[Fixed/Workaround] Asus G51J(x) CPU throttling investigation

thanks man, I'll make sure to keep an eye on the temps : )

 

Hey guys, been following this for a while....If I install this program that unchecks the mystery bit and allows for not so limited throttling, would I need to be worried about temps if all I do is game?? I've seen you guys running furmark and 3dmark to stress this machine, but would any gaming right now push it to where it would be considered bad for the system? I'll be running at the nvidia stock clocks (not asus') and was wondering if this would help with running graphic intensive games such as Crysis, as I've noticed some stuttering there.

My question is this, would this fix allow me to game much more fluently currently, while not pushing it the GPU and CPU to extreme temps???

 

If you are running OC clocks you run into the possibility of PSU blackout, as it won't have enough power to push an OC'd card plus full power CPU etc.

The throttling did not affect as much as some may think, so most probably you won't notice any huge gain at all. I actually noticed a couple of °C less on the GPU, but that might have been a fluke.

I want to try OCing my card as I have a 150w PSU so I bet I can maintain a higher clocked GTX260m without much issues... will probably try after I make a back of my data over the weekend.

 

hwmonitor used to show 55, up until i replaced it with the most recent version.

as i've said here and on the wiki, keep your eyes on temps no matter what (if you plan on doing something that didn't work before). removing throttling = removing thermal monitoring = mo more automatic shutdown/thermal throttling.
as ryzeki said, only a few situations will see improvement, the majority won't feel a difference in games or benchmarks.

 

I've also noticed the GPU now seems to be losing performance as more power is being provided to the CPU. Please do run the tests - I'm very much looking forward to your results .

Thanks, I only got my version a couple of weeks back - but that's true it may be old, will check for an updated version right now. Update: Hmm, nope - I have 1.15 and that is the latest version listed on the CPUID website. What version do you have thalanix? - Thanks.

Peter

P.S.

No problem, happy to help. Thanks - I haven't really been doing anything to change the BCLK - even sometimes when I have it throttled and I'm just running one single CPU app without much GPU usage it would up the BCLK - though I suppose I can restrict the max Processor use and turn Extreme Turbo off to try to help.

Actually I got another BSOD (4 this week now) a "System Service Error" when I was showing some pics/vids over HDMI on the big 42" LCD while running on battery - and I don't think I'd touched the throttle app then (though I may not have unchecked it possibly, I can't be sure). So my system seems to definitely be doing some funky things all of a sudden...

 

There is no known way for any software to under volt any Core i processor. This was easy to do in the Core 2 Duo / Quad line of CPUs but when Core i7 was introduced, that feature was eliminated from Model Specific Register 0x199. For Core i CPUs, that register only contains a requested multiplier. With Core 2 based CPUs, that register used to also contain requested voltage information so you could lower it with software but that's no longer possible.

I wouldn't be using Extreme Turbo mode until you find out what's causing the BSOD issues. If you have the time I'd run a few stability tests like Prime95 Blend to make sure everything is still OK. It's not uncommon for memory modules to slowly fail and start causing issues.

NOTE: This next section is still being debated.
I'm hoping we can find some evidence to prove this one way or the other.

I don't believe this is true for any Intel CPU. No matter what user register you "adjust" the CPU will still be able to do a thermal shut down when the core temperature gets up to approximately 125C. There is no way to turn this CPU safety feature off.

Has anyone taken their CPU up to 100C yet to see if ThrottleStop reports any multiplier throttling? Unfortunately I don't have any Intel documentation about this bit so I can't guarantee that thermal throttling will still work correctly. I think this feature is also at the hardware level so no software can interfere with it but I don't have any proof. Anyone want to be a guinea pig to test this out?

What is supposed to happen is the multiplier should start rapidly cycling between its default 12 and its minimum of 7 for an i7-720. This happens so rapidly that a lot of software can miss what's going on when it first starts to happen. ThrottleStop monitors internal high performance timers so even the slightest hint of throttling will immediately show up in reduced average multipliers at full load. If PROCHOT# is ever triggered, even for only a millisecond, RealTemp will be able to report that in its Thermal Status area which monitors this bit. This sticky bit is within the CPU so after a throttling episode, if that bit was set, you can start up RealTemp and it will read and report its status.

 

Throttling solved->More heat, less power to GPU->worse GPU performance? Damn it. Would getting a 150W adaptor solve this problem?

 

As I mentioned in a previous post, since disabling the thermal monitoring, I'm not seeing the stutters that I used to experience when facing large scale fights during online gaming. As for temps, HWMonitor is reporting that the CPU is max'ing at about 70c and the GPU at about 87c with the extreme turbo setting and the GPU clocked at 520. HWMonitor also reports that the CPU is using 55 watts, but I can't say if it is accurate.

 

Hmmmm Weird, I didn't notice any drop in GPU performance when disabling the thermal monitor.

My 3dmark06 remained basically the same, and I haven't tested OC. Where could I "see" the drop in performance? Maybe since I use a 150w PSU I haven't noticed a drop in performance?

 

i use a 120 and see no noticeable difference (in games, SM2 in 3dmark06 dropped 200 points).

HWmonitor also shows 55 now too, but i swear i remember seeing it at 45... intel lists it at 45.

 

That SM2 change is what I was referring to.

I just re-ran my 3DMark06 tests with 195.62 (which appears to have installed by what the Nvidia Control Panel says, which I managed to get installed via the Device Manager) - I didn't see that difference - at least with LCD brightness at 0% and no USB devices connected, no backlight keyboard - it may be that this reduced load is why there is no difference w/120W adapter (with everything at full and devices connected, system operating in hot ambient conditions, etc... then it may require more power and therefore we'd see that SM2.0 difference again).

Code:

                                       3DMks, SM2.0,  SM3.0,   CPU
With throttling no overclocks:         9603,  3997,   3985,    3167
With throttling GPU 615/2000/1600:     10656, 4564,   4503,    3183 (126W max)
Without throttling GPU 615/2000/1600:  10633, 4543,   4508,    3172 (127W max)

Over all tests: 87*C GPU max, 73*C CPU max

The difference between the throttle and w/o throttle tests is so small that it is within the normal range of difference between runs on the same settings. These are with the G51Jx btw, we might want to mention which when we post results/experiences since it seems there may be differences in how this fix affects things.

Peter

 

Hi all, I don't know if this worked before we did the 0x1FC fix or not - I have to run so I haven't gotten to run enough tests to confirm if it affects things or not. I messaged the guy who posted the info about Undervolting the i7 in ASUS laptops in the other thread. Here was his response:

I confirmed this - if you set the Multiplier Management settings, enable Voltage adjustment and then enable the Multiplier Management then it does change things.

If you lower the voltages Turbo Boost goes dead completely.

Running 2M SuperPi Benchmark:

At default Voltages = ~38s, HWMonitor reports Max 55.04W CPU usage
At 0.975V = 59.288s, HWMonitor reports Max 45.00W CPU usage
At 0.962V = 1m 04s, HWMonitor reports Max 36.50W CPU usage - Max Core Temp over run: 54*C

Thinking that it would not work when I first tried this, I set all the values at 0.900v to try to quickly see what he said about there being no effect if you set the consumption drastically lower - my computer froze instantly when I hit 'Apply.' Just in case, I shut down by holding down the power key for 5 secs, unplugged and removed the battery to make sure I reset things completely before rebooting and it booted up fine after that.

From these tests I think we can undervolt our i7s using this method.

Link to CrystalCPUID

Peter

 

Well hooray for this fix. Just a couple of days ago I noticed throttling with *background* + furmark... was about to make a serious decision to chuck it in and trade up to a g73.

So far, I've noticed no difference in 3dm06 score (still about 11700 @ 590/1500/1040)

furmark + prime 8 thread sent temps through the roof, 3 1/2 minutes and GPU hit 100C (that's at stock), all 4 CPUs 80C, that's when I lost my nerve, toggled 1fc, CPU throttled and temps went down to 70C almost instantly, GPU steadily dropped back to 95.

Asus throttled because the cooling can't hack it. I would bet my life on that being the reason.

BTW I reckon I've shaved a couple of degrees C allround by removing the silver foil off the inside of the the bottom panel (where you access cpu/gpu/ram/hdd) - the area gets bit hotter of course but I use a notebook cooler under it so it's all good. [edited so it made sense doh]

I could get my old C2D T9300 down to .925V on idle with CrystalCPUid but as soon as I stress tested = freeze or BSOD. Instead I used CPURightmark which did the same thing but could assign voltages with P-states which worked a treat. Unfortunately no longer

---

Unclewebb, this would be great to integrate into throttlestop, but really I would be happy with something that:
- saves/remembers the value for the toggle
- minimises when started with windows

Of course I would expect disclaimers plastered all over it if it makes you feel safer that some idiot noob isn't going to melt their lappy and blame you for it

 

might be the noobiest of questions (don't judge me lol) but if I'm only running DJ software (serato) will I still have this problem? I really won't be gaming with this laptop. This is my only deciding factor in buying the G51JX-A1

 

If you're not going to be gaming much, then why buy a gaming laptop with a powerful GPU, there are many others out there that will better suit you which trade graphics grunt for stuff like battery life.

You won't notice throttling due to this 1FC thing unless you perhaps try to run a graphics-intensive game with some really CPU-heavy software in the background. (Only time I've noticed it without deliberately running synthetic torture-tests that deliberately stress the CPU and GPU to their limits was when I was playing Borderlands while zipping GBs of stuff on Ultra compression in the background with 7zip.)

 

Ur right but ihavent found a laptop with a 16GB RAM option..... I figure it'd be nice to have that cushion

 

16GB is a lot even for video editing, are you sure you'll see a difference in audio?

the crystalCPUID results above look a lot like they're just killing the multiplier, but i'll wait and see before making a final judgment . time i caught up on some gaming.

 

Could be - I'm getting similar behaviour between ThrottleStop and CrystalCPUID. When I set for 0.962V on CCID it locks the multiplier at 11 right away - if I do the same with ThrottleStop it prevents the multiplier from going past that. Both show up in HWMonitor as 36.50W.

At 0.975V it comes up as 45W in HWMonitor, and the multipliers go up to 11.75 or so, but fluctuate. There are also options for changing the Multipliers in CCID - but I have not touched them. At 1.0V it seems to lock the multipliers at 12.

Will check out the power meter now to see if the wattage readings HWMonitor is reporting carry through to the wall draw. Personally, I don't care how it does it - if the power draw is actually being reduced significantly I might use this if I come to a circumstance where I need to use my G51Jx for work and need longer battery life. I know it wasn't designed for it and I have my netbook for day to day portable use - but it could come in handy to have it as an option.

Peter

 

Its not so much audio editing, the program requires high amounts of a raw data stream (I think low DPC its called) and I figure high ram would help....

 

That's my opinion too.

In the Core 2 design, requested VID voltage was in the lower 8 bits of MSR 0x199 and the requested multiplier information was in the next 8 bits. With Core i CPUs, they eliminated VID from this register and shifted the multiplier down to the lower bits.

Software that lets you adjust VID on Core i processors is very likely only adjusting the multiplier which is now located in the lower bits of MSR 0x199. The huge changes in performance are an indication of that. When you are adjusting voltage, you shouldn't see any change in performance.

Easiest way to test for this is to run MSR Tool and do a read on MSR 0x199, make a supposed voltage adjustment and then have a look to see if MSR 0x199 has changed. There is no longer voltage information in that register so it should be exactly the same. If this register changes then CrystalCPUID is misleading you.

 

SuperPi Mod 2M Test Set to Affinity only on CPU0
W/o Throttling Fix 0.987V clocks: 38.005s - 62W average power draw from wall throughout test (55.04W CPU reported by HWMonitor) - Turbo Boost going full force
W/o Throttling Fix 12x Multiplier: 1m 02s - 48W average power draw from wall throughout test (45W CPU reported by HWMonitor)
W/o Throttling Fix 11x Multiplier: 1m 08s - 46W average power draw from wall throughout test (36.50W CPU reported by HWMonitor)
W/o Throttling Fix 7x Multiplier: 1m 47s - 43W average power draw from wall throughout test (15.34W CPU reported by HWMonitor)

8x Multiplier 18.96W reported by HWMonitor
9x Multiplier 26.31W reported by HWMonitor
10x Multiplier 29.41.96W reported by HWMonitor
13x Multiplier in ThrottleStop (Basically Max Multi) 55.04W reported by HWMonitor

Ok, so we can see a steady curve of power use per performance. I know it's a given based on everything we worked on in the throttling front, but seeing the data I think there will be cases where it will be useful for me in this way

The power usage amount for the CPU appears to be estimates by HWMonitor as they do not match up cleanly with the actual power draw from the wall (and if you take into account power brick efficiency, and other usage, then you can see it is impossible that it is fully accurate). It changes the values instantly and there is no actual jump of those degrees in the rest of the system. If it were accurate then the jumps would have to not nearly as great. And the power left for the rest of the system - assuming the given minimum 87% brick efficiency - is 6W at max multi and 24W at min multi. So HWMonitor is not accurate really imho - but it's good to know the actual differences when we need that extra time on battery.

Peter

 

Oh and on the how useful is this throttling fix thing - I think for me anyway it'll be very useful. I use my laptop as my PVR via a USB ATSC Tuner and it re-encodes the recordings to compressed DivX files after capturing them and on my old laptop it would really slow things down and I'd often cancel it to increase the speed and thereby end up with lots of really huge HD files that I had to fix separately. Now having this throttling solved, I'm confident when I move everything over to my G51Jx that I'll still be able to game if not as before then a lot better than I would have were the throttling still in effect .

Peter

 

playing through dragon age again, that game now plays almost 100% smooth throughout with 2xAA (maybe on high enough clocks 4x). a lot of the chops, not just in that one city, were throttling related. mass effect 2 also gets a boost in some of the cluttered scenes (jacob's mission).

on battery, the greater majority of the time will be spent on 7x and won't make much, if any, difference.

 

It's good to know how it works though - For the "Maximum Performance Processor Control" and "Moderate Battery Saving Processor Control" up to 65% Max processor state in P4G it will keep the multiplier at 7x. 66%-72% 8x, 73%-83% 9x, 84%-90% 10x, 91%-98% 11x, 99% 12x, 100% = Unrestricted w/Turbo Boost. It does nothing up to 65% and then gets practically logarithmic near the end there . For "Maximum Battery Saving Processor Control" no matter what % it is, it never goes beyond 7x Multiplier.

It's good to understand that P4G has these % bars but it doesn't make any difference really - there are 5-10 states that are represented pretty unevenly by the % bars and there is no difference often between even greatly different %s unless they trip the next state (same for processor state as for brightness).

So now I know better how to utilize the utilities and when changes to them actually matter, or not.

Peter

 

without reading 100000 posts can someone tell me.... does this happen when the battery is out? is there an effect of the cpu score in futuremark. you should be getting right around 3100 points for the cpu. if it is being downclocked then the score should be affected greatly. If there a way one can see a result other then a probe telling you that it is doing this? ie... benchmark results far below expectations or the like.?

 

it will happen whenever the system goes over around 90-100W. single benchmarks won't show a difference as the CPU or GPU on it's own won't put it over that amount, but anything that uses a balance of both will. Dragon Age, BF:BC2, Dolphin-emu and pcsx2 are all perfect examples with stuttering related to throttling on a stock system.

 

you only need to read 1 wiki page...

http://g51jbsod.wikia.com/wiki/CPU_throttling

Yes benchmarks are lower that's how thalanix et al first showed it was reproducible. The CPU throttling causes fps to drop in the SM2 & SM3 tests which means it's not the "CPU Test" score that shows the hit.

Add to that Borderlands with 2 cores full of 7zipping in the background.

 

So is it safe to turn off throttling, turn on Extreme Turbo and overclock the GTS 360M to 640/2010/1610 to play games? Will it like reach 105 degrees? What's the max a CPU and GPU should go respectively?

Anyway, I decided to cancel my order and instead buy the G51Jx at an IT Show next week. At S$2998, it comes with 8GB RAM and Blu-ray! Is the RAM still 1333MHz?

 

there's no guarantee your 360m will be able to oc that high, so you shouldn't be expecting more than cataphract's. whether it's safe or not depends entirely on the ambient conditions, luck of the draw with thermal pads, and even the chips themselves. the majority of games won't be needing throttling disabled.

nvidia set 105 for the GPU and intel set 100 for the CPU. where the margin for early failure lies is up for debate.

 

Cataphract tried that overclock once and he said that it was stable. I'll try it next week when I buy the laptop. If I kill the processor and card, there's always warranty! Hehe. Maybe I shall get a notebook cooler.

 

Hey nfshp, I never said I did that - I said we should try that sort of overclock out with the 150W adapter and see if there is any effect. I'm still curious, but on my system anyway, I just can't seem to do it. It may be there is a GPU limit the same way there was the CPU limit and that since the 360m is cooler uses less power than the 260m I've been able to reach that limit already without having to have the more powerful adapter (where as G260M overclockers would stop overclocking because of temps and throttling).

I noticed that even when I under-watted the CPU to reduce the power drain to emulate a 150W adapter's power ceiling without having one, I still could not successfully complete a full run of 3DMark06 at 630/2000/1600.

I'm curious if it might have something to do with the ratios being off - the fact there is available power, it can run hotter (does not pass 84*C during the run) and there are no artifacts. I know reading up on the G51 260M overclock/optimization guide they gave specific ratios that should be followed - I'll give a try with the ratios suggested there and also the default ratio of the 360M at the ASUS clocks (since we still don't know the full nVidia clocks yet).

Will observe and report,
Peter

 

i haven't seen the 1:2.5 core:shader ratio matter at all since the g82 chips. 585:1425 is stable (not with the ratio), but 585/1463 isn't for me. the only way you can get higher clocks is slightly bumping the voltage, which won't be happening on these unless someone wants to go through and edit it all by hand (with no guarantee of working anyway).

side note: CPUID does change MSR 0x199 with the voltage.

 

Interesting, do you mean it was possible to do the MSR fix by changing the voltage settings in CPUID?

Its interesting because according to nVidia, the memory on the GTS360M is rated "Up to 2000 MHz" - and with the default ratio 615 proc clock exactly correlates with 2000 mem clock. In order to keep the ratio 630/2050 is necessary - and that is past spec - and trying it just now I was seeing some artifacts (V-Sync type and blinking mostly) along with the GPU crash right near the end.

So we know the GTS360M at those clocks is at its limit and with the 120W adapter I've been able to make the 3DMark06 score lower by running unthrottled with LCD at 100% brightness, keyboard backlight, etc... and my PDA connected to the system and charging via USB.

I think 615/2000/1600 is the ideal overclock for me without any voltage changes and has been completely stable in all my tests. I'm curious though where the 260M will end up with the 150W adapter and no throttling, so let's see when ryzeki gets to his tests

Peter

 

can someone confirm that this happens with the battery out please? I guess this is really a question for thalanix? does this happen with the battery out for sure?

 

battery in or out, charging or full. it happens when the system power draw hits a point that's way too low. this is with nothing connected, not even the mouse, at full brightness (kbd/screen).


the ratio used to be locked core:shaders, mem was only required to be as high as the heaviest shading scenes required, which gddr3/256 and gddr5/128 are more than capable to push out at stock.

 

can you remove the 2 extra memory modules and test it with only one at the same time remove the battery and disable wireless radio and bluetooth. Can you reproduce the same results given all the above?

 

whenever it goes over 90 or 100W. only 2 modules in a G51J, happens with either or both.

 

Out of curiosity, why?

Thalanix is right though if it goes over 100W from any combination of system use we can conclusively say the CPU throttles without some kind of MSR fix. The G51J (GTX260M, 2 HDDs, default 2 mem modules) has the issue as does the G51Jx (GTS360M, default 3 mem modules, max 1 HDD).

Peter

 

the throttling is connected somehow with 0x1FC and power, crystalCPUID wouldn't change anything.

on core 2's, 0x199 held the VID in the first 8 bits (al) and FID in the next 8 (ah). on core i's, VID was removed and FID was bumped down to where VID used to be (al). that's why you get a bogus multiplier under one of the settings pages.

 

can you email me please a simple way this can be reproduced using every day aps? like cad, or maya? something that does not need the aid of futuremark to overload the cpu. i may have missed it in the billion posts here but i cant read through all of them. just something simple and precise that does not involve multiple benchmarks running at the same time

 

Dragon Age @ 4xAA, usually in Lothering or Ostagar. Mass Effect 2, Borderlands, CoD4 and most likely any other game capable of getting the GPU >90% usage on stock (no vsync) while doing some multitasking (single thread compression). anything that uses a mix of CPU and GPU.

 

Thanks Thalanix - for some reason I kept reading the bits as 0x1FC until I saw the two side by side.

Why though? To report the issue to ASUS? To request 150W adapters? To explain it to clients? To decide to stock it or not? Or...?

Thanks,
Peter

 

The magic bit 0 in MSR 0x1FC is the bit that controls bi-directional PROCHOT. Turning this bit off prevents outside sources like power consumption or GPU temperature from throttling your CPU. I'm 99% sure that if your CPU gets too hot it will still throttle just like normal no matter what this bit is set to. Same with thermal shut down. Disabling this bit should allow your CPU to run according to Intel's temperature specs. MSR 0x1FC and bit 0 are not documented in Intel's publicly available documentation.

If excessive power consumption tries to trigger CPU throttling and you have this bit cleared, it is likely that your power adapter will automatically shut itself off. This will likely result in a system crash if you are at full load and it's possible that you could damage your battery. Some Dell XPS 1645 users have discovered this issue when running Prime95 + Furmark and then switching from AC to battery.

On the Asus G51J, the GPU tends to run hotter so the GPU core temperature could become your limiting factor and you might lose GPU stability or the GPU might shut down first when bit 0 of MSR 0x1FC is cleared.

 

Uncle webb is right. When i was testing my laptop with prime95 and furmarc it would crash when i unpluged the psu. So this is because the battery can't support that amount of power?

 

@CCz_Cataphract, how did you undervolt the i7? What program did you use and what's the voltages for the different multipliers?

 

Adjustable core voltage on the Core i7 is a myth. It has been proven that it does NOT work.

CrystalCPUID adjusts the register that used to contain voltage information in the previous Core 2 CPUs but the voltage setting does not exist in model specific register MSR 0x199 anymore. Software that adjusts that register on a Core i CPU will not work.

 

I wanted to be able to reduce the CPU power consumption so I could run my max overclock GPU tests so I was quite happy with the "UnderWatt" I suppose you could call it that I got by throttling the CPU with 7x or other multipliers. It does reduce the power usage (though not nearly as much as HWMonitor would have you think) but it also reduces the performance. So for my testing purposes and for battery life it's good to keep in mind - but unlike other undervolting because of the great difference in performance it isn't ideal (in typical undervolting you could reduce the voltages by some amount and still get just about the same performance with a cooler CPU and longer battery life).

So this throttling reduces the power use (W) of the CPU and it does so by reducing how much power the CPU consumes by it either requiring lower levels of current (A) or voltage (V) or both - but don't call it undervolting or it will confuse people.

Peter

 

Oh, so it's not actual undervolting!

 

Hey guys - first of all great great thanks for doing this work! You're awesome that You've managed to solve this issue. I have a question concerning that - do You think B:BC2 is affected by throttling? I'm experiencing strange smoothness in this game, it is not fully smooth like MW2 for example, but when I minimized the game to tray and checked TMonitor it was showing 933mhz on cores. Is this accurate reading which can describe the smoothness of the game connected to thrrotling?

 

Does buying a 150W adapter help squeeze better performance from the G51Jx?