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

Thanks unclewebb, will test it shortly. The problem is that I can see myself very easily reaching that 120W PSU limit while gaming if the Turbo Boost does stay constant and I overclock my GPU on my G51Jx. It's true that at Stock Clocks if you never do anything as draining as FurMark then you should be ok for now - but there are cases in today's games we can do it and it'll probably only get easier in future releases...

Thanks,
Peter

 

Toggle1FC
http://www.sendspace.com/file/07d762

There was a bug in version 1.0 so here's version 2.0.

The link on the previous page was changed.

I'm used to working with the XPS 1645 guys that have ATI GPUs. What is the power usage at the wall with this new trick while using any game you have? The Dell guys could draw maybe 110 watts from the wall while gaming which is only 100 watts DC max from the adapter so they were well within spec. It would be good to see some numbers on the Asus laptops with Nvidia chips.

 

throttle off, throttle on, throttle off, throttle on.

http://i137.photobucket.com/albums/q228/HoldFire/1fc-1.jpg
http://i137.photobucket.com/albums/q228/HoldFire/1fc-2.jpg
http://i137.photobucket.com/albums/q228/HoldFire/1fc-3.jpg
http://i137.photobucket.com/albums/q228/HoldFire/1fc-4.jpg

logs attached. in a game that throttles, exact same thing. not much else to show really.

 

A picture is worth a 1000 words.

When gaming, the difference between a multiplier of almost 20X and a multiplier of 7X will be huge.

Does Asus sell any higher powered adapters that work with these laptops? That and a super cooler would be perfect.

 

Yes, the G73 150W Adapter has been confirmed to work with a G51Jx. I think G72 adapters also work.

I tested with your app as well and it has the exact same effect as thalanix's method but with less work - thanks.

Peter

 

some of the resellers had 150Watters, and generics work too. temps are a good amount higher, but nothing alarming yet.

this is SM2 test 2 and a bit of SM2 test 1 in 3dmark before (no logs for that but you can see what happens. CPU throttled = GPU cut);
http://i137.photobucket.com/albums/q228/HoldFire/3dm06-gpuzex4.jpg

and a comparison to now;
http://i137.photobucket.com/albums/q228/HoldFire/3dm06-sm21-nt.jpg
http://i137.photobucket.com/albums/q228/HoldFire/3dm06-sm22-nt.jpg

big difference. unthrottled score: 11167 http://i137.photobucket.com/albums/q228/HoldFire/3dm06-res-nt.jpg
throttled score: 9473 http://i137.photobucket.com/albums/q228/HoldFire/3dm06-w7-hb.jpg

on 580/1450/975 clocks. 3 tests on highest brightness screen/keyboard, phone charging and xbox controller + mouse plugged in. no brick freeze, so i would say even an oc like this is safe for gaming.

logs attached for the new runs.

that was the method before the easiest way possible added it to the wiki page.
it doesn't make overclocks more stable, so whatever you had before will still be the limit now.

 

A 1700 point boost in 3DMark06 is very impressive. Finally some reward for all the work you've been putting in trying to solve this issue.

Now that this has been proven, I might add this option to ThrottleStop but the way your laptop is running, you don't even need ThrottleStop.

 

Just wanted to chime in on this, it's absolutely gorgeous and a million thanks to whomever posted to the wiki!

On Linux, wrmsr 0x1FC 2 gets rid of the throttling and wrmsr 0x1FC 3 has the obvious effect of throttling immediately.

I am very, very happy this is over. Cheers.

 

Excellent job guys, thanks very much for getting to the bottom of this.

Quick question, I have everest installed for hardware monitoring and it has alert capabilities, what is a reasonable maximum limit to set for temperate alerts while running this MSR fix?

CPU > 100 degrees and GPU > 100 ?

 

Thanks everyone! I can't wait to try this out tomorrow. I don't think the 3dmark06 score is really a good indication of how much this will help since it sort of stresses the cpu and gpu separately, but to think even with that that it increased 1700 is amazing.

@unclewebb: I would love to see this integrated into ThrottleStop! I enjoy running ThrottleStop for the logging purposes. Also, is ThrottleStop's voltage adjustment usable on all cpus, and would the non-throttling features work on all core 2 duos or only the sxps 16?

One last thing; what are the stock clocks for the 260? I looked on nvidia.com and it didn't have all the clocks listed for whatever reason.

Thanks again everyone! Now the only issue left is the heat, the two big issue are already addressed, I haven't seen any blue screens, and I very eager to give this a go tomorrow whenever I can get my hands on the G51J!

+rep whenever I can again.

 

550/1375/950 c/s/m. i tried breaking 12k on 3dmark06, but no luck. only up to 11700 with 580/1450/975

the 3dmark06 score is the difference with obvious throttling and without any throttling. as you can see on the SM2 - 2 screenshot, there's still a dip that wasn't there before (both on xp and win7), but i'd rather take that over the heavy throttle any day.

to improve heat, the only way is to mod the backpanel and/or the copper block mod.

 

Thanks for the clocks, good to know (I'll start there for the overclock since those are what they already should be).

I think it would be nice for one last bios update that brings the fans to their highest speed before 84 Celsius, that's just to high for me.

I've already modded the back panel, and if anyone has copper modded their G51J I'd like to see some pictures and give it a go as well. Maybe I'll end up trying it first though, and if I do I'll post pictures.

I just can't wait to try out a [nearly] fully functional G51J tomorrow!

 

out of curiosity, has anyone bludgeoned their power brick into turning off yet, or am i the only one?

 

I think I did that to my Dell XPS M1530, but maybe not, it just crashed once while doing something stressful on the psu lol.

What exactly happens when the psu turns off?

 

the notebook turns off, and the light on the adapter dies. have to unplug and replug back to the wall. i swear mine's gonna turn into the hulk overnight...

 

I've done this I couldn't see my adapter light, but I think that's what happened.

 

My G51Jx Tests:
Full bright, Mouse, etc.. GPU Overclocked 615/2000/1600
3DMark06 All Standard W/o MSR Fix: Extreme Turbo: 10371 SM2.0: 4452 SM3.0: 4384 CPU: 3046 - Max Power draw: 126W
3DMark06 All Standard With MSR Fix: No Extreme Turbo: 10427, SM2.0 4327, HDR/SM3.0 4517, CPU 2895 - Max Power draw 129W
3DMark06 All Standard With MSR Fix: Extreme Turbo: 10604, SM2.0 4356, HDR/SM3.0 4542, CPU 3101 - Max Power draw 130W - CPU Max 73*C, GPU Max 86*C

At the start of the 3DM06 test with Extreme Turbo, I had both my Laptops plugged into the same 120VAC 15A Travel Power Bar/USB Plug on my desk so I could see the Power Meter and record it and the screen display at the same time and after having this setup for 2 weeks with no problems, my Z96JS suddenly went to battery mode and wouldn't go back to AC until I plugged it in elsewhere. I've been very careful and haven't let the power draw go beyond 135W though so my adapter/G51Jx haven't turned off as of yet.

I did a test with the MSR fix and the GPU Overclocked to 630/2000/1600 - and the GPU crashed about 30 seconds into test #1. So I did another test without the MSR fix, Camera disabled, ODD disabled, no USB Mouse, Extreme Turbo disabled, but still with full brightness with the same clock speed - the results were: 3716, SM2.0 1362, HDR/SM3.0 1378, CPU 2868. Big help there . I will reboot and re-run the test to see if it was because the GPU was in safe mode after the crash and that's why the results were so low.

On my G51Jx anyway I get a feeling that the GPU is getting throttled more than the CPU - might be the drivers, still waiting on the first nVidia update. It could also be power, but it has another 5W it could draw on. Wish there was a way to underclock the CPU to see if there is a limit - or some way to remove its limits as we did for the CPU here. GPU overclock at 630 was fully stable on 3DMark Vantage - but maybe with 3DMark06 it just can't handle it well even with the power available.

I think though that with games/apps that test both the GPU and the CPU at the same time, 150W will really show an improvement in overall performance

Peter

P.S. Ouch thalanix, your throttled scores were much lower than even the stock ones from ryzeki's review http://forum.notebookreview.com/attachment.php?attachmentid=39693&d=1256323060 - I guess part of it was that you throttled it to the max where as he did his test as the system was with alone - glad the fix showed such an improvement for you - you deserve it

 

Unfortunately, Intel got rid of voltage adjustment on the new Core i CPUs. ThrottleStop VID voltage adjustment only works with Core 2 based CPUs.

The Clock Modulation, Chipset Clock Modulation and the multiplier adjustment features of ThrottleStop should work on any Intel Core 2 or Core i CPU.

When the Dell power adapters reach their power output limit, they simply shut down and then you have to power cycle them off and on by unplugging them so they start working again. It sounds like the Asus adapters work the same so during normal use you probably won't hurt them. Furmark + Prime95 is going to push a 120 watt adapter to its limit and maybe beyond depending on the power requirements of the rest of your laptop and what you have plugged into it.

Edit: I'm not sure about Asus but the power rating on the Dell adapters has nothing to do with the amount of DC watts that the adapter is allowed to put out before throttling starts. The new 130 watt adapter is limited to a lot less than 130 watts DC so it's hard to judge the throttling point by the rating on the adapter.

 

Good idea - I think I'll give that a try to try to get the most out of my GPU

Exactly - I think we should collect all the data on performance increases, lack of damage, how stable we are at stock clocks but that we can't get the most out of the system even at the stock clocks without the extra power overhead that the 150W provides. Right now even at stock clocks I can just about run out of power even just running 3DMark06 - if I had some things plugged in that would be a definite system shut down if I had my PDA, external HDD and other things plugged in to those 4 USB ports they provided to us.

Peter

I noticed that - we were suspecting ASUS had us limited on our 120W to a limit of 90W-100W depending on whose estimates we went by in terms of actual power use (either by estimating based on battery power, full load on devices and TDPs or with the other method of using a Power Meter and then taking the minimum efficiency into account to find how much power was actually being provided). That was before throttling happened - we could max out the adapters doing other things like burning DVDs, hooking up USB devices, etc...

 

the throttled scores were on an almost maxed oc. compared them with the MSR change to show that it does turn the throttle off. regular use only had throttling in Dolphin-emu and furmark.

i see MSR change lowered SM2 scores for you as well... interesting. looks like i won't be breaking 12k. the ultra-low scores are because of the driver crash . taking off the power limit had no effect on the cards themselves. if the bricks are ok with getting hard resets, do a prime95+furmark run, just to see how much it's putting out. i was able to get 3-4 minutes out of it while oc'd, so i suspect you'll be able to hold a little longer. seeing a 150W+ figure will bring tears

i can run 590 stable on vantage but not 3dm06, MSR or not. more than several runs with 580 clocks and not a single PSU lockup, so i think 135 is understating the limit by a good amount.

 

Well i have been playing around with throttle off and here are my findings on my G60 ( currently I have only one hdd -> Disco SSD Intel 80GB X25-M G2 )

(120W PSU)

IDlE -> 49 W
Full Load -> 173 W

When on Full Load I had turbo mode on, mouse keyboard and zalman ZM-NC-1000 on max speed plugged in , full brigthness on monitor and keyboard leds.
No PSU shutdown



3dmark06

Max -> 117 W



Was expecting more

The readings from 3dmark06 and on Idle appear correct, don't know why the 173 W

 

is full load with prime95?
i wouldn't be surprised if the brick efficiency goes into the cellar as it reaches it's max.

 

yes

http://img22.imageshack.us/gal.php?g=ohoh.png

12k yupy , but due to resolution limitations i can't call it a win yet

 

On my G51Jx I ran a FurMark+Prime95 test with full brightness, Extreme Turbo, Wifi+BT and 615/2000/1600 GPU overclocks. I shut off the test after 14 minutes, there was no system shut down by the PSU - though it was quite hot it still wasn't what I would define as 'burning'. This is definitely taking it beyond spec though - 137W wall draw would be minimum spec for this adapter to give 120W in DC output.

The peak wall draw was 165W - I never saw anything beyond 144W sustained though for more than a few seconds, it often went up to 159W, 160W in spikes and never below 130W.

The GPU peaked at 94*C at the end of the test after a pretty steady rise. The CPU maxed out at 74*C at some point during the test.

Throughout the test the Turbo Boost readings were going up and down, but they remained at 2.97 GHz/2.82 GHz for the majority of the test I believe.

Peter

 

Bleh lost my post - trying again. A few minutes after doing the FurMark + Prime95 tests I decided to bring up Oblivion (one of the two games I installed to try to repo the 0x124 BSOD issue) and see what the Power Usage was like. Oblivion at all max only brought the system up to about 121W with other USB devices plugged in and keyboard backlight, full brightness, etc.... and within 4 minutes I got a BSOD- yay! But then I looked and it turned out only be a 0x00000109 BSOD error instead (booo!).

Looking up the error I came up with "A bugcheck of 0x109 = CRITICAL_STRUCTURE_CORRUPTION and tells us that the Windows Vista Kernel has detected corruption within itself - and it is notifying you of such via BSODs.

Somehow within your system a driver has managed to modify critical kernel code. The interesting part about seeing a 0x109 here is that there should not be any 0x109's within Vista x64 b/c kernel patches are not allowed in x64 unless they come from Microsoft via a hotfix."

Will see if it happens again.... otherwise sys is going well, but I think we definitely need 150W adapters or we're going to wreck these puny 120W ones if we keep pushing them this hard

Peter

 

Can you post a screen shot of ThrottleStop when testing? When the top Clock modulation and multiplier boxes are not checked in ThrottleStop it will be in monitoring mode only. How high does the BCLK go with these laptops in Extreme Turbo mode?

 

Here's a couple of shots with the ThrottleStop app running while testing. The last is a quick snap I took with my DSLR while running the first 14 minute test earlier.

http://www.stuffopolis.net/UploadFolder/ThrottleStop1.jpg
http://www.stuffopolis.net/UploadFolder/ThrottleStop2.jpg
http://www.stuffopolis.net/UploadFolder/ThrottleStop3.jpg

Peter

 

Seriously think this is 1 bug I do not want to get fixed
Don't want my p.brick to burn .. and I like my pc in solid form not in liquid form as it melts away.

 

CCz_Cataphract: Can you post a screen shot of CPU-Z while testing? I'm interested to see what it reports for BCLK. If that widget is reporting 2.96 GHz then the BCLK must be overclocked beyond the 133 MHz default.

Edit: I think it is probably at about 141 MHz. If so, the 2.96 GHz is a little overly optimistic. It's very difficult to get an average multiplier beyond 20X during any 1 second sample period. You need to sample for a very short amount of time to ever see the full 21X that an i7-720 is capable of. Background activity is constantly waking up other cores and bringing the average down. Intel's Turbo documentation recommends a 1 second sample interval.

sama98b: These tests with Furmark are a little on the extreme side but I think for normal gaming this is going to be a great fix without burning down too many houses.

 

I read the title and it says throttling is solved, but what's the solution?

 

Hi guys, I just got my G51J back from RMA, they replaced the motherboard in it and I think something is messed up with my CPU temps after that... after running furmark+prime95 for just about 1 minute the THRM and CPU temp were reaching 80 C How could I simply check if the thermal paste on the CPU is not screwed up? Maybe they put it badly and I shoud send it back to repair this. After running WEI in windows CPU temp was strongly over 70C. What temps should I have after full load?

 

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

with or without changing the MSR? either way, doesn't seem like anything unusual.

 

that is a really great effort guys,
by the way, since I have the 920 xm cpu, does the fix also work for this processor or should I follow different steps?

 

doesn't hurt to check, probable answer is yes. keep in mind your power limit will be a lot less because of that 920.

if that's oc'd, those are some nice temps.

 

Excellent! Nice clean solution. I'm not seeing stutters during games any more. Thanks all for your time and effort.

 

It was indeed a Bus Speed of 141.3 Mhz, I have a screen shot of the Intel Gadget showing 2.96GHz and the CPU-Z app showing the BCLK but it sounds like it makes sense to you so I won't bother posting it. The Gadget remained at 2.96 GHz for a few seconds, I'm not sure exactly how often the BCLK was changing, I only grabbed the shot of it.

Thanks for checking them out - They were OC'd - I can tell you for sure about the last one, it was taken near the end of my first long test w/this fix:

By the way, a minute or so after getting the BCLK reading for unclewebb my system experienced a BSOD again - I hadn't seen any for the first few weeks with the system but that makes 3 in the past couple of days. This one was an "IRQL_NOT_LESS_OR_EQUAL" message (0x00A)

Peter

P.S. WJamesLord, what games were you seeing stutters in btw? Thanks.

 

Dragon Age Origins in the starter city had some chops.

i've gotten those kernel BSODs when i pushed my 4890 + e8400 on a 450W PSU, never IRQLs.

 

Thanks CCz_Cataphract for confirming that. Your MHz numbers make more sense now that I know the BCLK is at 141 MHz.

These messages "IRQL_NOT_LESS_OR_EQUAL" can sometimes be overclock related. Might want to try running at the default 133 MHz BCLK and see if that makes any difference.

 

That's about par for the course. Anything above 93 is probably starting to worry, if it hits 100, quickly RMA it

 

It's good to know that throttling only happens in two cases, that most people won't encounter (at least, not in the near future).

 

and any game while overclocked, and dragon age in Lothering (only from the limited set of games i play). i expect BF:BC2 to also be affected.

 

It would be nice if we could undervolt the processor. undervolt=minus watts.

There's no way to undervolt our g51/g60 processor? Asus really locked it?

 

there's no way to undervolt an i7 with software, period.

 

Thought guys with throtlestop and dell, that dell with i7 and the 90w psu, could undervolt their i7s.
Or you telling that it just doesnt work on our asus?
I have seen a post in this forums that a guy undervolted his i7 to 1.25v and the and now the comsumption is minus 10watts, i think it was in the N71 forum and a guys with that dell.

 

if you could find a link, ill have a look. from what i remember a while back it wasn't possible.

 

"if you want to know which power supply you have, just multiply output volt by intensity
for instance, I have 120W= 19V*6.32A

I managed to undervolt a little the core i7 720QM, but it seems that ASUS has blocked the posibilities of undervolting.
In load, I managed to reduce Power consumption from 55W to 45W (according to HWMonitor)

Did anyone tried also ?

Max temps dropped approx. 10°C for CPU and 4°C for GPU (Since the GPU is located near the cpu)

Didn't tried to undervolt the gpu (wainting first for official catalyst for mobile)"

Here: http://forum.notebookreview.com/showthread.php?p=5932762

I don't now how this guys has done this, but he didn't explain...

 

I tried out that small checkbox program, and it worked. I had furmark and prime 95 open, and when checked, all cores (observed in hwinfo32) read 933 MHz, and when unchecked, bounced around to all different (and much higher) frequencies.

I can't wait to get some benching done now. This is also with insanly high gpu overclocks. I had to put them down to 575/975/1400 for heat reasons though. It easily went past 100 Celsius with the higher overclock, but it wasn't throttling!

 

no undervolting there. 45w is what it's at be default.

 

Maybe not. But would be nice, the problem is if it is possible or not.

 

It's still interesting though - my HWMonitor always shows 55W and I haven't been able to get it to the correct value. I'm kind of curious how he did it. Not saying he actually undervolted or changed the power - but maybe fixed HWMonitor to give the right result which would be nice.

Every thing I've ever seen has said the i7 can't be undervolted. Of course what we're doing now is rather novel as well - who knows, maybe we can find a way

Peter