An Analysis of the Asus G51's Temperature Issues

Okay, basically I've seen alot of people saying how the Asus G51 is running really hot and so thus, I investigated reasons as to why that may be occurring in hopes of illuminating why it is appearing so.

All G51 owners should know that when you start playing a game, the fan starts off at a somewhat medium speed but then kicks in after a while to full blast before going back down to medium. This repeats throughout the process. It really isn't very noticeable but after a while you get the general pattern.

Anyways, I have used GPU-Z to log my GPU temperatures every second and then played the following games:

1) Crysis at 1366x768 with very high on all settings, no AA. My GPU was overclocked to 576/925/1441.

2) GTA IV at 1366x768 with everything maxed out. GPU was running stock.

And here are the temperature plots:



Crysis at those settings with an overclocked GPU is basically the most that we can realistically expect the laptop to be pushed to nowadays. Thus, the maximum temperature of 93C here really isn't too bad at all and probably wouldn't be achieved with any other game, even at max settings.

We see that temperatures quickly rise to around 91-92C and then it achieves an equilibrium there. Note that the slope significantly decreases once the temperature hits 89C. This indicates that the maximum fan speeds get turned on at that point. This is basically what we expect and really isn't too surprising at all.

GTA IV, however, is a different story. We see that the temperature hits 89C pretty quickly, causing the fans to go at max speeds. However, the temperature output of the GPU is not big enough for an equilibrium to be established. Instead, maximum fan settings cause the temperature to decrease until around 80C.

At that point, the fan decreases in speed. Unfortunately, the fan speed here is too low and the temperature creeps back up to 89C, turning on the max fans and dropping the temperature down again, creating an oscillation effect.

What is the significance?

Most people here who use HWMonitor only really care about the max temperature that it lists which means that in most games, the max temperature displayed on it only occurs for a fraction of the time spent gaming. Instead, most of the temperature is between 80C-89C in an oscillation.

What are the implications?

Should someone be able to turn on the fans to maximum speed at a much lower temperature, lets say 70C, then the maximum temperatures would be much lower. We can probably expect that with most games, an equilibrium of somewhere between 70C-80C would be achieved. Those temps would be very good.

Now, that still won't prevent the temps from going into the low 90s with something like Crysis running with an overclocked GPU, but if you don't push your GPU to the limit with every game that you play, the average gaming temperature would be generally much lower.

How do we turn on the fans at lower temperatures?

I've done alot of researching and the answer is that we cant right now. There exists no 3rd party tool out there that can control the fan settings and at the same time, is compatible with the G51.

That means that it would require a BIOS update either from Asus (unlikely) or custom written by someone to change the fan settings.

 

good analysis TevashSzat. that's the idea i had from the start with the g51. but i got replies saying that it wouldn't keep temps lower, but only take a longer time to reach ~90c. but from your analysis, they are proved wrong. i really hope someone makes a bios mod. the g51 seems like a really great gaming laptop. it already beats the sager np8662 due to the fact that the cpu can be overclocked and it has lower temps with a quad. asus do something right and make a bios fix!!!!

 

i tried this with my g50v. i dont remember what i used, but if you search for threads containing dsdt + acpi and my name, it will show up. i didnt post any results when it did work (probably should have, but no one followed it), though i did edit the post with milestones.

increasing the fan's RPM earlier didnt reduce overall temps, it just took longer to get to it.

i dont know the mathematical formula behind it, but;
1) the fan can only move some volume of air per minute
2) the volume of air will have some amount of heat energy

in a semi-closed system like that, the net temp of the air in the case (which determines the temp of the hardware) will happen when hardware heat (TDP) = heat energy expelled (function of volume+temperature), which will happen at a certain temperature. that temperature is around 90C +- room temp at the highest fan speed (for my t9400 @ 2.95 and 720/1800/1035 9700M GT).

 

You think that the GPU runs hot in the G51 try a G71. I have the unique advantage of owning both the new BB models and being able to test them side by side in the same environment (room temps etc) and the GPU in the G71 runs hotter than in the G51 when playing the same games at the same settings.

Ive been running them against each other and it was one of those things that made me stop and scratch my head because I would not have guessed it. However I have tested them a few times and each time the G51 GPU ran more than a few degrees cooler and never maxed as high as the G71.

CPU is a different story as the G71 CPU ran a few degrees cooler, which again makes the issue with its GPU running hotter even worse, as the G51 can keep the GPU cooler with a hotter CPU.

 

Well that would be the case if your temperature before the mod was like my Crysis graph, which reached an equilibrium. Turning on the fans earlier in that situation would only make the temperature reach its max slower.

For most games however, the max fan settings is enough to push the temperature below the threshold for medium fan speed. That implies that with max fans, you can reach an equilibrium at below 80C but you never do since the medium fans kicks in before you reach that equilibrium, bringing the temperatures up again until the max fans get started.

 

Why not try crysis on stock and make a comparison graph?

 

Well....I'll do so as soon as I have the time to since I have to play for around an hour nonstop to get good graphs and I usually end up playing in bursts since I'm like multitasking during gaming sessions....

Anyways @thalanix I'm really interested in what you did and have been reading through your posts but I still haven't really understood how you edited the dsdt/acpi stuff. I'd highly appreciate it if you can give me a few links to either programs or resources that you used.

 

Good analysis which shows that htis idea can help in some games to lower the temp.

BTW, is there a 3rd party soft which could be used to edit the BIOS. I dunno but isn't it just a matter of changing a value from "80" to "70"...

 

nice post I was wondering about the temps for the G51, but how about u try the 186.XX official driver. Some ppl say it runs cooler =/ or you could try the DOX 182.05. Then again im not sure which one is better for the 260M. This is just a suggestion ^^

 

Well, NiBiTor is the one usually used for vBIOS but laptop fan controls are controlled by the BIOS....I've been looking at thalanix's attempts with the G50 and I think that it may be possible but it seems horrendously complicated given my knowledge of programming and such.

Well, switching drivers would cause a maximum change of maybe 1-2C which really isn't too significant. If I can find a way to change the fan settings, I could potentially drop the max temps under most games by 10C.

 

another possibility hit me: nibitor has the vbios "compensation value" (rep. temp = real + compensation) value locked in the gui, but i know it can be changed with any hex editor. since it doesnt look like a critical piece, i wouldnt mind to see what happens if you flash after increasing it (say, from 8 to 18).

i still cant find what i used to extract the dsdt table, but im fairly sure it was some unix program. once there, i just went through changing every "set" of values to lower. the worst that happened was xp/7/ubuntu/bt4 refused to boot with the new table, but it was still possible to go into the bios and reflash a working one.



heres a better method, just make sure you change the path to the key;
http://eddietse.blogspot.com/2008/03/patching-dsdt-table-for-8510p-to-reduce.html

^- would be a non-permanent windows-only solution (i dont see crysis on linux, so no need for a full bios mod) with 0% risk to damage your system. worst come to worse, just F8 at start and "last known good config". if someone follows that, attach the extracted dsdt so everyone can have a look.

no need to be a programmer, but some logic would help to determine what is what in the 15,000 lines of values. most of if will be mechanical "look through -> 'this looks cool, lets change it' -> change the table -> reboot -> check if it worked".


another necessary edit: when i was changing it, i don't think the starting value was 93C. im pretty sure it checks the raw values (not the ones you see in hwmonitor), ie. processor temp is some value from limit and gpu temp was the real value and not the driver-reported value.

 

If you do figure out how to change the fan settings this will keep a lot of laptops cool, but then again why hasn't this already been implemented by the manufacturers who make laptops? It doesn't make sense there must be a reason why it is not done in the first place. Maybe to prevent the fan from running too long and wearing itself out that will cause its overall performance to die down and your GPU might no longer have a fan to keep it cool anymore =/. Therefore causing alot of ppl to replace their laptop fans of their GPU when they overheat n fry XD.

Hey im just thinking outside of the box though. The fan idea is still good XD

 

electric motors are designed to last a long time. unless you change the circuitry to "overvolt" the fan (which is a solid idea for the brave), having it run at full shouldnt hurt. spinning up is the most stressing part.

 

oh ok that sounds true. Then again as I said before its a wonder why these gaming manufacturers didn't even think of doing it in the first place =/

 

it would lower the temp a bit in non-gpu stressing games, and hopefully stabilize it like the crysis graph. constant 90C is better for the gpu than oscilating between 82 and 88 (google: thermal cycling).

why they didnt implement it, i dont know. maybe to prevent unnecessary noise which sometimes could be a big issue, especially since over-passive fan profiles on linux-based machines can and do have the gpu hit 70+ (backtrack 4, im looking at you) [edit] before even turning the fan on.

 

Probably mostly because they didn't want people to complain about the fan sounds and hoped that the average user wouldn't check at the temps alot.

Yeah....the thermal cycling issue sounds like it could be of concern. Anyways, I've found this site which appears to be able to extract the dsdt tables....

Haven't really spent that much time into looking into this since I am really busy at work and thus, at home, right now.

 

I just registered to say that I think manual fan control is a great idea, I am eagerly awaiting someone to do the hard work and make this possible for the public

 

is there any standard of the temperature for a gaming laptop?
sth like: under 75C can be regarded as safe, higher than 8X C should be dangerous?

 

There really is no standard across the board. You have to compare like models.

If you hit, say, 90C on a netbook's GPU (pretending that they have GPUs) that is bad.

But hitting 90 on a G51? No sweat. I'd go so far to say that 95C on a G51 is no sweat. Just try to let it cool down every hour and half, if you're so inclined.

Anything below 85C for a G51 is considered artic-level chilly for the GPU.

 

being the lazy kid that i am, i just smooshed some arctic silver 5 on my cpu and gpu and my temps 93 max playing games. i need to actually go in there and properly apply the thermal compound to see what i can get my temps down to.

been playing a crap ton of TF2 and L4D.

 

i would say unless necessary, dont. thermal cycling is the killer, not temps.

if it has unix source, then maybe that's what i used. either way, should work.

 

Really?

I mean, extreme flucuations--in your example, you talked about 88 and 82 and back again. I'm thinking that an hour and a half isn't too "soon" to let your GPU a rest.

Is there an ideal cycle? I do know that keeping it running in the 90s range for four hours definitely won't help your component life.

 

I actually think a stable 90 is better than the oscillation he's seeing. Bumps contract and expand, man

 

im sure most have seen the "8600m gt died" or "g1s died" threads

and another thing to note: if anyone plans on a soft-dsdt mod, make sure you change the lower limits as well.

 

in GTA, could it be because some areas are not as graphics intense as the others? I mean crysis is intense everwhere.

 

Well, that cycling can be seen in Assassin's Creed and Mass Effect as well. All of the games that I've been playing except for Crysis simply haven't been able to significantly stress the GPU yet even at stock clocks.

 

I think what you meant is stress the GPU "At All Times". Most games have areas that are easier to render. Example: The Normandy in Mass Effect vs The Citadel. I can probably think of some place in AC but that game is so God damn boring, I gave up.

 

So it's not just me eh? It was a mind numbing experience to play Assasin's Creed. If I wanted button smashing fights I'd play tekken, and otherwise it was a VERY dull game. Amazed at how it had all those sales.

 

so, u mean we shouldn't OC both GPU and CPU?

 

im all for oc'ing, the 8600m's had a manufacturing problem (bad bumps/voltage regulators) that died early from (supposedly) thermal cycling.

 

No, it just means you shouldn't overclock carelessly.



Anyway, some temp data. Using the the 3DMark06 configuration in my signature, I hit 79 on my CPU and 97 on my GPU. Note there is no undervolt, and this is the hardest I've pushed my CPU. The notebook is mounted on top of a Zalman NC2000, which is not active.

My temperatures show that I stay at the 95-97 range for about fifty seconds. I stay above 90 for less than two minutes.

I'll try to get some Crysis tests in, since 3DMark fluctuates as it loads tests.

 

You can't really use 3DMark to see the fluctation in temperatures since not all of the tests will stress the GPU so you will get naturally fluctuations just from the presence of CPU tests.

 

Well, ran some Crysis. Temps maxed at 96 after forty-five minutes. (This is from the beginning of the Paradise Lost chapter to the start of the Ascension chapter. 1366x768, all high except for very high textures, no AA.) CPU is X9100 undervolted to 1.250 (I think I can go lower, but I'll get to that later), at stock 3.06 GHz clocks. GPU is 565/1410/970.

I hovered mostly around the 94-95 range with ten-second spikes at 96 here and there. This is on top of the Zalman NC2000, which is off.

 

Interesting undervolting data.

When I got my X9100, I ran a stock 3DMark06.

1366x768
CPU: 3.06 GHz, multiplier at 11.5, voltage at 1.350
GPU: Asus stock
Score: 11203
Max temp: 91C

I undervolted and undermultiplied (is that a real term?) and got a very good score still.

1366x768
CPU: 2.793 GHz, multiplier 10.5, voltage at 1.175 (Clock detected in CPU-Z; 3DMark06 still shows 3.06 GHz)
GPU: Asus stock
Score: 10945 (-258 decrease)
Max temp: 87C

I might not be able to pull a 10000 at 1280x1024, but, damn, that -4C is really nice.

 

Your memory is high in proportion to your core/shader. You will get a better score with 600/1500/900.

 

I'm running it at the Asus stock [under]clocks, which are 500/1250/799

The standard clocks are 550/1375/950, which I assume are perfectly stable, and I just increase in increments. I take each number and, say, add 2.5% to that number.

I'll try another run with different clocks, then. I was assuming that the GeForce standard clocks are where I should build from, but I guess not.

EDIT: Hmm. Using the exact same CPU setting and using 1280x1024, I get 11964, versus my 12139 using 605/1510/990

Then again I finished the trial at 95C, two degrees cooler, for a loss of 175 points.

 

http://forum.notebookreview.com/showpost.php?p=4943629&postcount=627
http://forum.notebookreview.com/showpost.php?p=4971113&postcount=684

 

My 605/1510/990 ran stable through a 3DMark06 test with no artifacts or crashes.

So, from what I gather, I crank up core/shader until I get artifacts (I guess starting out at stock 799?), and then start raising the memory clock until I get artifacts there?

 

I would use 900MHz as the base and start working your way up from there. I am about to say stop once your GPU hits 88*C, but I know you Asus guys love cranking your cards into the 90's so I'll let you have your fun. Just don't surpass 95*C. Keep your core/shader ratio at 1/2.5.

For example, start with 600/1500/900, then raise the core/shader from there and see how high it will get until it becomes unstable. Then try 620/1550/930 and raise the core/shader, and then 620/1550/940, and then 620/1550/950, and so on. However, I expect that you will probably hit a ceiling around 640/1600/950. Gauge which set of clocks will allow you to push the core/shader the highest as those two factors are what boosts your FPS. The memory does almost nothing for you since 1366x768 doesn't need that much bandwidth.

 

I'll try it one of these days. I don't need that power, since 565/1410/970 runs Crysis well, though I still get some mad heat. Maybe... 94-95 with an undervolt to 1.250 and a Zalman NC2000 running full blast, after an hour of high settings at native.

But if I get what you're saying, 565/1410/800-ish would be ideal.

EDIT: Saw your edit. I assume the same method would work with a minor core/shader overclock like what I usually game with. Does a higher-than-you-need memory clock cause a lost of unneeded heat?

 

http://forum.notebookreview.com/showpost.php?p=4939991&postcount=612

To add to this, higher memory clocks produce more heat, especially when you near the memory's maximum rated speed. Unless you raise the voltage to counteract the effects of thermal instability, the stable temperature ceiling for your card will lower. In other words, say your card is stable at 300/600/280 up to 90*C and you are already using the most out of the given voltage. If you raise the memory frequency to 300/600/300, you will start to notice that your card will become unstable at say 84*C.

GPU memory is the most sensitive component on a GPU. The GTX 260M uses Samsung memory rated for 1000MHz maximum as reported from the serial number on the memory chips, K4J1032400 HJ1A or similar. Anything beyond 1000MHz would cause damage to the memory that you would start to see in the long run. I suggest that the highest any G92 user would go with their memory is 950MHz for the sake of stability and long term reliability, and just raise the core/shader if they want to squeeze every bit of performance out of it. If you want to rule out memory as the cause of instability, lower the frequency to 930MHz or 900MHz and find the highest stable clocks for the core/shader.

 

This meshes well with this page which shows a max of 950DDR on MXM3 (look down a few paragraphs):

http://www.notebookcheck.net/NVIDIA-GeForce-GTX-260M.14559.0.html

 

Does the notebook itself (the keyboard and palm rest) actually get hot? or just the cpu or gpu?

 

The left side of the keyboard (CPU) and the left side of the palm rest (GPU) gets a little hot when gaming. Its minimum compared to other laptops (HP Pavillion, Toshiba Satellite and Compaq Pressario for example)