Positive experience with ACC on Phenom II

[chew*]

I still see temps in AOD and and CoreTemp with it enabled. I'm not sure why everyone else doesn't...

I'm running the latest bios for the board, version 802 - released in January. The release notes for it state "01. Support ACC for AM2+ 45nm CPU". Could something have been done in that bios that makes it more usable? I cannot report anything other than what I've seen on my PC. I have nothing to gain or lose by reporting my findings. I'm just reporting what I'm seeing. As I've mentioned, setting 3.7Ghz on core 3 or all 4 cores in AOD previously resulted in an instant reboot. With ACC enabled, this does not happen and it is 3.7 is completely stable.

--Matt

Maybe you need to play with ram timings and northbridge VID then, that will probably yield the same results when dialed in.

[Steevo]

My 2 cents on ACC.

It is not necesary with PH II, yes I know it enables a core that was disabled because it was "BAD" maybe the NB is weak on it, or maybe it doesn't scale well or maybe it runs really hot with all 4 who knows........

What ACC does is adds voltage to the chip, note the fact when enabled temps dissapear from AOD.......probably either shuts the thermal safeties off or AMD did not want you to know they were increasing voltage etc creating more heat............If it helps your PH II then maybe you need to fine tune your voltages a tad more

The only way ACC did anything for me was on a bios that though my chip was a PH I etc, did not support PH II. It clocked higher but at the cost of performance with a bios not supporting the chip.....

ACC does more than add voltage, it changes the timing of the cache, core and communications bus. The extra pins were tied to the new SB chipset. They have not released why it needs to be tied to the SB, some working close with AMD have mentioned internal cache timings as the biggest issue and those caused by a poor implementation of frequency control. However AMD is very tight lipped about the workings, and the pinout of the Phenom's.

[Jethro]

Where do you get that ACC just adds voltage? I suspect its doing something with timings and or drive strengths personally. Why would it be called advance clock calibration if it only altered volts etc.?

From a Rage 3d article:

After talking with a very smart Friend, we've come up with two possible ways in which ACC might interact with CPUs to produce the results seen since its introduction (it's a one or the other situation, albeit the second proposal would indirectly involve affecting the first aspects as well):

*

Change parameters dealing with asynchronous clock domains
within the chip/clock crossing, by increasing/reducing the number of sync clock cycles - this would correlate nicely with the increased OC’ing headroom aspect, as well as provide anecdotal evidence that indicates higher ACC values induce a slight reduction in performance in something like Prime 95, the core with the highest ACC modifier finishing last, whereas the opposite holds true for negative ACC values, in which case the core with the lowest ACC value finishes first - however, more testing is required to fully verify this.
*

Adjust skew/clock slew/ drive strength (adjusting drive strength affects clock slew anyhow) - this possibility correlates with the fact that "weak" CPUs seem to receive the most benefit, whilst good CPUs get almost none, whilst also explaining why extreme ACC values can cause boot failure, and why high positive values are recommended for high voltages (albeit it's hard to know what it is you're adding X% to)

Be aware that the above are at best slightly educated guesses ... there simply isn't enough data out there to deduce exactly what's going on. For all we know, miniature dwarven craftsmen could travel via the SB-CPU link to tweak the latter - at this point in time, the exact mechanism remains a mystery. However, for most people the only thing that will matter is that it pretty much works.

[mattkosem]

Maybe you need to play with ram timings and northbridge VID then, that will probably yield the same results when dialed in.

It doesn't. I've tried. It even happens with the memory set to 6-6-6-24 at 800mhz and the NB VID set to 1.40v. Believe me, I would employ some other solution if one worked. This is the only thing that has cured the issue for me. My OC on core 3 improves substantially with this change and this change only. If I had a video camera, I would record it and show you, but I don't. This brand new mobo might be faulty, and could be providing weak voltage to core 3. If that is the case, I will know on Tuesday when my new mobo arrives.

--Matt

[Jethro]

http://www.rage3d.com/reviews/mobo/a.../index.php?p=3

[G0ldBr1ck]

Where do you get that ACC just adds voltage? I suspect its doing something with timings and or drive strengths personally. Why would it be called advance clock calibration if it only altered volts etc.?

From a Rage 3d article:

After talking with a very smart Friend, we've come up with two possible ways in which ACC might interact with CPUs to produce the results seen since its introduction (it's a one or the other situation, albeit the second proposal would indirectly involve affecting the first aspects as well):

*

Change parameters dealing with asynchronous clock domains
within the chip/clock crossing, by increasing/reducing the number of sync clock cycles - this would correlate nicely with the increased OC’ing headroom aspect, as well as provide anecdotal evidence that indicates higher ACC values induce a slight reduction in performance in something like Prime 95, the core with the highest ACC modifier finishing last, whereas the opposite holds true for negative ACC values, in which case the core with the lowest ACC value finishes first - however, more testing is required to fully verify this.
*

Adjust skew/clock slew/ drive strength (adjusting drive strength affects clock slew anyhow) - this possibility correlates with the fact that "weak" CPUs seem to receive the most benefit, whilst good CPUs get almost none, whilst also explaining why extreme ACC values can cause boot failure, and why high positive values are recommended for high voltages (albeit it's hard to know what it is you're adding X% to)

Be aware that the above are at best slightly educated guesses ... there simply isn't enough data out there to deduce exactly what's going on. For all we know, miniature dwarven craftsmen could travel via the SB-CPU link to tweak the latter - at this point in time, the exact mechanism remains a mystery. However, for most people the only thing that will matter is that it pretty much works.

He didnt mean that voltage is all it dose. Just that thats all that Phenom II benefits from. All the other functions of ACC were designed for Phenom I, Phenom II does not benefit from the Clock Calibration.

[Jethro]

Some Phenom II triple core's gaining a core with ACC on auto might just be considered an benifit i think Is it possible that this feature could come into play at some point even if it isnt a factor now? Its obviously helping Matt!!

Anything from AMD stating that it will NOT effect Phenom II's?

[chew*]

To each his own I call it placebo effect.......Your sources are people taking educated guesses.

My sources are a little more direct than that and I quote " ACC offers no benefit whatsoever with Phenom II "

It is not designed for phenom II it is not made to work with phenom II etc etc, so the only thing that it is applying is voltage......

NB is directly tied to the MMC I have tested this, at X NB vid volts my memory would not pass memtest, In test B with more than X NB vid volts it passed fine and was cinebench stable.......

It's possible that the mobo manufactureres are trying to code it in however and you guys are seeing a fluke......however Phenom II does not support it.

1.40 NB vid volts is not trying it.........taking a cpu to the max voltage the board allows is trying it....

I would love to see direct comparison benches of performance at same clocks with it off and on.

I have tested with it on BTW and although the thermal monitor went blank my K probe does not lie, a +2C jump on the edge of the IHS.

since when did gaining a bum core become benefit? If cpu with only 3 cores can do 4.3 gig and the same cpu with 4 cores can only do 3.3 gig its not much of a benefit. same scenario if cpu can do 2800-3000 NB and can only do 2400 with the 4th core enabled, wheres the benefit.

[Jethro]

Whether one percieves it as a benefit or not ACC is certainly having an effect on Phenom II's.

[chew*]

Whether one percieves it as a benefit or not ACC is certainly having an effect on Phenom II's.

Well unless we see comparison tests with and without thats sort of an open mouth statement.

I can clock higher......so its a benefit....hell I can clock 1 core higher than the others..........won't do me much good performance wise in Wprime though... see my point....

Bench comparisons at same speed enabled and disabled will determine if its truly a benefit.

Yes its definitely having an effect, lol free cores for everyone....

[Jethro]

As im reading some folks are having very positive results enabling cores on there phenom II's. Why are you so cynical>? Its all in good fun. Just chattin about it a bit man lighten up.

[Rolle2k]

Even though i also have read that ACC won't work on Phenom II i now probably have to try it, maybe tomorrow.. i have also trouble with one core.. i just have to test a little more, and if it doesnt help then i have to try this way.

[mattkosem]

Well unless we see comparison tests with and without thats sort of an open mouth statement.

I can clock higher......so its a benefit....hell I can clock 1 core higher than the others..........won't do me much good performance wise in Wprime though... see my point....

Bench comparisons at same speed enabled and disabled will determine if its truly a benefit.

Yes its definitely having an effect, lol free cores for everyone....

Well, if you can come up with a reasonable, clear, and concise plan that you would like me to execute, I'd be happy to run it. I'm only running ACC on one core. Enabling it across all cores kills the OC on cores that are otherwise working fine. I can run side by side single threaded tests chosen by you on the core with ACC and on one without. Make me a list, and I'll run through it. As you can see in the prime screenshot that I posted, all 4 cores were at the same exact point in the test after over 15 minutes and they all reached that step at the same time. I have nothing to prove, but will humor your curiosity if that's what you crave.

--Matt

[mattkosem]

Here's a quick super pi 1M test. The one on the left is from core 3 and the one on the right is from core 1. The tests were run one right after the other. They're well within a reasonable distance of eachother. The core with ACC even took a small, but negligible lead.

--Matt

[mattkosem]

And here's 32M. Only a .07% difference.

--Matt

[chew*]

PI is to inconsistent, my second PI run is almost always faster than my first.

You would need to bench maybe 3d, and multicored....cinebench gives preety consistent results when run repeatedly and is highly cpu bound.

Try running that with ACC off an on 3 times each and we will average results.

Not trying to be cynical just posting the facts that have been relayed to me.

[mattkosem]

PI is to inconsistent, my second PI run is almost always faster than my first.

You would need to bench maybe 3d, and multicored....cinebench gives preety consistent results when run repeatedly and is highly cpu bound.

Try running that with ACC off an on 3 times each and we will average results.

Not trying to be cynical just posting the facts that have been relayed to me.

Your second pi 32M run is consistently faster than the first by a consequential amount? On 1M I can understand that, but 32M runs over a long enough period of time that things like that are avoided. I've seen variations of as much as .08s between pi 1m runs on faster intel cpus, but never variations of that magnitude on 32M runs.

As I stated, I can only run single threaded tests for you to test this. ACC is only enabled on one core, and enabling it on the others at the same level causes the other cores to be unable to clock as well. If you want to see cinebench tests, I can run it once or twice with affinity set to the core with ACC and a again on a core without.

That stated, I'm still sticking with pi32M being a consistent enough bench to show these results. If you didn't miss the 32M benches that I posted and would still like to see cinebench on a single core, please specify the version that you would like to have run and I can run it.

--Matt

[chew*]

matt just run cinebench, render all cpu's with acc on that one core, then run it without acc on, if it won't pass at that speed just drop the clock down for both comparisons.

I'd like to see this because I think the difference will be most noticeable when all the cores are synced together so to speak, etc with acc on I do not think they will work in harmony as well.....

PI no matter what it is is way to inconsistent, I get diff times every single time I run it, cinebench however scores identical repeatedly and is influenced by cpu speed only, ram clocks etc do not benefit it.....

[mattkosem]

Ok. I'll have to run it at 3.5Ghz then. I'll kick Cinebench R10 off with both configurations when I get home.

--Matt

[mattkosem]

Here you go. 14767 with ACC enabled at +2% on one core, 14467 with it disabled completely.

--Matt

[chew*]

very interesting, thx matt.

[Rolle2k]

Funny thing.. my "bad" core always hanged the computer at around 3,54GHz, or prime died under running so i had to restart it.. and now i have enabled acc +2% on that core.. haven't run more than 1h 30min at same voltage, but so far it seems stable. Normally it would have errored already.. i'll post later when i have primed for longer time!

[gOtVoltage]

Funny thing.. my "bad" core always hanged the computer at around 3,54GHz, or prime died under running so i had to restart it.. and now i have enabled acc +2% on that core.. haven't run more than 1h 30min at same voltage, but so far it seems stable. Normally it would have errored already.. i'll post later when i have primed for longer time!

I knew ACC was good for more than a High oc.

If anything it brings stability for CPU's that seem to have uneven power distribution on all 4cores. Thus allowing you to tweak the Volts to compensate for stability.

Even ACC dosnt allow me Higher Overclocks vs Not using it on same Mobo . But it does Come in handy for supplying cleaner power across all 4cores when tweaked right.

[mattkosem]

Just wanted to report back. I'm running the M4A79 Deluxe now. I don't instantly reboot at 3.7 without ACC anymore, but I think that's due to the flat 12C drop in temps that the water cooling provided. Prime blend reboots the machine within 2 iterations at those settings. Setting the +2 on the problematic core has, once again, corrected that issue. Blend running 1024k FFTs pushes me to a max of 40C at 3.7Ghz with the NB at 2400. I'll see what I can do about pushing those numbers up further when I have some more time.

--Matt

[Lastviking]

With my Phenom II 940BE and Dfi 790FXB i first need to press reset after enabled acc and then it works to boot in to windows but i lose some stabilty with (acc enabled) becuse i geting blue screen next time i restart windows.

Maybe its the board or the cpu.