CPUgenie - New Undervolting Tool

Long time has passed since last RMClock update, which left many people with 45-nm Core 2 CPUs without adequate tool for multiplier management due to lack of support for non-integer multipliers in RMClock.

CPUgenie 1.0 is a small but powerful tool that allows:

- Fully configurable Load-based CPU Multiplier Management
- Per-Multiplier Voltage Control / Undervolting
- Support for Super LFM modes
- Different CPU management policies on AC powers and battery
- Per-Application CPU policies, allowing you to e.g. lock the CPU to the highest speed when you run specific application
- Voltage Optimization Wizard, tool that aids finding of lowest stable voltages using stability tester
- Power Monitor, tool that allows monitoring of important CPU parameters (temperature, voltage, etc...) as well as Application-based CPU usage and ACPI BIOS info
- Support for non-integer multipliers
- Support for desktop and mobile Intel Core 2 CPUs
- 32-bit and 64-bit version (with signed drivers)

With CPUgenie, it is possible to reduce CPU temperature under full load and even save power at the same time without sacrificing the performance. However, before experimenting with undervolting please make sure you understand the process and respect the guidelines (e.g. full backup before testing)

Check it out: http://www.cpugenie.com/

Reducing temperature by 26 C on Sony Vaio Z: http://www.clockmod.com/index.php?/articles/view/notebook_temparature/
Undervolting the Intel DG45FC Fly Creek mobo: http://www.clockmod.com/index.php?/articles/view/undervolting_intel_dg45fc_fly_creek_htpc_system/

 

Unlike RMClock and CrystalCPUID (which works with non-integer multipliers), this thing doesn't appear to be free; it says "Order now" and offers a 30-day trial. IMHO, it's not worth paying for something like this when you can get something very similar for free.

 

Indeed, tool is not free but it IMHO does the job of automatic performance control pretty well and has a very nice UI and testing wizards - I've found CrystalCPUID to hang my GM45 notebook and voltage information it gives is quite wrong. RMClock works OK but w/o fractional multipliers and its unsigned 64-bit driver requires elimination of driver signing checks in Windows in order to work...

Anyway, having extra choice can't be bad.

 

I've heard that CrystalCPUID usually hangs the system when one selects the multiplier (but its fine for me). Voltage control has been fine and I have never heard of any bugs or glitches related to it.

RMClock does everything perfectly, and the lack of support for half-multipliers doesn't make any difference.

One is not going to lose much CPU power with a half-multiplier less. At most, it'll be 133MHz less per core (penryn-refresh). 133MHz, more or less, won't even make a difference when encoding AVC HD vids.
And RMClock runs fine for me in Vista 64.
Plus there is an excellent guide on Undervolting using RMClock.

If CPUGenie would have been free, then I would have probably tried it out....

 

Andy how did you manage to make RMClock work in Vista 64? I tried and got nothing but a bunch of errors.

 

You know did anyone actually try this? The program seems to suggest that it can drop your VIDs below Intel's "hard" limits (ie 0.9250v for the P8400), but neither CPUZ or PCWizard seem to be reading even my RMClock undervolts properly, so I can't tell if it actually works...

 

I find RMClock on Windows 7 32 bits consumes a little bit of CPU I wonder if it's normal.

 

I tried the demo(?), all it does is unlock the voltages like the regfix for RMClock does, it doesn't actually change the voltages since:
1. CPUz never saw any changes in voltage
2. it never crashed (.95V -> .712V, seems fishy)
3. temps didn't differ when stress testing with Orthos

 

Gonna try this out later, looks really cool. How much does a copy cost?

Use CPU-z 1.41 and it will show the correct voltage.

 

Darn, you too? Really seemed too good to be true I've been trying to figure a way to get around the lock for a while now.

You know, I originally had 1.41 (or I thought I originally had 1.41), and it was telling me 1.038v while RMClock was telling me 0.9250; I upgraded to 1.50, and it was still telling 1.038v while RMClock was still telling me 0.9250. I'll try 1.41 again, but shouldn't the latest version be the most accurate?

 

No, they didnt update the support for the mobile chips after v1.41.

The latest CPU-z reads my T2300 (yonah) & PM725 voltages fine. Its the Merom and Penryn Chips that have issues after v1.41.

Also to show the right clocks on CPU-z (Superlfm clocks). You have to press F9.

 

RMclock did nothing for me (T5800 on an Asus N50Vn-A1B), and I couldn't figure out how to get CrystalCPUID to work. I'm gonna download the trial for this right now, mess with it later when I have more time.

 

It does work for me (not unlocked voltages, of course - all CPUs I have are VID locked), I installed it and verified the voltage changes with RMclock running in the background in monitoring mode only.

Funny, CPU-z also does not report VID changes for me in RMClock - but RMClock and CPUgenie do confirm each other's change of VID. I think the way CPUgenie and RMClock read voltages is quite different than CPU-z - CPU-z is more reliable for cases when you have special VRMs like on desktop Mobos, where you can change voltages in BIOS. I guess CPU-Z people invested a lot of time trying to figure how to get those voltages.

But, RMClock, CrystalCPUID and CPUgenie read and write the VID registers and report voltages from Intel's reference table. However, as most laptop mobos are based on reference design, this is no problem - as voltage you get setting the VID is the real voltage. In case of desktop mobos like ASUS, this might not always be the case.

Btw, when you click "Unlock Voltages" it tells you that it can't do wonders and that most CPUs are VID locked anyway via. popup window - it even advices not to do it.

But regular voltages (within VID range of your CPU) work quite well - at least @CPUs I tested at home (mobile Penryn and Merom)

@edit - screenshot of warning updated

 

Blah, no point in using this over RMClock then; I'd really like to see someone get those locked VIDs working... to the extent that I'd pay for it XD

Might just be your version. Flipfire was right in that 1.41 supports just about everything; I had 1.44 previously and upgraded to 1.50 - neither of those read my P8400 voltages correctly.

 

Unlocking VID-locked CPUs?

It won't happen because of the way Intel did the lock The only way to get lower VID-s for those CPUs is to hw. mod the VID pins themselves.

Which is a shame, really - all my desktop CPUs run perfectly stable @min. allowed VID for all multipliers, which is usually 1.1 or 1.05V. If Intel did not lock the VIDs to ridiculously high values those CPUs would most likely run stable at the same voltage as my notebook CPU

Just checked - I have 1.44! I'll download 1.5 right away. Thx!

 

No, no, it's 1.41 you want. 1.50 still fails.

 

Dunno why people find it difficult to use CrystalCPUID ?!



Just open Intel Enhanced Speedtstep Control > Enable Voltage Change > While changing voltages, open CPU-Z 1.41 or Core Temp and monitor the voltage change in them. Nobody gives a rat's about notebook CPUs, so CCPUID added only desktop-type VIDs, and don't have an option to select between Mobile and Desktop type CPUs.

After getting the perfect undervolt while stress-testing, that particular undervolt will be for the highest multiplier. You only need to undervolt the highest multiplier, because that is the frequency at which the CPU will reach its full theoretical TDP, and one needs to undervolt the highest multiplier to reduce "that" TDP. Once done, save the shortcut on desktop. When you login, just double-click on that shortcut, and the undervolt will be automatically applied.

Now I started using CCPUID over RMClock, 'cause RMClock used to screw up my overclock+undervolt. CCPUID worked like a charm, and I haven't seen a single crash since the time I started using it.

In the RMClock Tweak file, it is clearly stated that - "NOTE: Your CPU may not actually accept all VID values from the "unlocked" VID range.". So just messing up the registry won't necessarily get you a couple of extra votages to play with. Some CPUs that do not go through proper calibration and inspection in the factory usually can go a VID or two lower (or higher). If you wanna play around with voltages, get a "proper" ES.

 

Just tried it out.. i like it. Its more informative and flexible than RMclock.

Probably buy the full version.

 

It is not difficult, but it is quite limited in what you can do, at least if we are talking about Intel EIST (dunno about AMD, last time I had AMD CPU was way before undervolting was popular).

RMClock and CPUgenie offer much more options - e.g. per-multiplier voltage management and support for SuperLFM states. Also, CPUgenie's load-based multiplier management can be tweaked to give more performance or to have more aggressive power savings, etc... and could be overrided on per-application level.

Also, there is a thing that worries me - CrystalCPUID has BSOD-ed my system few times when I was playing with multiplier options, something I find absolutely unacceptable and I never saw that with neither RMClock nor CPUgenie no matter what I selected/tested (of course, not counting trying absolutely insanely low voltages that bring CPU down due to instability).

To me, CPUgenie looks like polished RMClock in terms of better usability with nice GUI, useful monitoring&testing tools and support for fract. multipliers and per-application settings. Whether that is worth $14.95 they ask for it - this is something I'd say it goes to personal choice

And.. yes, signed 64-bit driver is a plus, too.

 

Whoa, whoa, that's news to me: do ES chips really let you utilize the full range of RMClock VIDs? If I got an ES P8400, could I go below 0.9250?

 

Only >some< ES chips can do that. If they are too close to official release they might already be locked, I guess for testing sake. But you can always try with your ES, try to get lower than 0.925V and if VID monitor shows that voltage really went down, that means that you are lucky

However, if you get too "young" ES you might end up with bunch of hardware bugs that won't be in official errata - so I'd be very careful with that.

Too bad Intel does not sell "VID unlocked" chips, even as "Extreme" editions - it is quite outrageous to extort money for something which is, erm..., already there - but at least we'd have a choice to get it

Btw, I've seen some Ebay sellers selling re-packaged ULV Core 2 Duos in Socket M - that is one way to get super-low VID, but at the expense of lower frequency and who-knows what other problems due to "expert" repacking

 

Yep, only true/proper ES CPUs which do not have any hardware caps, can be tweaked in terms of voltages and multipliers. I've had 3 B1 X9000s and all 3 had unlocked multipliers and VIDs. I never did any overclocking and undervolting through software, only via the BIOS. Bugs and all are usually few, well, Intel respects their hardware testers and wouldn't send them buggy stuff to test upon. Its usually Ax - stepping stuff you would worry about, as those go through min calibration in the factory. I have an A1-stepping T6400, which dies when stressed for 2 hours straight -.-

 

Where do I have to press F9 for this to work? Pressing it when CPU-z 1.41 is the active window doesn't seem to do anything - it still says I have a 199.5 MHz FSB and 1196.6 MHz CPU clock speed. I'm pretty sure RMClock is actually giving me a 600 MHz CPU, though, as performance-wise it seems to be in the 600 MHz range when I lock the multiplier/FSB there.

Santa Rosa T7500 processor.

Tried CrystalCPUID just now. It hung my system at 1.0125V, my lowest voltage with RMCPUClock (and I did remember to disable RMClock first). So I tried it at 1.05V, and I got an instant reboot. RMClock is definitely preferable for me.

Tried CPUGenie as well. It did not freeze my system or cause a reboot, even at 1.0125V. RMClock is confirming its undervolt as accurate, and temperatures are in line with what RMClock gives. I doubt it'll go lower - RMClock caused a reboot at 1.000V - but this program does seem to have its undervolting down. Might give it a run for a few days, or let its auto-tune run and see what it thinks about me voltages.

 

I'm running a T7500 in a D630, and the lowest voltage I get in CCPUID is 1.063V - which in mobile terms is 0.950V (the lowest shown by RMClock). Well, all Meroms I've owned have had the lowest voltage of 0.950V/1.063V....

I am perfecly undervolted at 1.175V in CCPUID, which in mobile terms would be 1.0625V.

 

So CCPUID uses a different scale than RMClock? I know RMClock displays different voltages before you switch it to a mobile chipset, but I would assume that when you tell it to apply a certain voltage it's actually applying that voltage. CPUGenie seems to give 1.0125 right away when I tell it to give 1.0125.

It looks like you add 0.1125 to get the "desktop" equivalent? So if I tell CCPUID to put my voltage at 1.125, it will actually give it as 1.0125 and shouldn't crash? It just seems a bit odd that it would give a different voltage than it says it's giving - not sure how much I like that.

CPUGenie seems to be working well. I've noticed a couple different things:

*I can overclock above the stock voltage with CPUGenie, which I can't with RMClock. I'm not sure quite why I'd want to, but it does seem to be working. At least the temperatures are higher. RMClock says I'm still at 1.2375, but the temperatures are a few degrees higher at the 1.25V CPUGenie says I'm at than they were with its stress test at 1.2375V.

*CPUGenie is giving me higher temperature readings than either RMClock or I8Kfangui. The latter two tell me my hottest CPU is at 91C right now; CPUGenie says I'm at 96C. There's a difference at lower temperatures, too, though not always as much. I'm not sure what the reason for this discrepancy is, nor which is more accurate. But I am pretty sure I don't want to overvolt!

*If I "Unlock voltages" I can also go below 0.85V (my usual minimum). At SLFM it doesn't crash at 0.8375. But neither CPU-z 1.41 nor RMClock report the 0.8375 figure - they both still say 0.85.

*Sometimes when I close a CPUGenie window, it completely exits the program and I get a "did not close properly" message when I restart it. This has happened even when it's not performing power/voltage management.

*Looking at this article, supposedly the tested E5200 Pentium Dual-Core had a maximum temperature of 56C both at 1.2 GHz and 1.025V, and at 2.2 GHz and 1.025V. That seems a bit fishy to me - frequency should have an impact on temperatures, too. Not as great as voltage, to be sure, but nearly doubling the voltage should have some impact on the temperature. And it did if you look at the original temperature column. 1.3 GHz with 1.1375V gives 59C, while 2.2 GHz with 1.1375V gives 76C. I think they just copy-pasted the first result for the CPUGenie Temp column.

 

A lot of programs have different temps and the reason for this is Intel's reluctancy to publish TJMax values for all of their CPUs.

Intel CPU temperature is calculated by reading one specific MSR (register) and subtracting that value from CPU's TJMax constant (this is the maximum temperature above which CPU halts execution).

Problem is, TJMax for Intel Core 2 CPUs can be 85C, 100C and 105C, depending on the model and stepping. E.g. most new 45nm parts have TJmax of 105C.

Now, the only proper way to do it is to keep the table with all TJMax values and use the right offset - this works for mobile Intel C2Ds well, because Intel seem to be publishing their thermal limits - but for desktops, for many models it is just guessing.

So, if RMclock is using TJMax of 85C for your CPU and CPUgenie is using 105C, there lies a difference. Also, some sources claim that you need to apply some offset as well, which is not verified... but u get the point, this is really "trial and error" kind of stuff thank's to Intel's paranoia

Then it is most likely not working - btw, CPUgenie also has its own voltage monitor and it will also show that voltage is not changed in case of VID lock.

 

This is the best undervolting program. Plain and simple.

 

It looked very promising, but unfortunatelly just before finish, at the 9.5 Multiplier testing, BSOD came and after restart I got these errors...many hours of testing without success ,(

 

Does it even support half multipliers?

 

That's out of question if I was running automatic voltage optimization wizard, don't you think?

 

What makes you like this program better than Rmclock? Even though it supports half-multipliers, I still like rmclock better, mostly because it consumes less system resources (6mb for rm clock, 38mb for cpu genie on windows 7 64bit)

 

I have downloaded cpugenie and it seems to adjust voltages correctly (verified by cpu-z 1.41). However I couldn't manage to disable P-state transitions.

What I would like to do is to keep my processor at the highest Frequency level (that is 2.53 GHz for a T9400). However when the CPU is in idle, the frequency goes down to 1.6 GHz. I would appreciate any help...

Thank you,



-

 

Well, it is better for me only because of half timers. I also think the voltage indicator is more accurate on this program.

 

I noticed this behavior only on default voltages with Intel's 45nm mobile penryns. Even if you set the highest multiplier - e.g. 2.4 GHz, sometimes the frequency goes down to e.g. 1.6 GHz without any way to prevent this by software.

But, if you decrease the voltage for 2.4 GHz (in my case, from 1.137V to 0.987V) - frequency stays unchanged.

This behavior is the same with CPUgenie and RMClock. I guess this is some kind of "backward compatibility" thing for boards without proper ACPI P-state tables, so they can also save some power.

 

Hi,
same problem here. Everything works fine, but everytime i restart my Notebook my config gets invalid and CPUGenie starts at factory defaults ... Compared to RMClock half multipliers are working for me, but because of my "invalid config" its not an option ...

 

Thanks for the reply. + rep

I did set the voltage to 1.100V at 2.53GHz (the default was 1.26V). However it did not work either. It is still jumping to 1.6GHz when the CPU is idle.


--

 

It's called Intel Speedstep. To force the maximum speed (using RMClock), just make it use the Maximum Performance profile (and check the highest multiplier you can in that profile).

 

Wow CPU Genie is actually working for me

At 1.250 volts, the temperature is a constant 73 C when stressed, I just tried it at 1.175 volts, and it's staying between 65 and 67 C. This makes it better than RMclock. It WORKS.

Edit: CPUz 1.41 and CoreTemp both read the same temps as CPUgenie!

Double edit: It's stable at 1.0 volts!! 57 C!!!!!!!! (that's with a 5 minute stress test. 8 hour stress test coming tonight.)

 

No news and/or updated version?

 

Tried this out last night. Nice program. Clear design. Easy to use.

I ran the 3-minute per voltage setting test (overnight), which ran without a hitch. This morning, I did a 30 minute test on the voltage that the overnight test suggested for my maximum speed, and the test failed. So the program isn't kidding that users shouldn't set voltages based only on the overnight test.

Anyway, all in all, I like the program. The testing wizard seems stable and much less labor intensive than doing it manually. If ever I don't need my computer for more than 30 hours, I'll run the standard test and use its settings.

 

So, after a few days of playing around with this, I have uninstalled CPUgenie and won't be purchasing it any time soon. I've mostly done this because undervolting doesn't really give me any more battery life with my t9300 core duo. The only significant improvement that I get is a heat reduction at max load. Since my CPU just doesn't work that hard on a regular basis, it's not really enough to warrant keeping the program.

My overall assessment of CPUgenie is as follows:

Good
-----
1) Undervolting wizard -- automates the process of finding proper voltages with a minimum number of bluescreens (I had none). Run the standard or safe test, though. The three minute per multiplier test, as the program says, is only for finding ballpark numbers. If you want to go more quickly, run the three minute test, then run individual 30 minute tests on the voltages that it finds to test full stability. I found that the next higher voltages were perfectly stable.
2) Half-multipliers -- full support for half-multipliers
3) monitors -- excellent real time monitoring of heat, voltage, and load
4) stress test -- built-in stress tester seems to work perfectly
5) GUI -- nice looking GUI, mostly intuitive

Bad (or at least not yet great)
-----
1) power management -- this is a little buggy; doesn't seem to mesh correctly with standard OS power management settings; clunky interface with OS power management
2) startup settings -- no matter what you click, the program always starts automatically with Windows upon restart

Final word
----------
I'll keep my eye on this one and would keep it in a second if it were free. I'd still love to have a program that will undervolt at half-multipliers without issue. I can't get CrystalCPUID to work reliably (sometimes it works, sometimes it works then all of a sudden sets the voltage lower than I want, sometimes it gives me a bsod). RMClock seems to integrate well with the OS power management. But it doesn't support half-multipliers, and the work-around (disabling IDA) seems to overclock the t9300 to 13x on occasion. This probably isn't all that big a deal because CPU-z reads that the chipset actually hits 13.5x on occasion with standard BIOS settings (can anyone explain this to me?), but it makes me uncomfortable, especially when it's undervolted.

So yeah, good program, just not flawless, and thus for me, not worth $14.99.

 

Version 1.1 seems to be out - looks like they've been fixing some issues:

From their forum announcement:

Changelog:
==========

Version 1.1

- Improved CPU driver performance, less time is spent in kernel mode
- Improved CPU driver robustness, CPU voltage/frequency change is now atomic and cannot be interrupted by process scheduler
- Fixed a bug where CPUgenie CPU management works but start page reports that CPUgenie is not managing CPU
- Added “Launch Power Monitor” context menu command
- Fixed problem where CPUgenie dialog might become black when Aero compositing is enabled
- After testing of a voltage when CPUgenie does not manage the CPU, Windows CPU driver is enabled
- Fixed a bug where old voltages were shown in UI after Voltage Optimization Wizard successfully ends
- More robust min/max VID/FID probe routine
- Fixed a bug where CPUgenie voltage tester would crash after computer restarts because of low voltage
- Fixed a bug that prevented CPU frequency management under some configurations
- Reduced Memory Usage


Version 1.00

- Initial Release

 

^^^ Thanks for letting us know 1.1 version...

Seems like they have fixed all the bugs that I have encountered before. Now I can run my T9400 at full speed (2.53GHz) with 1.100 Volts.

I will use for a few weeks and if I cannot see any more bugs, I will purchase it... I think it deserves $15 bucks...


--

 

It deserves more then $15. The program is simply amazing. BTW: I reported those bugs. I am helping them out and finding problems.

 

I run my T9400 with 0.986 volts @ 9.5x multiplier (2.53) without any problem. Temperature went down about 18C.

9.5 ... 0.986V
9.0 ... 0.925V
8.5 ... 0.900V
8.0 ... 0.875V
etc.

 

you are lucky... mine crashes at 1.0V-2.53GHz.

I haven't tried the voltages in between 1.0-1.100 V. Just picked up 1.100 randomly and it works fine...


--

 

Does it actually unlock the voltages or only in theory? I was monitoring it with rmclock which was still showing 0.950V whenever I select anything below that.

 

No - "unlock" just unlocks the GUI - allowing to request lower-than-advertised VID.

However, unless you have an unlocked ES CPU, nothing will happen - CPU will just set the lowest-advertised VID.

Unfortunately, the only way to get around this limit is to hw-mod the pins. Notebook vendors are very unlikely to release BIOS-es with custom VRM control.

 

i have a QX9300 andy oddly my min is 1.050 unless i enable full range VID.

I tried 1.050 and 0.950 @x6. oddly the temps are still the same and the power consumption by AC meter is still the same. 70ws (idle is 58ws)
i'm running 10min test now. 20mins test = 23hrs.

i'm using CPUz1.51 when i press F9, it shows 508mhz,

 

I have same problem too. Anyone could solve that problem??