CrystalCPUID

[LiamC]

I have been playing around with this (thanks Jan), and it is indeed a cool thing, but it locks up my system after a few minutes.

Does anybody know of any fixes?

The only thing out of the ordinary is my system clock, 265 MHz vs stock 200

keeping voltage the same and dropping the multi, machine still locks.

Is this a known issue? Is there a forum I should be reading?

s754 2800+
Zalman 7000B AlCu
Abit KV8 Pro with v 1.7 BIOS (latest)
Latest C 'N' Q drivers from AMD
CrystalMark v 4.0A

 

[Jan Kivar]

Are You running a multiplier lower than the stock 9x?

Do You get a blue screen, or just a total lockup?

There are few known bugs:

If Cool 'n' Quiet is enabled in Windows (Power Scheme set to "Minimal Power Management") and the user is using a multiplier lower than the stock one (say, in your case 8x) to be able to use higher FSB, the driver is stupid enough to push the multiplier back up to 9x, which will cause too high clocks for the CPU and it freezes.

If You read Sudhian, You'll know that at least the MSI board cheats the half multipliers by rising the FSB additional 10 MHz. Which would jazz things up nicely...

Can we assume that the computer itself is stable at 265 MHz FSB? 24h Prime95 session should check for that.

Also, the BIOS revision matters much - back in the days when my MSI K8N Neo Platinum was the high-end board from MSI, MSI forums were filled with posts comparing new beta BIOSs (they still are, but the release rate has really gone down). There some beta BIOSs sucked above 220 MHz, whilst with other versions 270 MHz was still fine (on a same computer).

Cheers,

Jan

 

[Jan Kivar]

Also, with my board & CPU I couldn't go from 1GHz/0,85V straight back to 2GHz/1,5V with ClockGen, it just freezed when I tried. I had to first set the voltage to 1,1V and then to 1,5V. With ClockGen, one can make .ini files where the settings are just listed like in .bat files. A shortcut can then be made to launch different settings. I just ended using ClockGen twice (with a .bat file) when I wanted to use 2 GHz.

I didn't try to find out what actually caused the freeze, but IMO it hung because I set the voltage from 0,85V to 1,5V.

Cheers,

Jan

 

[Bozo]

Crystalcpuid ???


Bozo :mrgrn:

 

[ddrueding]

I just upgraded to the newest version (4.something) and it's still working great. If you want more information, go to Jan's thread @ SPCR

I am (kind of) interested in playing with some overcloacking potential here, but it doesn't look like multipliers above stock work? I take it the CPU is locked? Is there not a way to dynamically change FSB? Am I asking too much? Are there enough question marks in this post? Why am I posting under the influence again?

 

[LiamC]

Bozo,

http://crystalmark.info/?lang=en

It uses Cool 'N" Quiet, SpeedStep or other technologies to change a CPU's multiplier & voltage on the fly. That way, your PC runs at full speed when you need it, but throttles right back when you don't.

Jan,

Yes it's stable @ 265 (at least with every other test I have been able to throw at it). I was playing around last night and it's harder to make it freeze using lower FSB speeds, but it still happens. Total lock up. No blue screen.

I was able to intercept a Windows error report and there was a Null pointer exception thrown somewhere in the code, but I didn't have a real good look at what was going on.

Power management is set to Home/Office. I'll play around with these settings.

I set Crystal to go from full speed to 7.0x multi (1.3V) & 4.0/5.0 multi @ 1.1V. But now that you mention it, it jumps straight from either low state to the high state. I'll play around with these settings.

I did notice in the AMD processor guides, that the intermediate P-states (power states) are not defined for my particular stepping--maybe something is actually broken in the silicon?

Don't read Sudhian. I'll see if I can find a a few other BIOS revisions to try.

Thanks for your suggestions.

ddrueding
Reading that thread on SPCR (that either you or Jan linked in another post) is what got me started the first hit is free

You're right about higher multipliers being locked.

 

[Jan Kivar]

Power management is set to Home/Office. I'll play around with these settings.

IMO there's nothing to the other settings. Setting to "Minimal Power Management" makes Windows do the same thing that CrystalCPUID does, and things go wrong there. I wonder if the predefined states for each model are encoded to the CPU or the driver? Ultimately, hacking the AMD driver would be better for regular users, but I like that I'm able to disable the multiplier management and run at the lowest setting when I want.

I set Crystal to go from full speed to 7.0x multi (1.3V) & 4.0/5.0 multi @ 1.1V. But now that you mention it, it jumps straight from either low state to the high state.

Yeah, and now that You mention that You have it going straight from low to max, I remember that I have the same setting in use. So maybe that's not the culprit after all.

You should disable the multiplier management and test each state for stability. The options have a setting about what state to be used when the multiplier management is disabled. You should lock to each setting and run Prime95 for 12h to make sure that each individual state is stable. Prime95 is IMO the best stability test program (and the only one I've tried ). What program(s) do You use?

I did notice in the AMD processor guides, that the intermediate P-states (power states) are not defined for my particular stepping--maybe something is actually broken in the silicon?

I don't think this would be a possibility. But You could test this by setting the state switching to work only between min and max setting, thus skipping the mid state completely.

Cheers,

Jan

 

[LiamC]

Well I have found that I can clock the bus to 240 MHz (2160 MHz--stock is 1800MHz) and CrystalCPUID will run all day no problems. 245 MHz bus and the lock-ups start.

At 2160, I can set CNQ to run:

9 x 240 @ 1.5V
7 x 240 @ 1.3V
4 x 240 @ 1.1 V.

At the lowest setting, with a Zalman 7000B AlCu, the CPU is operating at 1 to 2 degrees above ambient, and the whole system is quieter than my Cyrix 166+ SmoothWall box (noisy P/S).

I tried just changing the multiplier (without changing the voltage and this made no difference). Memory timings made no difference. Power settings made no difference.

I am going to torture test higher bus speeds with Prime and see if this is a limitation of my system or CrystalCPUID

 

[LiamC]

Update: Clocking the bus at 240 MHz didn't solve the problem. This does seem a little odd eeing that everything else is happy @ 265 MHz.

For the last day, I reverted to just the AMD driver and setting the Power Options (Control Panel) to Max. Battery. I get the same result (well almost) as CrystalCPUID, it switches faster, and so far, is happy at a 260 MHz bus. This seems like a better result.

 

[CityK]

I'm trying to find a thread, in which I believe Jan talked about the bios table readings etc., in reference to the use of crystalcpuid.

With CnQ, the idle temp readings I'm getting in this toaster (soltek 3901) are ~53C (1000@1.1V, default CnQ setting). This is with a 939 3000. I'm unable to check what they are now with crystal set at 1000@0.9V, as the hardware monitoring utility goes all screwy once crystal starts (i.e. core temp 255C, fan speed 13K....uh-huh).

So far, I can't seem to go lower then 5x multipler or lower then 0.9V.

Also still trying to figure this program out. Not as straight forward as I would like.

How are you guys running Crystal? Are you just placing it in the Startup folder with cmd line arguements?

 

[ddrueding]

I actually put in a registry key (basically the same as the start-up folder) that includes arguments to start minimized.

I haven't played with overclocking yet, but I think I'll start now that my 6800 has show up and I'm doing a lot more DIVX encoding...

 

[LiamC]

Yeah, I've been meaning to update this post. The AMD driver + Power Management Options still tanked (system freeze) just like using CrystalCPUID. I start Crystal with the command line args.

I now do things manually. run at high bus normally, and when I want to run cool, I drop the FSB speed via the Abit software and then start Crystal.

 

[Jan Kivar]

I'm trying to find a thread, in which I believe Jan talked about the bios table readings etc., in reference to the use of crystalcpuid.

Hmm... what exactly is this bios table? SPCR has this list of CnQ-supporting motherboards. I've read that some motherboards actually require that CnQ is turned off in BIOS in order to be able to fully use 3rd party programs. Could You be more specific about the table?

How are you guys running Crystal? Are you just placing it in the Startup folder with cmd line arguements?

Yep.


There is another utility to try: RMClock Utility. A new version has been released recently. Seems to have nice state change/CPU load graphs. I haven't tried it yet, but judging from this thread it seems to use 200 Mhz step (=full multiplier, there is option for enabling half-multipliers), and it interpolates the voltage linearly between the two user-set end-points. Which isn't optimal if one is trying to minimize the power consumption on CPU's behalf. FYI, I just glanced the thread - maybe there is more info than this (same applies to the RM forum also).

Cheers,

Jan

 

[CityK]

Hi Jan, thanks for responding.

I'm trying to find a thread, in which I believe Jan talked about the bios table readings etc., in reference to the use of crystalcpuid.

Hmm... what exactly is this bios table? SPCR has this list of CnQ-supporting motherboards. I've read that some motherboards actually require that CnQ is turned off in BIOS in order to be able to fully use 3rd party programs. Could You be more specific about the table?

I think I may have confused some things. I think I remember reading somewhere (most likely SPCR) that by dropping the voltages using CnQ or Crystal, some hardware monitoring programs get all messed up because they rely on some bios table of values, but that table is no longer valid given the changes instituted by Crystal etc. This initially appeared to be the case for me, as my (vendor supplied) hardware monitoring app was reporting wildly inaccurate values. However, I think that must have been caused by some other incompatiblity, as that has since been resolved and I now get readings that (I believe) are correct.

Now, I say above that "I believe are correct" only because I remember a discussion, likely here or on SPCR, between I believe you and dd, about some bios reporting wrong temps (overstating by ~15-20C). I'm wondering if that is the case with mine or not - given that while using CnQ and Crystal, and running at 1000MHz@0.9V, idle CPU temps are generally ~51C, while the mobo sensor reports low to mid 40s. Pheeeww! Its hot in there. Also, the hardware control for the Icy-q fan seems to be somewhat useless - once the fan increases speed, it never throttles back down, even in face of decreased temps. Looks like a job for speedfan or something.....although I have yet to check the bios more carefully for any options.

I'm starting to really like this little box, but my concern is to drop both the noise and heat levels, which as you know, are often conflicting goals....which makes this somewhat of a challenge from the start given the initial state of conditions and, due to the system's size and confined space, the limited options for modifications.

SPCR has this list of CnQ-supporting motherboards.

I notice my board isn't on their list. I don't know if that list is still actively updated or not, but if it is, given that I don't have an account there, and if you or dd felt up to it, could you be so good as to submit a working candidate for me --> Soltek SL-B9D-FGR (EQ3901-300P).

There is another utility to try: RMClock Utility.....I just glanced the thread - maybe there is more info ...

Thanks, I'll look it over later on.

 

[ddrueding]

I found that through my overclocking adventures, MBM5 started to give really inaccurate results. The latest version of speedfan still works great, though. It also does temp reporting for my GPU, and would do fan control for it if it had a fan

 

[Jan Kivar]

Hi Jan, thanks for responding.

With pleasure.

I think I may have confused some things. I think I remember reading somewhere (most likely SPCR) that by dropping the voltages using CnQ or Crystal, some hardware monitoring programs get all messed up because they rely on some bios table of values, but that table is no longer valid given the changes instituted by Crystal etc. This initially appeared to be the case for me, as my (vendor supplied) hardware monitoring app was reporting wildly inaccurate values. However, I think that must have been caused by some other incompatiblity, as that has since been resolved and I now get readings that (I believe) are correct.

Good to hear that. As David already mentioned, MBM doesn't like when the CPU voltage changes, especially when power-save features (STR) are used. I've seen voltages from 0.50 to 2.67 with MBM, while SpeedFan works fine.

Now, I say above that "I believe are correct" only because I remember a discussion, likely here or on SPCR, between I believe you and dd, about some bios reporting wrong temps (overstating by ~15-20C).
...

It's actually easy to test: just fire up some CPU stress utility, like Prime95 or CPUBurn (F@H doesn't quite produce maximum heat stress). If the CPU temp gets higher than 70°C, there's a good chance that the BIOS shows too much. With my motherboard (MSI K8N Neo Platinum) it occasionally showed the correct temperatures (normally if the computer had been booted once) with BIOSs lower than 1.3. With the current BIOS (1.4) I can see similar effect, but luckily the variation is only 1-2°C when idle. Haven't tested what the variation would be when loaded, but I'd expect it to be the same.

The funny thing here is that even though I use STR, temps are shown correctly when I exit STR. IIRC the BIOS actually has to enter&use an offset to the temperature reading with NewCastles, and the offset seems to be sustained even when the computer is basically unpowered. Hibernation should IMO do the temp rise effect again, since no CPU state information is stored in hibernation.

I haven't tried the BIOS fan control in my board. With my current setup I can't hear the CPU fan (Zalman CNPS7000-AlCu@5V) because my power (Zalman ZM400B-APS) masks out the noise.

I'm starting to really like this little box, but my concern is to drop both the noise and heat levels, which as you know, are often conflicting goals....which makes this somewhat of a challenge from the start given the initial state of conditions and, due to the system's size and confined space, the limited options for modifications.

I don't remember if You have already explained your box in some other thread. I manage to keep the CPU temp under 50°C even with the HSF at 5V. I have the Nexus 120mm case fan as only exhaust (12V). Without CrystalCPUID the temp was near 60°C, so it makes quite an improvement.

I notice my board isn't on their list. I don't know if that list is still actively updated or not, but if it is, given that I don't have an account there, and if you or dd felt up to it, could you be so good as to submit a working candidate for me --> Soltek SL-B9D-FGR (EQ3901-300P).

IIRC the list is maintained by a forum member (jojo4u if memory serves me right) based on a thread in the forum. I'll check what the current situation is, and mention your board there, if David hasn't already done so.

Cheers,

Jan

 

[ddrueding]

Go ahead Jan, I'm still a noob over there.

 

[LiamC]

Well I have been doing some troubleshooting. Seems this CPU will run at up to 2.34 GHz--with the proviso that CrystalCPUID will freeze and Hearts of Iron will fail every second time when saving its save game files. Unreal timedemo will freeze after 30~50 iterations as well--I've always taken this as a bad sign and it comes through again.

When I get the bus down to 235MHz, All the problems disappear. HT clock, Memory clock, AGP clock were all locked/within parameters, so it just seems that the CPU itself can't run reliably at too high a speed. The failure isn't critical, more of a soft error. All my previous overclocking trips have ended in hard locks/errors, so this wasn't as obvious--and is more insidious as this way, erros could be happening but not noticed.

At 2.1GHz (235 FSB), everything passes and the CPU runs at 1.4V. I can't feel any difference between 2.3 and 2.1 GHz, it's more a psycological thing.

 

[time]

Well, this is going to depress you Bill, but I'm currently running:

s754 2800+
stock cooler
Asus K8N
CrystalCPUID 4.3.10

The BIOS lets me boot at 2.53GHz and complete 3DMark. I can get well above 2.6GHz with ClockGen, finally blue-screening at 2.7GHz (300MHz FSB). For this limit testing, I boosted the voltage to 1.65V without even trying lower values, and didn't attempt 3DMark etc.

Back in the sane world, I've set the FSB to 266MHz to achieve 2.4GHz at 1.5V. My middle Crystal setting is 1.6GHz (6x) at 1.2V and the minimum setting is 1.067GHz (4x) at 0.85V. The middle voltage is a guess, but I know the minimum needs more than 0.8V at that speed.

CPU fan speed while I type this is 1095rpm.

Oh yes, I'm also running with all three DIMM slots populated with Transcend/Samsung PC3200, currently at 221MHz.

 

[LiamC]

Well, this is going to depress you Bill,

You got that right!... :eekers:

 

[CityK]

Oh yes, I'm also running with all three DIMM slots populated with Transcend/Samsung PC3200, currently at 221MHz.

Actually Time, your mem speed is 217. A number of apps incorrectly calculate the mem speed when using memory ratios. Allow me to demonstrate:

First off, the correct determination of memory frequency when employing a memory divider on the A64 is given by:

Mem Frequency = CPU clock speed / mem divider
              = (HTT x CPU multipler) / ceiling[ceiling(CPU multiplier) x "base HTT:mem ratio"]

The cpu clock speed is obviously straight forward.

On the otherhand, the way the A64 mem controller calculates the memory divider is not as straight forward. First thing to note is that the memory divider has to be an integer, which is accomplished by the first ceiling function. The second, and perhaps even more interesting feature to note, is that the inner ceiling function insures all fractional CPU multiplers are rounded up to an integer value. The negative ramifications of this effect will be illustrated by example to follow. Third, the ratio (in the memory divider side of the equation) is actually the inverse of the "mem:base HTT" ratio in the BIOS. For example, if one sets a 5:6 (in otherwords 166:200) ratio in the BIOS, a corresponding ratio of 6:5 is used in determination of the memroy divider.

Now, taking your example, we have a HTT=266, CPU Multiplier=9, and a base HTT:memory ratio of 6:5 (although you didn't devulge this, it is, by deduction the only possibility). Thus, your:

memory frequency = (266 x 9) / ceiling[ceiling(9) x 6/5]
= 2394 / ceiling[9 x 6/5]
= 2394 / ceiling [10.8]
= 2394 / 11
= ~217

Apps that do not calculate the speed correctly do not use the ceiling fn. For example:
memory frequency = (266 x 9) / (9 x 6/5]
= 2394 / 10.8
= ~221

Now, about that half multiplier example, how about we use your values again but with the exception of using a 9.5 CPU Multiplier:

memory frequency = (266 x 9.5) / ceiling[ceiling(9.5) x 6/5]
= 2527 / ceiling[10 x 6 /5]
= 2527 / 12
= ~210

Shocking isn't it!

Just to give a final example, suppose CPU Multiplier = 8.5, HTT=240 and the BIOSes "Mem:HTT" ratio is 5:6, then:
Mem speed = (240 x 8.5) / (ceiling[ceiling[8.5] x 6/5])
= 2040 / ceiling[9x 6/5]
= 2040 / ceiling[10.8]
= 2040 / 11
= 185MHz

If you work the last one through but use a lower 8x CPU multiplier, you will observe that memory gets driven at 192.

 

[time]

Wonderful stuff, CityK!

Actually, the app that incorrectly calculated the mem speed was me. I was trying to finish the post, thought I'd add the memclock, and guessed wrongly that ClockGen was mistaken.

In fact, ClockGen and CPU-Z both display the correct frequencies, and your arithmetic is spot on.

It's worth pointing out that memclock actually changes when Cool 'n Quiet or CrystalCPUID shifts gear. For example, with a multiplier of 4, your equation yields:

memory frequency = (266 x 4) / ceiling[ceiling(4) x 6/5] = 1064 / 5 = ~213MHz

Fascinating. Thanks for explaining it.

 

[CityK]

Oops! Yes, sorry, I forgot to mention that the mem controller only works/utlizes multipliers 5x and up, so both ClockGen and CPU-Z are obviously programmed correctly.

I'll go dig up the article from which I originally gleaned this info a little while back. Actually, I think I'll create a new post with reference to it and and another article I just read on the weekend. The two are so complementary and provide excellent results. Unfortunately, some of the conclusions/analyses are flawed.

 

[CityK]

All hail Jan's journalistic debut!

 

[blakerwry]

That's pretty good from an article, informational, and visual standpoint.

Good job!

 

[Jan Kivar]

Thanks guys!

Jan

 

[CityK]

I was unaware that CrystalCPU was independent of C'n'Q. I had been under the impression that C'n'Q was actually a requirement for Crystal.

I note that having unistalled the AMD driver AND disabling C'n'Q in the BIOS, Crystal continues to function perfectly.

That last point makes me pause to think what the C'n'Q BIOS function even does or what the special need of the AMD driver is?

 

[Jan Kivar]

I don't know for sure, but the BIOS' Cool 'n' Quiet option might relate to BIOS showing a "table" of available FIDs and VIDs to the OS**. Setting CnQ disabled in BIOS should result in failure for all programs that rely their operation on this table ("programs" such as the AMD's driver). The program can't find the table (or the table has only one allowed state) and thus cannot operate properly.

This is not the case with e.g. CrystalCPUID, since it just sets the FID and VID according to what user wanted. E.g. if the FID was set too high for the given CPU, the computer freezes, and one needs to to power cycle (cold-boot) the computer, or even reset the BIOS. I don't know if CnQ needs to be enabled in BIOS for some motherboards to allow CCPUID to operate. (So, I said in the article to set CnQ enabled.)

Windows uses the driver to select & set the FID and VID according to the CPU usage. (I don't know anything about drivers, and how they interface/operate with the "kernel", so my description is probably more or less off.)

If You want to try, check if the AMD driver still works when CnQ is disabled in BIOS.


Cheers,

Jan

**I stumbled on something like this while I was searching for similar apps for Linux. (The reasons for the search of similar Linux-apps can be found on the article discussion thread at SPCR.)

Some Linux-users have determined that the reason why CnQ doesn't work on some motherboards (at least with Linux) is related to corrupted/missing CPU PST tables in BIOS. (Anyone know what these PST tables are? AcronymFinder has 40+ suggestions for PST...)

 

[CityK]

Anyone know what these PST tables are?

I'm thinking power state something...and upon typing "power state s0" into google, the very first link is a msdn document that talks about Power State Transitions. So that's my guess.

If You want to try, check if the AMD driver still works when CnQ is disabled in BIOS.

I'll check this out later tonight.

 

[CityK]

If You want to try, check if the AMD driver still works when CnQ is disabled in BIOS.

I'll check this out later tonight.

It didn't work. So I guess your far fetched theories might be on to something Jan