown and will run until it errors out, then you can count how many cycles each core completed....

EX: If core1 completed 5 cycles while all the others only did 4 try raising core1 by 1 step on ACC, or lowering all the others (you'll have to play with it, try to keep them close to zero)...

Theres no doubt there are ALOT of variables here, and you may never get a perfect 1234x4 run everytime, but you can at least tune it to were it's close and that has always helped me with stability!

We're all Overclockers here, give it a shot...
At least for me it's a whole lot better than just setting random ACC numbers...

[Smartidiot89]

So turning up you meen +% and turning down -% ?

Absolutely worth trying, anything that takes my CPU above the magical 3,7GHz stability point.

[Daveburt714]

Yep... Raising means plus on ACC and Lowering means minus

If for example, Core2 keeps moving up the list, raise it's ACC..
If it was @ zero make it +2 that should put it back inline.

[nguyenquochuy]

Thank Daveburt714, it's really worth to try as I'm having some trouble trying to make my unlock 550BE stable at 3.85GHz

[Gener_AL (UK)]

Nice post, Just the infomation and confirmation I have been looking for on ACC.
I recently purchased a 955BE (coming form a 9850BE) and have managed to get the 955 stable
For gaming and benching at 225x17 NBx12 HTx10 DDR@1066
3825/2700/2250/1200 (1.45/1.25/1.2/2.1 v)
And was looking for for a way to get more stability with higher overclocks
The two remaining things in bios that are a bit of an odd mystery for tweaking for me have been the skew and ACC for the quads i have tested.
every quad I've tested had different skew and each quad 9950 9850 940be and this 955 I have right now have had different ACC settings when set to auto.
I have a feeling that ACC & skew are linked to the clock multi for the CPU

Note following tests on M3A79T (790FX)1102 bios


Stock Clocks all options auto

e.g
940BE skew auto: channel a=300 delay b = 450 delay
955BE skew auto: channel a=150 delay b =300 delay
9850BE skew auto: channel a & b= 300 advanced (from what i recall)
Tested On UD4P (790x) as only board i have tested that shows you values for auto setting
9850 ACC: auto = +2
9950 ACC: auto = -2

Questions I have to ask are what CPU are you using dave, and rest of posters in this thread, and what are the default ACC setting & skew setting using Auto option.

Would be nice to build up a profile for these settings with phenom's
It may prove to be nothing but theres a link in the settings & product there somewhere I am sure of it.
sorry to hijack thread a little , but are not skew and ACC linked in thoery or at least provide a similar effect?

[jlacy76]

Thanks this is news we can use. I can pass 32m fine but 1024 BSOD or hangs all cores.

[Tom128]

I will probably give this a try just to play! Are you saying this would effect Phenom 2's as well?

[keithlm]

Are you saying this would effect Phenom 2's as well?

It seems that awhile back reviewer posted a "fact" about ACC not being needed on PhII's and everybody else started quoting that "fact" as being true. Thus another internet myth was born.

Probably because "not needed" and "not useable" aren't the same thing.

[Daveburt714]

It seems that awhile back reviewer posted a "fact" about ACC not being needed on PhII's and everybody else started quoting that "fact" as being true. Thus another internet myth was born.

Probably because "not needed" and "not useable" aren't the same thing.

Well said keithlm! I've never really tried (or seen much need) for using ACC on my PHII 940, maybe I should play with it...

But, unlocking cores on X2/X3's and tweaking using this method has shown some benefits for me. It may not be able to get you more, but it does seem to settle down what you have.

At least on most boards you need to activate ACC to unlock cores anyway so I applied the same technique I used on the 98/9950's...

These "Old Timey" skills did seem to help with OC'ing my 720 and 550.
I would think the fact that they were downgraded to X2/X3 would make ACC more usefull, and personally, I've found that to be the case...

I'm just curious, has anyone else tried this, and did you see any improvements?

[ELItheICEman]

I played with it last night. When running as a quad, my 550BE will not boot at 3.9 with all ACC set at -2% - but using this method, I set to -4%, -2%, -8%, -4% and was able to boot and bench at 3.9. Still not stable, but I'll keep playing with it.

[shaolin95]

Interesting, maybe I can get a bit more out of my 720.....

[demonkevy666]

I get stable at 4.0ghz on 1.535 for core 1 core 2 and then core 3 can't do any thing core4 sometimes errors.

[Daveburt714]

I played with it last night. When running as a quad, my 550BE will not boot at 3.9 with all ACC set at -2% - but using this method, I set to -4%, -2%, -8%, -4% and was able to boot and bench at 3.9. Still not stable, but I'll keep playing with it.

Glad to see someone at least found it somewhat usefull..

It may just be my chips, but I've had trouble when I get into ACC numbers over +/-6%... The closer I can keep them to 0% has always produced the best results for me personally...

Instead of those settings have you tried: 0%, +2%, -4%, 0%.... ?
Same ratio, but closer to 0%.

If you don't mind, try that and see if you can still boot/bench @ 3.9Ghz.

[Smartidiot89]

I played with it last night. When running as a quad, my 550BE will not boot at 3.9 with all ACC set at -2% - but using this method, I set to -4%, -2%, -8%, -4% and was able to boot and bench at 3.9. Still not stable, but I'll keep playing with it.

Nice results man

ACC might not be of use to most but for people with unlocked cores "it makes sense" ? Waiting for more results

[ELItheICEman]

I'll see what I can do. Got carried away this weekend with things to do, but hopefully tonight I'll have a few minutes to sit down and fiddle some more. I think this thing's got 4.0 in it somewhere, even if it doesn't do more than boot and validate.

[Smartidiot89]

I am gonna pop in here again.

You guys seems to believe that ACC can help unlocked CPUs to become more stable... I will be getting a 550BE from a friend which can be unlocked to 4 cores but the 4th is unstable (3rd seems to be benchstable atleast). Could this be something ACC potentially could fix for 24/7 stability or benchstability across 4 cores?

[chinook]

This is definetely worth giving a shot. But what about prime95? I've always seen one core lag behind on iterations and i was wondering "what the hell?" but now it seems a bit more clear that acc could help on that. have you tried this with prime95?

[Trynity]

ive tried to get the cores back in order, but with no luck

Whats the setting you start with? 0% 0% 0% 0%

ACC Disabled?

When ACC is disbled my order is 1 3 2 4 but if i push core 2 to get it back in range i get 4 2 3 1 .....

thanks for help

[ganastasiou]

try to slow down other cores,then go a step up at all cores when u find the sweet spot.

[mav2000]

Tried thia dn I got 1,3,2,4 so went into bios and set everything to zero except core 2 which i set to -2. Then tried again and got it to 1,2,3,4. But when I ran it again it now hows 1,3,4,2 and keeps changing on every run....am running 32m. Now it shows 1,4,3,2....

[Daveburt714]

Whats the setting you start with? 0% 0% 0% 0%

ACC Disabled?

When ACC is disbled my order is 1 3 2 4 but if i push core 2 to get it back in range i get 4 2 3 1 .....

I would start with whatever "auto" sets your ACC too, then go to "Per Core" and set all cores at that setting to start with...

Of course ACC must be Enabled...

Tried thia dn I got 1,3,2,4 so went into bios and set everything to zero except core 2 which i set to -2. Then tried again and got it to 1,2,3,4. But when I ran it again it now hows 1,3,4,2 and keeps changing on every run....am running 32m. Now it shows 1,4,3,2....

You can try +2,-2,0,0. But there are no guarantees that will work either...

Guys, go back and read the 1st post again...
Your probably never going to get a perfect run where all cores stay completely in sync, there are just too many variables. It takes time to figure out where you'll get the best results, but I've never managed to keep them all in perfect sync!

In fact I've noticed the sequence change when I change my OC and had to tweak ACC all over again.
This was never intended to be the "Holy Grail" of ACC tweaking, just a better option than random guessing...

Like I said it's helped me but it does take alot if fiddling and ofcourse YMMV...

[kink]

I doubt the usefulness of this method. wPrime simply starts one thread per core and probably sets the affinity mask so that each thread runs only on one dedicated core. The problem is that there are always loads of other threads running (other applications, services, OS itself), which are scheduled to execute on some available core; another active thread may therefore interrupt the second wPrime thread, then the 4th one, afterwards the 1st one etc. There has been quite a change to the thread scheduler in Win7, so this whole thing additionally depends upon your OS too. I did a 1024M test on Win7 and the order wasn't stable at all, no surprise there...

[Trynity]

Of course ACC must be Enabled...

yes .... my failure, i mean ACC set per Core 0 - 0 - 0 - 0

not disabled @ all

regards

[Daveburt714]

I doubt the usefulness of this method. wPrime simply starts one thread per core and probably sets the affinity mask so that each thread runs only on one dedicated core. The problem is that there are always loads of other threads running (other applications, services, OS itself), which are scheduled to execute on some available core; another active thread may therefore interrupt the second wPrime thread, then the 4th one, afterwards the 1st one etc. There has been quite a change to the thread scheduler in Win7, so this whole thing additionally depends upon your OS too. I did a 1024M test on Win7 and the order wasn't stable at all, no surprise there...

kink: I agree with you on 99% of what you said, I don't doubt all the references to Affinity, services, apps, OS and interupts...
Which is why I mentioned all the variables...

No matter how much you tweak ACC using this method your NEVER going to get a perfect sequence run using WPrime or P95 for that matter!

The 1% I dissagree with you on is the usefulness of this method...
Although you'll never get it perfect, keeping things closer to inline has helped me settle things down a number of times.

I just thought I would throw it out there, if people find it usefull, great, if not, it seems to work for me... So I'll keep using it...

It is what it is...