CrystalCPUID + Abit NF7-S help please!

[mattthemuppet]

Hi all,

just swapped out my old Gigabyte 7VT600 mobo (no fan control or undervolting) for a rather nice Abit Nf7-S (w/ fan control and undervolting), but CrystalCPUID doesn't want to work. It worked fine on my old mobo (the Sempron 2400+ has been multiplier unlocked), dynamically adjusting the multiplier, though there wasn't much point as I couldn't undervolt. I've now downloaded the latest version of CrystalCPUID (45 I think) and it doesn't work, neither did the previous version. I can undervolt my CPU to 1.525 in the BIOS which is perfectly stable, but whenever I change anything in CrystalCPUID it just blank screens me or freezes the computer.

Any ideas? Could there be something in the mobo BIOS that's stopping it or some other conflict? I'd really like to get this working as it'd make what's now a really quiet PC very quiet indeed

Any help's much appreciated!

matt

[akio]

Hi,

According to this site NF7-S is on the list which multiplier can't be controlled by software such as CrystalCPUID, i'm afraid.

[mattthemuppet]

thanks akio! just done a bit of digging and it doesn't seem like any nForce2 board supports on the fly multiplier, whatever program's used can't believe after all the research into a new mobo and the time and hassle of switching my old one out, I made this mistake! grr

[akio]

mattthemuppet, I don't know well but nVIDIA seems to offer an alternative to CrystalCPUID.

NVIDIA System Utility

Just try googling "NVSU".

[edit]
i found more, but, i don't understand these utilities' functions and differences exactly.
NVIDIA nTune
good luck

[p'jem]

Matt,

try 8RDAVcore, it doesn't allow multiplier adjustments but you can

-adjust FSB on the fly
-adjust Vcore on the fly
-Set StopGrant state to enabled even if the BIOS don't support the option
-Control the output of two fan headers
-easily program it to automatically apply certain presets when certain temperatures or cpu utilisation thresholds are reached, and to run on startup.

At one stage I had an XP1700 in my NF7-S with x6 multiplier set in the BIOS. When idle, it would set 1.15V, 133FSB, and virtually stop the fans altogether. If CPUload exceeded 90% for 5 seconds, it would apply the 1.3V, 176FSB preset, and give the fans a little more juice.

Sadly you can't really adjust the FSB over more than a 33mhz range or so with any stability, and there doesnt appear to be any solution to that.

Enabling StopGrant only seemed to lower temperatures by 2 or 3 degrees on my Barton & Thoroughbred chips, but takes a huge 15 degrees off the Spitfire Duron 700! This is a shame because on the first version of the Nforce 2 chipsets this option made a really big difference to the thoroughbreds too, but it seems NVIDIA crippled it on later versions.

These NF-7 boards don't seem to POST reliably with any CPU, at any clock speed, below 1.3V, but once you're into Windows you can let 8RDAVCORE wind things right back for you.

Anyway, has anyone managed multiplier adjustments on the fly with an NF7-S?? I tried the NVTune and Nvidia system utility but the option was not to be found, though my Barton was mutliplier unlocked. Maybe it'd be possible if I had a mobile XP chip, or I close the L5 bridge?

Here's hoping

[mattthemuppet]

hi p'jem, thanks for the tip! Right in the middle of trying to get 8rdavcore working stably - sorted 3 presets (140, 166 and 184Mhz at 1.275, 1.4 and 1.525V) for my Sempron 2400, all Prime95 tested for >6h, set up autoFSB
and put the program in the start menu, then had multiple screen freezes and reboots until finally windows wouldn't start (missing ntoskrnl file ). No idea what happened, though thankfully we've been able to repair windows and we're just backing up again. Perhaps it was that 1.3V minimum problem?

As for the multiplier on the fly - sorry, but that's a big big no no on this board. I've unlocked my chip to a mobile (ID'd as an Athlon XPM although it isn't) and could adjust the multiplier on the fly with my old Gigabyte via KT600 board, but anytime I tried with the NF7-S it froze the screen. Sorry

[p'jem]

Sorry to hear that it ate your operating system!

The gap between your highest an lowest settings was quite big though, so you were pushing things a bit. Also I think it depends on your CPU and FSB - my 1700 was pretty amenable, I dont know how close it was to its max FSB but I know it woudnt complete 3D Mark at 200 FSB on three known good motherboards, at any multiplier or voltage.

My Barton isn't so co-operative. It's gone two weeks without reboot on 5x200, 1.2V. But when I tried programming the BIOS to BOOT at 183, and AutoFSB to scale back to 166 when idle, things got ugly, even if I left voltages untouched.

On the subject of voltages, I'm concerned that if you change the voltage and FSB simultaneously, the FSB will increase almost instantly. However, the mobo voltage regulators will take a few milliseconds to charge up the capacitors by the CPU socket to the new (higher) level, and stabilise at that point with no overshoots or undershoots. In the meantime your PC might crash.

In AutoFSB, your can set the Vcore delay to a higher value than the FSB delay, which will give things time to stabilise on the way up. On the way down though, that'd make things worse, as the cpu voltage gets cut before the FSB, so your CPU is starved of core voltage while still running at full speed for a few hundered thousand clock cycles.

Here's how I got around the percieved problem ;

I created 4 presets

* Voltdown - Set voltage to 1.15V and FSB to 133
* Voltup - Set voltage to 1.3V, dont change FSB
* Clockup - Set voltage to 1.3V and FSB to 176
* Clockdown - Don't change voltage, set FSB to 133

....and these rules

If CPULOAD > 90 For 5 Seconds ---> Voltup
If CPULOAD > 90 For 10 Seconds ----> Clockup
If CPULOAD < 40 For 5 Seconds ----> Clockdown
If CPULOAD <40 For 10 Seconds -----> Voltdown

That way, when clocking up, I'd bump the voltage 5 seconds before the FSB. On the way down, the FSB would come off first, then the voltage.

Didn't help with my Barton though. Maybe that's because at 200FSB stability is too critical for unexpected changes in bus speed. All my autoFSB crashes were when exiting from a game, as it happens.

Project for tomorrow - see if those NVIDIA utilties do the job better.

For now though, it's surprising how playable the likes of Half Life 2, Far Cry, Halo and F.E.A.R. Demo are with a 1 Ghz CPU aided by a DX9 card and Soundstorm audio!

[mattthemuppet]

I can't 100% point the finger of blame at it, but it's a remarkable coincidence if not Nearly finished fixing everything, so fingers crossed it won't be a lasting problem.

What you said make as it was when it went between two states that it started freezing up. I'll sort everything out, back up and then have a fiddle to see if I can get things working better - it offers so much potential that I can't just forget about it! I'll report back with the results...

[p'jem]

Had a play with NVTune today.

Only seems to give control of FSB and AGP speeds, and memory timings, though you can program it to apply X preset when Y application runs.

Of more interest was its automatic tuning mode, where it increments your FSB one 1Mhz at a time while running diagnostics to find the safe upper and lower limits of the system.

Going down, it stopped at 180Mhz and said that was my lower safe limit. No crashes!

Winding upwards was a little hairier - 3 crash/reboots for it to discover an upper limit of 220 and which memory timings are safe to tighten further.

Given this info I'll try getting AutoFSB to switch between 180 and 220 when I'm feeling brave enough, see if it holds together in practice.

It's only a 20% clockspeed change though, even if it's possible to reduce the core voltage a little further at the lower speed, I'll not be saving more than about 30% power dissipation. Not quite enough to Save The Earth given that it runs 24/7 and that STOPGRANT state doesnt work on Socket A!

[p'jem]

I lowered the FSB of my system in the BIOS to 133Mhz to see if this allows AutoFSB to get more than a 20% stable clock speed change on the same CPU and mobo.

NVtune recommends a lower limit of 120 now, and after a couple of crashes, found an upper limit of 166.

This is indeed a bigger clock speed change - 27%, and you could probably get an even bigger one if the system was set to POST at 100FSB.

However, we now have a lower FSB on the "full speed" setting, which means a higher clock speed is needed to get the same gaming performance. Does this cancel out any gains made this way?

-------------------------------------------------------
::Warning! Some rather shonky math follows!!::
-------------------------------------------------------

from -
http://www.tomshardware.com/cpu/2004122 ... #directx_9

3D Mark 2005 CPU Test
Barton 3000 2100Mhz 200FSB = 743PTS 0.354 pts/Mhz
Barton 3000 2167Mhz 166FSB = 703PTS 0.324 pts/Mhz

0.354 / 0.324 = 1.1 ie. 20% FSB Increase worth 10% more performance

220fsb is 25% higher than 166FSB, so 12.5% performance lost AT SAME clock speed

Barton 2500 1833Mhz gives 636 points
Barton 3000 2167Mhz gives 703 points
10.5% more performance from 18.2% more clock speed

% increase in clock speed to get 1% more performance at same FSB = 18.2% / 10.5% = 1.73

Additional clock speed required to make up for lower FSB = 12.5% x 1.73 = 21.7% more core speed needed

so instead of 220 x 5 = 1100Mhz

@166 FSB we'll need a core speed of 1100 x 21.7% = 1339Mhz

so we need 8x multiplier

166 x 8 = 1328 Mhz (gaming setting)
120 x 8 = 960Mhz (quiet setting)

Wheras before we had

220 x 5 = 1100 Mhx (gaming)
180 x 5 = 900Mhz (quiet)

------------------------
Conclusions
------------------------

1 - the higher FSB still results in a CPU dissipating slightly less power in quiet mode, even though the range of adjustment is smaller

2 - the higher FSB gets the same performance in gaming mode with considerably less power dissipation

3 - performance in quiet mode will be much better at the higher FSB. This is important because my AutoFSB scripts can't bring the CPU up to full speed in less than 10 seconds.

However, the noise floor of my system is the Seagate 7200.7 HDD, the real reason I'm bothering with AutoFSB is because I'm trying to cut the energy wastage having this thing on 24/7.

The RAM and Chipset probably dissipate just as much power as the CPU in my system, so overall AC draw will be lower in idle mode with the reduced FSB. Even more so because I'll probably be able to undervolt the RAM and Chipset too.

Or maybe I should just get out more often!

[mattthemuppet]

blimey, that's impressive!

interesting to see empirical back up for the inefficiency of overclocking. Guess that implies that underclocking doesn't cut performance as much as you'd expect?

did you take any temp measurements at those various settings? it'd be interesting to see if lowering FSB is more effective at lowering temps than lowering multiplier. Did your lowest Vcore change at all?

[p'jem]

[quote]did you take any temp measurements at those various settings? it'd be interesting to see if lowering FSB is more effective at lowering temps than lowering multiplier. Did your lowest Vcore change at all?[/quote]

Nope, l left voltages and multipliers unchanged so as not to affect the results of Ntune. It was just an experiment to see if there was a tradoff between FSB and the amount of on-the-fly adjustment AutoFSB could give you. If I decide to adopt a lower FSB permanently then I'll start optimising Vcore and taking temperature readings.

You've got a point though, I can hit 1000Mhz either by

200 x 5
or
100 x 10

...but I havven't checked out if the lowest stable Vcore is the same for both methods, or if both give identical temperatures at the same Vcore.

At the moment, I am assuming that minimum Vcore is proportional to clock speed only, and that CPU heat dissipation is proportional to Vcore and Clock speed, but in reality it's possible that FSB affects both.

At the moment, CPU temperatures never exceed 40degress. At some point however, a new game will come along that forces me to raise the multiplier and Vcore, making this more of an issue.