HD 4000 throttling down to 350 - 650mhz after 10-30 seconds of GPU load...

[idata]

Hi everyone,

I am getting really frustrated trying to convince Fujitsu that the system they built me is down-throttling my GPU clock speed whenever it is being used to it's fullest for any short period of time.

First my specs:

Fujitsu T902 tabletPC

i7 3520M 2.9gHz-3.6gHz

16 gigs of 1600 RAM

512gig SSD

HD 4000 GPU (no dedicated GPU option)

windows 7 pro

So let's say I boot up Skyrim (low settings) and just stare into the distance to get a relative constant stress on the GPU. The FPS will start out at around 29-30fps as I would expect from this system compared to other tests and reports online of other similar builds with the HD 4000. However, at around the 20-30 second mark, the frame rate will drop considerably (12-19 fps) and stay quite consistently low (except for the rare spike up only to fall again shortly after) until I exit to menu.

If I run GPU-Z sensor software during a test like this, it will reflect the exact moment when the FPS drops with the GPU clock cycle dropping from the 1250mhz where it should be, to 650mhz. It is not a gradual drop. There are never any numbers fluctuating inbetween, it simply goes from 1250 to 650 and sometimes as low as 350. The temp on the GPU core gets up around 82-86 degrees on the average before it throttles down. It then sits at around the mid-70's for quite some time.

When using "Real temp" to monitor the CPU temps as I run this test, the CPU cores often get up to 92, 94, sometimes even as high as 96 degrees! Is this system down-throttling due to overheating?? Or is it perhaps a Chipset problem? perhaps relevant is the fact that the computer does not get very hot to the touch at all. The fan still dumps hot air out the side, but the surfaces stay quite cool.

How could I narrow down (or help my tech guy narrow it down) to find the problem? I was thinking of use throttle stop to keep it from throttling under load... thinking that if it didn't help and it still throttled the same as always, I would know it was a temperature issue, NOT a chipset issue or software related, because I read that intel will override throttle stop if the system is getting too hot, and down throttle for safety measures anyway.

any help or advice would be greatly appreciated!!

[JEFFREY_F_Intel]

It definitely sounds like thermal throttling. The cpu contains a power control unit which monitors temperatures and if they get too hot will disable CPU and GPU turbo - forcing them to run at lower temperatures. Possible cases are:

1) inadequate cooling solution - not enough airflow or fan not kicking in when it should

2) heat sink not making good contact with the cpu.

If you run cpu benchmarks (superpi, cinebench, etc) do you see the same temperature/step function drop?

[idata]

Thanks for the reply!

I have Real Temp installed and ran the Prime95 sensor test that it has, and within 5 or so seconds, it was showing 99 degrees on the CPU cores according to Real Temp, so I stopped the test to protect my system, and wasn't able to see if it was downclocking the CPU as well.

but, since it's GPU is integrated, if I see the GPU clock cycles are throttling down to 650mhz, doesn't that mean that the CPU is also throttled down? I thought they share the same component? I'm not really sure how the processor/integrated GPU works on a physical level.

Is it common for smaller-form laptops using higher end processors to have cooling systems that simply can't adequately cool the processor when fully loaded?

I'm taking it in on Tuesday to a guy who has a lot of experience fixing fans and cooling systems on similar type units. Hopefully he'll find out something like the heatsink improperly seated or the paste not applied proper. He sounded confident.

How would I narrow it down if it was a chipset thing??

[idata]

Luke,

You should use a C or F designator when refering to temperatures.

RealTemp is software based, so do you have any equipment to get another a comparison value of the temperature?

Electronics stores sell a product called Freeze It, or other brands are available of a chemical spray that will lower the temp of the object that it is sprayed on. Use a brand that does not leave a residue. A pipette is included to focus the spray. Use a spray that is designed for cooling not just a dust off spray

You should be able to push the GPU away from the throttling temp, noted by your readings with the spray, and varify this with several cooling cycles / heating cycles. Short blasts of the spray should be sufficient to lower the chip's temp, especially if you don't let it get too far past the throttled temp.

Let the community know of your results.

[idata]

i using i5 3317U run clock GPU 350~650Mhz

can't up than 650Mhz why???

laptop acer s3-391

[idata]

Hi ot,

I've never heard of Freeze it.

Are you saying to use it to confirm whether the temperature is really what is causing the throttling? Or use it as a workaround solution?

Also, would I have to open up my computer to use "freeze it"? I mean, how do I access the processor to spray the stuff on, and is there any risk of doing damage to the chip when using this??

Assuming it works to lower the temp effectively while the GPU is being stressed, I would be monitoring the GPU clock cycles while blowing this stuff on to see if in fact the GPU would perform better? I guess that is one way of narrowing it down. I'll have to look into the stuff.

you wrote:

"You should use a C or F designator when refering to temperatures.

RealTemp is software based, so do you have any equipment to get another a comparison value of the temperature?"

sorry. The temps I reported above are in Celsius.

I also have monitored with throttlestop software and GPU-z and it has shown very similar temps. So, three separate programs have seemed to confirm that my CPU and integrated GPU get up in the high 80's (Celsius) regularly when stressed. However, my GPU still throttles at exactly the same time and way even when it's only in the mid-high 70's. Perhaps the BIOS has a very aggressive thermal throttle setting.

I have also managed to change the pattern and improve things a little by manually throttling my CPU back to 2.0 ghz (using throttle stop). When I do that and then run a test, it shows that the HD 4000 throttles back to 650 for a few seconds, and then goes to 1250 for a little while then throttles back for a few seconds, etc essentially a reverse pattern of before. It happens like clockwork though, and I can't tell if it reacts that way because the lower clocked CPU is keeping it from getting as hot, or if it's more to do with enabling more power available to distribute to the GPU. any thoughts on that??

here is some visual aids to help understand how my system compares to other HD4000 integrated GPU's in a typical usage scenario when the GPU is under load. In this case, Battlefield 3 is being run, and my cpu is being compared to a properly functioning ULV i5 3427U in a smaller ultrabook form factor. Even smaller than my setup. As you can see, there is no excuse for my system to be behaving like this. In all respects this system I am comparing it to is inferior, yet the throttling is not happening and it is running smoother than my full voltage CPU. hope this helps in understanding what I am talking about.

[idata]

Luke,

Yes you will have to open you machine to test it properly. Would you spray the Freeze It directly on the heat sink. 80C is 176F thats hot, and will burn your finger if you hold it there for too many seconds. But is it really that hot? You can't rely on just the software at this point. The sofware monitors core temp, but what is the heatsink temp for this use a IR non contact thermometer or thermocoupled meter to verify. Seems like its heating up pretty fast if that the case, maybe they forgot to put thermal paste on the heat sink, maybe the fans arent coming on when they should. Maybe its a bad CPU, or fan interface, is something blocking the air flow internally? Hard to guess until you narrow it down with the freeze it and monitor the temps and clock cycles. Don't shock the chip by rapidly freezing it so it looks like a snow ball, little bursts of spray and you wont harm it. Monitor the temps, cool it slowly. Here is the other kicker, HD4000 produces more heat than the HD3000, so if your laptop is using the same heatsink arrangement for the HD4000 as they did for the HD3000 this could be an issue. You could start with a cold machine too, keep spraying freeze it on it and see if you can prevent the throttling.