Undervolting my AMD X2 BE-2300 (Brisbane 45w TDP) w/RMClock

[davidh44]

AMD X2 BE-2300 1.9Ghz (Brisbane 45w TDP)
- Overclocked to 2.03Ghz (9.5 x 214Mhz)
- 1.25v stock, 1.0v startup

Here are my results undervolting with RMClock:
5.0x 0.800v
6.0x 0.800v
7.0x 0.825v
8.0x 0.925v
9.0x 0.925v
9.5x 0.950v

And some questions:
1. At 9.5x @ 0.950v with RMClock, am I actually using less power than the default AMD CnQ minimal setting of 5.0x @ 1.0v? (i.e. how much of an effect does multiplier have on power consumption?)

2. Is it true I need to stock with whole-integer multipliers (e.g. 8.0x instead of 8.5x) for stability even though the BE-2300 uses a 9.5x multiplier?

3. Given the voltages above, how many p-states would you utilize?

4. I left all other settings in RMClock at default. Are there any other settings I should change?

[juamez]

1. I think the power used at 9.5x @ 0.95V settings are the power used at 5.0x @ 1.0V times (9.5/5.0) times (0.95/1.00V)², so that would be 1.71475 times that power, IF we take into account that the power scales perfectly linear with the clock and the power scales squared with voltage. I don't think that's 100% the case, but maybe somebody can come up with a better calculation.

[Esben]

1)

Usually when I calculate comparisons between speeds and voltages, I first calculate reference power consumption, which I call PN (Power Number).

The formula, which roughly corresponds to reality is:

PN = frequency*voltage^2

So your first PN is:

PN1 = 1 * 1^2 = 1
PN2 = 2.03 * 0.95^2 = 1.83

Thus your computer consumes 83 % more power, undervolted, at full speed than default C'n'Q. The only effect the multiplier has, is it controls the resulting CPU speed.

2)

I've never heard about requiring whole-integer multipliers for stability. Though specific whole-integer multipliers can give a memory speed closer to the rated memory speed.

3)

I'd choose 5x, 7.5x, and 9.5x, but defining more will not hurt.

4)

If your system is stable, then no other settings need to be changed.

[davidh44]

Thanks for the replies.

I've found that although each P-state by itself can be very stable, using the Performance-on-demand profile with multiple P-states causes stability issues. It's probably due to the frequent switching between different frequencies and voltages? I'm going to try increasing the voltages across the board to see if that helps.

[dragmor]

Thanks for the replies.

I've found that although each P-state by itself can be very stable, using the Performance-on-demand profile with multiple P-states causes stability issues. It's probably due to the frequent switching between different frequencies and voltages? I'm going to try increasing the voltages across the board to see if that helps.

I don't use RMClock but see if you can make sure it ups the voltage before it ups the CPU speed, doing both at the same time maybe whats causing your problems.

[astrayan]

Thanks for the Brisbane info. I presume this also holds for the 4850e etc. Why AMD have to come up with new nomenclature for each chip, it shows how far they think ahead.

I only use 2 p-states. I think anything else is abstruse, and unfathomable, given the way you will never have something that consumes 60% of your CPU for a sustained period. It's usually all or nothing.

In my experience, rather than in my theory, voltage has a separate effect on idle consumption, which I suppose is additive.

64W idle 1GHz .875V
83W idle 2Ghz 1.45V
81W idle 1Ghz 1.45V

[davidh44]

I only use 2 p-states. I think anything else is abstruse, and unfathomable, given the way you will never have something that consumes 60% of your CPU for a sustained period. It's usually all or nothing.

I noticed that when I had multiple p-states, it never did stay long enough at the intermediate p-states to seem useful. And RMClock seems less stable the more p-states I add.

With just 2 p-states, I find that I'm still getting random reboots and BSOD, and am slowly increasing the voltages to see when it'll go away. Each p-state by itself runs just fine.

In my experience, rather than in my theory, voltage has a separate effect on idle consumption, which I suppose is additive.

64W idle 1GHz .875V
83W idle 2Ghz 1.45V
81W idle 1Ghz 1.45V

Interesting, thanks! Makes me wonder whether I should just run it full speed @ 0.95v instead of doing the 0.925v low-speed / 1.0v hi-speed p-state combination I'm currently using.

[davidh44]

Thanks for the replies.

I've found that although each P-state by itself can be very stable, using the Performance-on-demand profile with multiple P-states causes stability issues. It's probably due to the frequent switching between different frequencies and voltages? I'm going to try increasing the voltages across the board to see if that helps.

I don't use RMClock but see if you can make sure it ups the voltage before it ups the CPU speed, doing both at the same time maybe whats causing your problems.

Good idea. Maybe definiing an intermediate p-state with higher voltage will help smooth the transition? Although I think RMClock can jump from lowest to highest p-states without having to go through the intermediate p-states...

[plympton]

With just 2 p-states, I find that I'm still getting random reboots and BSOD, and am slowly increasing the voltages to see when it'll go away. Each p-state by itself runs just fine.

David,

I think if you do some searches here, you'll find a bunch of people with BSOD & Brisbane & RMClock. Not sure what the culprit is, but there's definately a trend when using the dynamic frequency switching - I had the same problem with a 5000+ Black Edition. Pulled my hair out for a night, and then sold it at a loss the next day just to get it over with. It was great that it could go to 3 GHz on Air, but if I couldn't scale it down to 1 GHz at idle, it wasn't worth it to me.

As an aside, before C'n'Q was the rage, I ran a socket 939 at 2 GHz @ 1.0 volts instead of running C'n'Q - I got the same power use as running at 1.0 GHz using stock volts, so I was a happy camper. You could either fix-volts w/ undervolt, or run C'n'Q at stock volts. Given the 2 choices, I'd do what you suggest - run fast undervolted.

-Dan

[davidh44]

I think it has to do with the frequency switching slightly before the voltage (like dragmor mentioned). If I choose "perform single-step transitions only" instead of the default "allow direct multi-step transitions", and make sure the voltage of the previous p-state falls within the range that will at least start up the next p-state, then it's OK. I'm not sure how much of a performance penalty there is in doing only single-step transitions though.

There are plenty of settings for p-state transitions in the Advanced CPU Settings tab of RMClock that I'm sure would remedy the problem, but I don't think there are many people that actually understand how to adjust those settings (my not being one of them).

[Esben]

I've found that although each P-state by itself can be very stable, using the Performance-on-demand profile with multiple P-states causes stability issues. It's probably due to the frequent switching between different frequencies and voltages? I'm going to try increasing the voltages across the board to see if that helps.

I also had this problem on my 690G and X2 4800+ setup. I couldn't find a fix, except increasing voltage. Now it runs as a file server at a constant 1 GHz, 0.8v.

[davidh44]

I'm going to try CyrstalCPUID (the other well-known undervolting utility) and see if I get more stability when using "performance-on-demand."

I don't know why I didn't think of using it before when RMClock was giving me so many issues...

Anyone else want to do the same and report back the results?

[Esben]

I'm going to try CyrstalCPUID (the other well-known undervolting utility) and see if I get more stability when using "performance-on-demand."

Thanks for testing it, please report back. Which PSU are you using? On my unstable system, the PSU I used isn't the best available
That might influence transition stability.

[RedAE102]

I'm going to try CyrstalCPUID (the other well-known undervolting utility) and see if I get more stability when using "performance-on-demand."

Anyone else want to do the same and report back the results?

I have been for a while, long after I had to give up RMClock altogether. My results are here:

viewtopic.php?p=410542

[Aleph]

I had the same experience with my two computers X2 2350 on Asus 690G and x2 5000+ BE on Gigabyte 690G.
Regular BSOD or automatic resets when RMclock changes p-states (performance on demand).
I changed the memory to the kind that was listed in the memory support list of my motherboard.
This was a slight improvement.
I changed to "perform single-step transitions only" with slow switching 300/400ms.
This was stable most of the time.

Later i realised it could be motherboard related as i have two 690G mediacenter motherboards.
These motherboards are cheap.
These motherboards have due to low cost a simple power design for the cpu (mine had 3 phases).

As i looked to the more expensive boards, i noticed that they had more powerphases, and better components (often cooled) for the cpu power.

I now have a mediacenter board with 4 Powerphases and better capacitors (MA78GM-S2H).
So far i have been able to overclock my BE to 3100mhz with multistep transitions and undervolting.
1000mhz 0.9V and 3100mhz 1.275V, and it is rock solid.

I hope this will help.

Regards.

Edwin.

[davidh44]

Edwin - interesting report, thanks.
CrystalCPUID has proven to work well with dynamic p-state transitions and different voltages on Brisbane CPUs. The only reason I started with RMClock was that seemed to be the most popular one. Once I used CrystalCPUID, I found I liked it much better, and it allowed me to make transitions between 0.8v and 1.25v without randomly crashing like it would with RMClock on my 3-phase motherboard.

I would consider switching to CrystalCPUID even with your higher quality 4-phase motherboard. If it's stable with the cheapest 3-phase motherboards, it'll definitely be stable with yours and alleviate you from having to worry about whether that random crash was due to RMClock or not.

[plympton]

Edwin - interesting report, thanks.
CrystalCPUID has proven to work well with dynamic p-state transitions and different voltages on Brisbane CPUs. The only reason I started with RMClock was that seemed to be the most popular one. Once I used CrystalCPUID, I found I liked it much better, and it allowed me to make transitions between 0.8v and 1.25v without randomly crashing like it would with RMClock on my 3-phase motherboard.

I would consider switching to CrystalCPUID even with your higher quality 4-phase motherboard. If it's stable with the cheapest 3-phase motherboards, it'll definitely be stable with yours and alleviate you from having to worry about whether that random crash was due to RMClock or not.

Is there a tutorial for CrystalCPUID for undervolting, etc..? I have played with it, but never seem to "get" it - the interface is REALLY confusing for me.

-Dan

[alecmg]

Is there a tutorial for CrystalCPUID for undervolting, etc..? I have played with it, but never seem to "get" it - the interface is REALLY confusing for me

Funny, I gave up on RMClock for same exact reason.
CrystalCPUID has only one page to set multipliers, voltages, time and load between changing p-states. It is as simple as it gets.
Also a useful feature for autostart to add to shortcut "/cq /hide", then it starts with working power saving and works in tray. Remember not to close the main window, but rather to click tray icon and choose "Hide"

[davidh44]

Is there a tutorial for CrystalCPUID for undervolting, etc..? I have played with it, but never seem to "get" it - the interface is REALLY confusing for me

Funny, I gave up on RMClock for same exact reason.
CrystalCPUID has only one page to set multipliers, voltages, time and load between changing p-states. It is as simple as it gets.
Also a useful feature for autostart to add to shortcut "/cq /hide", then it starts with working power saving and works in tray. Remember not to close the main window, but rather to click tray icon and choose "Hide"

I think CrystalCPUID is simpler as well, and gets updated by the author much more often. The only downside is you can't take the voltage below 0.775v...not an issue with desktop chips, but mobile chips could benefit from being able to go lower.

If you're new at this, the only screen you need to configure is the "Multiplier Management setting." Click on "enable voltage" and make your own adjustments to the multipliers and voltages. Leave everything else at default.

[loimlo]

In fact, there's a CrystalCPUID tutorial on the SPCR. I also think that RMCLOCK is more difficult to use than CrystalCPUID.

http://www.silentpcreview.com/article231-page1.html