What does it do and how does it really work. Is it helpful in overclocking at all?
:shrug::shrug:
Printable View
What does it do and how does it really work. Is it helpful in overclocking at all?
:shrug::shrug:
Magic, I don't know and yes.
Any idea how to use it?Quote:
Originally Posted by MrMojoZ
Nobody really knows how it works. The leading guess is that it adjusts running parameter tolerances. Yes, it's generally helpful in overclocking so long as it is properly implemented by a motherboard and its BIOS.
How to use it? You get a range of positive and negative values. You tweak the values for each core until you find the sweet spot that allows the greatest stable overclock--too much and too little are equally bad. Positive values help OC.
It works by taking the voltage and breaking the molecules down into much smaller particles before injecting them into the CPU's intake bus. This makes the CPU more efficient so it can burn more Voltage easier.
(Ooops. Too many car analogies for me.)
Many people believe it is just a form of clock skewing.
Basically you turn it on... and it works. It can be fine tuned also as Particle mentions. But it can easily add 300Mhz to an overclock.
al it is is a clock skew, as mentioned above,
nothing that boards 10 years ago werent using, just re hashed for the masses in 2008,
like everything i suppose,
and yes it works, but only on CPU's with week tolerances lol
Needs more angst.Quote:
Originally Posted by soundood
http://en.wikipedia.org/wiki/Clock_skew
I wonder why AMD was being so secretive about this, refusing to explain how it works... Could this maybe also benefit intel cpu's?
Somehow I doubt it takes 6 pins to achieve clock skewing. Gotta be something more to it.
Actually the fact that they released it and are still secretive....Quote:
Originally Posted by Miss Banana
That makes me wonder if this is only half of the whole. Is the SB750 going to do even MORE with the Deneb??
(And apparently many Intel motherboards DO have clock skew settings.)
repoted for flame baitingQuote:
Originally Posted by MrMojoZ
i would say differant, six pins sounds just right, to enable some type of skewQuote:
Originally Posted by dirtyd
The 6 pins on the CPU previously "unused" are hocked up to the sb750.
I've also heard something about JTAG interface.
Setting wise seems like auto works best for my 9600be.
My Asus M3a with SB600 has options for "AI clock screwing" or something
like that:p: and the most I could get with Win xp64 was 2500mhz
The same cpu (9600be) with the same OS (x64) is now running ~2.9ghz on
the DFI 790gx so I would say it does work.
My 9850be got killed either by the board (above) with ACC or by AOD - maybe both (conspiracy?) I don't know.
All DFI boards that i know of have the DOS Skew control settings, but all that does is handle EMI interference generated by noise from voltage going through the board. Way way back when in the before pentium era that setting actually did something. With better filtered power supplies and better manufacturing process for motherboards, there isnt a need for this setting any more. I highly doubt that the ACC aforementioned is this skew control. Hooking pins up to a south bridge, to me that sounds more like bull since the south bridge functions as the link to the lower buses on the board the pci slots, pcie slots, then the north bridge is the traffic cop that directs things from the upper buses and built in connectors to the back of the board. With AMD the memory controller is on the chip so the NB all it does is regulate traffic of data from inputs to the cpu and outputs to the SB chipset.
I really dont see how hooking up 6 unused pins to a south bridge chipset did anything. I cant blame AMD for wanting to keep this newfound thing quiet, but maybe it is something that is enabled in the bios and they tied it to boards with the SB750 chipset to sell more motherboards. I think it is great that ACC gives good performance boosts in overclocks, but i think they are feeding a load of crap to the general public that has no idea how the internals of a computer works. Maybe later on they will release and disclose what exactly is the real cause of making the boards with SB750s oc so much better.
Must do it differently in the great white north!:rofl:Quote:
My Asus M3a with SB600 has options for "AI clock screwing" or something
like that
Hmmm I am inclining to share the same notion.:shrug:Quote:
Originally Posted by daseto
When I was trying to make the failing core stable, I increased the ACC to +2 for that core and it worked perfectly BUT then the next core would start to fail when it was not failing before. For instance, my ACC is set to 0,0,0,0. During overclocking, core 1 kept failing. I then set ACC values of 0,2,0,0 BUT THEN, core 2 started failing instead. The next set of values that seemed logical for me to try were 0,2,2,0 = Core 2 failed. I am going around in circles 0,2,4,0 = core 2 failed. 0,4,2,0 = core 2 failed again. I am lost.
KEY:
ACC values corresponding to Core 0, Core 1, Core 2, Core 3 are 0,0,0,0 respectively for the above example.
Try raising them in synchronized steps to start with, such as 0000 to 1111 to 2222.
Also, I might have missed this, but what motherboard do you have? I know that at the very least, the ASUS M3A78-T does NOT have properly functional ACC while some other SB750 boards do. Perhaps yours falls into that list of problematic ACC implementations.
You don't want to know :rofl:Quote:
Originally Posted by RiverRicer
Anyone else heard that Nvidia is going to support ACC in the "near" future?
Read it on some site, that collects a bunch of news bits together - forget the name:confused:
Why would AMD not be secretive about it? It's not like Intel is sending their
HighK Metal Gate Hafnium etc. formula to anyone interested.
I set the DQS clock sqew to 255 on my NF4 expert,it supposed to help with high HTT/Ram speeds and afaik the "traditional clock sqewing" is just for that - to help deal with "high speed" (ddr1 mostly) ram and cpu intimate relations.;)
It also has a page full of "spread spectrums" etc, ....don't think it's related.
Anyhow looks like the ACC "mystery" remains - as nobody knows exactly
what and how, just why;) and hey - why not:up:
edit*
For the guys saying that 6 pins is "bs"- why does ACC only work for Phenoms and not X2's ?
Well, leaving it on AUTO does help OCing but I want to maximise its use. Perhaps it could be a useful tool. It seems so because the core which was failing for me consistently wasn't failing anymore when I ACCed it by 2%.:clap: But then, another one starts failing :shrug:
Leaving it on AUTO does require less vCore to sustain the OC though.
I was not aware that it did not function properly on that board. Can you provide a link with details as I am curious now. I have seen others with good results using it so i am curious as to what the problem may be.Quote:
Originally Posted by Particle
Well if all ACC is is a form of skew control then it is possible to tie the bios of the mobos to only enable certain functions with certain chips plugged in. Not all boards have skew controls in the bios that you can change. DFI is one that has had it in the past, dont know if they have that on a SB600 board, might be fun for someone that has one to let us know and fiddle with it and see if they can get similar results as on a sb750 just by messing with the skew control.Quote:
Originally Posted by SocketMan
As I recall, the K8s are missing (2 of the?) pins required to do ACC.Quote:
Originally Posted by SocketMan
Are you sure that you aren't thinking about the M3A79-T? That one does work wonderfully while the 8 (790GX) does not.Quote:
Originally Posted by Ben_Brown
Welcome to the ever-growing club of AMDers that are trying to tweak ACC. My experience is identical to your own, Sandeep. Try to help the weak core and...boom...there goes a different core. Without the ingredients for the ACC 'secret sauce', we'll never get a handle on it. I gave up and set the damn thing back to 'auto' after two evenings of repeated attempts to find some sort of consistency with various settings. Consistant it ain't!:down:Quote:
I am lost.
Ah yes, over site on my part. Thanks for clearing that up. :up:Quote:
Originally Posted by Particle
be carefull on this forum, with comments like that ...lol, youll be an outcast before you can say, intel what?Quote:
Originally Posted by daseto
lol, i totaly agree with you, some of the rubbish that AMD has been spewing is beyond me, the whole release of boards with the SB600 chipsets, to then release the exact same boards with a SB750 chipset is pure robbery.
in fact i would suggest that they would be liable for recycle issues from varios councils and law makers, purely in the fact the didnt offer an 'upgrade' option, like BFG or EVGA, to comply with ROHS standards
but hey i am hated here so my opinion will mean nothing, just another intel fanboy with an AMD rig ehh?
About the 6-pin link to the SB750, and the details you ,mention above regarding the NB/IMC/SB BUS controls etc....Quote:
Originally Posted by daseto
The one thing that is significantly different between K8 and K10 is that the clock gen for the K8's was located in the CPU NB, and there were initial problems with the K8 X2's which AMD released the Dual Core Optimiser fix for...
But on K10 (Phenom) the Clock Gen has been removed from the CPU NB and is now situated in the motherboard NB for 7xx series chipsets...
One of my theories about ACC is that its a combination of Clock Skew tolerance tweaking (i.e. the +/- values for the cores) ,and a new clock gen that has been incorporated into the SB750 which uses a direct link to the 6-pins on the CPU to negotiate with the cores....
I have nothing solid to base this on other than the fact that its been reported for some time that there were issues with SB600 and also SB700 internal clock gens.... so the fixed clock gen in SB750 in combination with the tweaking of the clock skew tolerances, allows for increased OC for those CPU's that were previously shown to have one or more weak cores...!!!
I think that ACC has been provided by AMD as a gesture of goodwill if you like as a workaround for customers who have bought Black Editin Phenoms that were sold with the intention of OC'd (hence the unlocked multi's) but that proved to be poor OC'ers as a result of the weak cores....
This is my theory anyway...!!!!
I have no
soundood, I don't think it's so much what you say as how you say it.
Abrasive, challenging--like waving a red towel in the middle of a bull pit:
"lol, i totaly agree with you, some of the rubbish that AMD has been spewing is beyond me, the whole release of boards with the SB600 chipsets, to then release the exact same boards with a SB750 chipset is pure robbery."
Same thing only debate-inviting instead of provocational:
"I completely agree with you. AMD's current explanation of how ACC works smells fishy to me. They released motherboards with the SB600 chipset and then quickly re-released with the SB750. That strikes me as bad business and insulting to the customers."
I do disagree with that statement though, unless you aren't really serious about SB600 to you looking like a scam. That southbridge was their leading model until the SB7x0 chips were finished up, what--two years later? It doesn't seem like a crime to me that they're doing development cycles and releasing updated technology as they invent it.
I think I remember reading that part of those 6 pins is used to enable higher multpliers on the phenom BE's that werent available on previous chipsets before?
Is this true?
As far as I'm concerned ACC doesn't do anything.
lol they all have multiplier that go up to 25 on BE'sQuote:
Originally Posted by GenTarkin
I recall those 6 pins also use a pci-express lane or something.
It's an ASUS M3A78-T issue.Quote:
Originally Posted by Jimmer411
I thought it was to provide an additional Voltage/power to the CPU.Quote:
Originally Posted by demonkevy666
I also remember (sort of) a discussion ( here at XS at the start of this year)
about the "future" boards,now known as the 790xx +SB750, where
a newer clock generator was going to be added to the south bridge.
edit*
Found a pick
http://www.digit-life.com/news.html?10/63/37
Hi guys
Well i did some ACC testing to night , my 9850BE 3 of core are good running cores but core number 4 runs like crap . Without the ACC 2.7ghz was all i could get with my old board . Now with the a79a-s ACC set at +2% i can get 3.0 ghz 24/7 trouble free but as soon as i try to clock up any higher i drop core 4 when running prime .So i put ACC to the real test . I bump the cpu up to 3.105 ghz. set the volts for the cpu to 1.4v to give ACC something to work with than i started. I kept the 3 good running cores at +2% than i stared bumping core 4 up +4% core 4 crash as soon as prime started bump it +6% same thing , bump +8% same thing , bump it to +10% bam! prime up and running on all 4 core i let it run throught 3 test passes on the 3 test change over to AOD stability test run that for 30 min than i do a 3Dmark-06 run and then jump right into a vantage run pass on everything NP so now it look like the 4 core are all blance out now , Tomarow night i see what happens if i try to bump it up to 3.2ghz . The hole point being if one of your cores are crashing before the other try cranking in the ACC on the single bad core to see if you can bring it in line with the others. ACC go up to +12% my thinking is if they did want me to use it they shouldn't have put it in the bios . Just something to think about .
Follow up so as i trun the power off on my pc and look at it i say to my self Hummmm it a Phenom cpu do you think it will restart again ???
hell no even when it pass all the other stuff earlyer even when i get prime to right on the core 4 the dran thing will not repost so even when
using ACC getting the cpu blance out still not a sure thing when going for the higher over clocks . I had to declock back to 3.0 ghz to get it
to repost . Well it getting late need some sleep nite all.
Anyone else having success with ACC? Please share ur experiences.:(
I've done a bunch of ACC testing using the 4-thread version of Prime. My observation is that the 'weak' core is actually completing the Prime iterations faster than the 'good' cores. In the case of my 9950, core 2 (third core) seems to get there faster, but also crashes first when pushed. Using that logic, I reduced ACC on that core, ie. I'm currently running 2-2-0-2. However, it doesn't seem to make a big difference from 0-0-0-0 or 4-4-4-4. What I have found is that a setting on a particular core >±4 from any other core causes boot-up problems. For example, if I set ACC at 6-6-0-6, no boot.
Since all my testing was done using bios, not AOD, the no-boot scenario affects my mental stability! :D My max, rock-solid o/c so far is 3.25@1.4vcore, 250fsb.......not very impressive, but a solid, Prime blend stable increase over stock speed. I've run the chip at 3.5 with about 1.55vcore, but couldn't find stability with Prime or OCCT.
In my testing of my bad core i can running 3.0ghz with ACC set to +2% all time no reboot trouble run prime np stable all the time . That a .300 ghz pick up from no ACC, to get 3.1 ghz i than have to bump ACC to +10% on core 4 it will run stable i can run prime np play game it all good . Now when i trun it off it will not reboot i back ACC on core 4 to +2% declock the CPU run it for a bit than i can once again reclock the ACC back up again using +10% acc on core 4 .It like the hole system has to reset it self for it to work again with ACC setting set high or it get hot some how with a high setting just not sure .
@River: What happens when you keep the delta lower, like 6-6-4-6 or 8-8-6-8?
I've tried some higher + settings on the cores while keeping the deltas close, but my chip doesn't seem to like anything over +4 on any one core. ACC settings are highly chip-dependant and it would be a whole lot eaisier if AMD would just come out a say what ACC does, or at least how to adjust it for various core scenarios.:mad:Quote:
Originally Posted by Particle
On the other hand, I'm happy with a rock solid +650 o/c. Without ACC it would have been a lot tougher.:D
http://67.90.82.13/forums/showpost.p...1&postcount=53Quote:
Originally Posted by RiverRicer
kazgirl's post is spot on.
Just one minor niggle - K10 also had an ondie clockgen, it seems. Only replaced by SB750 when possible.
Oh, and it seems that somebody at Coolaler (again) is reviewing the new MSI 780G + SB750 board and an ACC enabled BIOS.
Hey guys keep posting your ACC experiences. Hope we can unravel this mysterious ACC:eek: