Help: rMBP 650M GPU Throttling on BootCamp

x3n0n1c said:

Intel Power Gadget can be used to monitor CPU clocks in OS X. I have not looked for a way to monitor GPU clocks, though i'm sure some method exists somewhere.

Those would have to be checked.

Click to expand...

checked it. throttles CPU when GPU is under load (used furmark). or after a minute of only-CPU stress test.
edit: Also, it throttles the GPU too when tested with furmark. (checked the voltage indicator. It drops after a while)

exnihilodub said:

checked it. throttles CPU when GPU is under load (used furmark). or after a minute of only-CPU stress test.
edit: Also, it throttles the GPU too when tested with furmark. (checked the voltage indicator. It drops after a while)

Click to expand...

So there ya go.. lol

recently..

okay guys, I've tried something and got some interesting results on my side:

I've ramped up the fans with "macsfancontrol" prior to starting a game and cooled it to a little bit lower than my machine's usual idle temps. Considering the Mac's fans start up slowly when under load, I thought this might help me distinguish between the "Thermal/Power Consumption" possibilities.

And surprise the temps could actually go HIGHER to 100's! and the machine stayed at the 80-85W for a long before throttling down again. So i ruled out the power limit possibility.

I was thinking about one of the sensors may have a lower thermal ceiling, or some sensors could even be hidden from the OS. After numerous experiments with simulating cold&heat, i've come to the conclusion that the throttling happens whenever the sensor "Temperature #8" hits 50-51 C's (has nothing to do with CPU/GPU temps as far as i can see, because the throttling could happen when they were at 80C or 100C, And CPU/GPU temps and this Temp #8 always moves inversely proportional to each other.). Unfortunately AIDA64 gave no label to temp #8 so I've found out that Heatsink #1 at HWMonitor, gives the same reading with temp #8 in AIDA64.

So my final conclusion for now is either the Heatsink has a throttling capability, and its ceiling is as low as 50C,- or there's a non-reporting sensor that's in close proximity to the heatsink and gets higher temperatures, but relative to the heatsink.


Here are the screenshots for the interested:

exnihilodub said:

okay guys, I've tried something and got some interesting results on my side:

I've ramped up the fans with "macsfancontrol" prior to starting a game and cooled it to a little bit lower than my machine's usual idle temps. Considering the Mac's fans start up slowly when under load, I thought this might help me distinguish between the "Thermal/Power Consumption" possibilities.

And surprise the temps could actually go HIGHER to 100's! and the machine stayed at the 80-85W for a long before throttling down again. So i ruled out the power limit possibility.

I was thinking about one of the sensors may have a lower thermal ceiling, or some sensors could even be hidden from the OS. After numerous experiments with simulating cold&heat, i've come to the conclusion that the throttling happens whenever the sensor "Temperature #8" hits 50-51 C's (has nothing to do with CPU/GPU temps as far as i can see, because the throttling could happen when they were at 80C or 100C, And CPU/GPU temps and this Temp #8 always moves inversely proportional to each other.). Unfortunately AIDA64 gave no label to temp #8 so I've found out that Heatsink #1 at HWMonitor, gives the same reading with temp #8 in AIDA64.

So my final conclusion for now is either the Heatsink has a throttling capability, and its ceiling is as low as 50C,- or there's a non-reporting sensor that's in close proximity to the heatsink and gets higher temperatures, but relative to the heatsink.


Here are the screenshots for the interested:

Image


Image


Image

Click to expand...

I don't see a full load scenario? you're getting 100c without even under full load???

You need to run LinX + something that will peg the GPu 100% then report temps/CPU and GPU clocks via Intel Extreme Tuning utility and GPUz.

Or just run the stability test..

EDIT: This is what I want to see. These graphs, while running the test, will clearly show thermal/power throttling.







Power from the wall (not charging) = 132watts

ok I did as you said and noticed two things

AIDA64 gave a throttling warning when the CPU hit around 104 C's. but that was not the thing that was bugging me. It's pretty much acceptable to have some kind of throttling at 100's.

What was bugging me was that when the purple line called "Temperature #7", hit 50-51C's, the clock rate of CPU dropped significantly --regardless of the CPU heat or load-- and never backed up (since that purple line takes a long time to go down). That's why I gave it two kicks of stress tests. One long, one short. Here's the power/temp graphs:





EDIT (forgot to take a snapshot of clock graphs, sorry. Just made a test again, yielded the same results and now Im posting the clock graphs

exnihilodub said:

ok I did as you said and noticed two things

AIDA64 gave a throttling warning when the CPU hit around 104 C's. but that was not the thing that was bugging me. It's pretty much acceptable to have some kind of throttling at 100's.

What was bugging me was that when the purple line called "Temperature #7", hit 50-51C's, the clock rate of CPU dropped significantly --regardless of the CPU heat or load-- and never backed up (since that purple line takes a long time to go down). That's why I gave it two kicks of stress tests. One long, one short. Here's the power/temp graphs:

Image

Image

EDIT (forgot to take a snapshot of clock graphs, sorry. Just made a test again, yielded the same results and now Im posting the clock graphs

Image

Click to expand...

So looking at the last graph, I see it throttled for a bit and never recovered. I wonder once it hits that magical temp on some sensor (which you can't pinpoint) and will never allow full clocks again unless the laptop is cooled down. I do see some throttling not related to thermals towards the end of your test. I wonder if there is a temp sensor in the power brick? 50c sounds low.. but maybe they are monitoring the power brick temp so nobody burns themselves? I would stick that in the fridge for 10 minutes and then plug it in and see if that temp has dropped.

ha1o2surfer said:

So looking at the last graph, I see it throttled for a bit and never recovered. I wonder once it hits that magical temp on some sensor (which you can't pinpoint) and will never allow full clocks again unless the laptop is cooled down. I do see some throttling not related to thermals towards the end of your test. I wonder if there is a temp sensor in the power brick? 50c sounds low.. but maybe they are monitoring the power brick temp so nobody burns themselves? I would stick that in the fridge for 10 minutes and then plug it in and see if that temp has dropped.

Click to expand...

as I've said earlier, I cooled the machine down before opening up the game and waiting for the throttling to occur ( I always use the same scene for testing). It let the machine go to higher temps (as in my latest stress test. It starts to throttle at 80C otherwise) and it stayed at full clocks for a longer period before dropping down again. so I think it's definitely a temperature problem, Im just not sure about which temp sensor that's about, or if the sensor is hidden from the OS completely.
-----
Long story short, yeah it definitely lasts a bit longer before downclocking, when I cool the machine down prior to tests. And can use the 85W power limit efficiently in that time. I think I'll just wait a while before I go to a house with air conditioning and try my chances there. Will report back with my findings.

I messed around with everything on my 2013 MBP w/ 750m. Nothing will work to stop the throttling. It's wasted effort unfortunately.

As others have said, the update that prevented MacBooks from draining the battery for extra power pretty much ruined these laptops. Wouldn't surprise me if it was part of Apple's plan for everything to have integrated graphics.

The power adapter is the culprit. They'll never bump it up any bigger as it'd go against their "slim is sexy" agenda.

iStrikeRx said:

I messed around with everything on my 2013 MBP w/ 750m. Nothing will work to stop the throttling. It's wasted effort unfortunately.

As others have said, the update that prevented MacBooks from draining the battery for extra power pretty much ruined these laptops. Wouldn't surprise me if it was part of Apple's plan for everything to have integrated graphics.

The power adapter is the culprit. They'll never bump it up any bigger as it'd go against their "slim is sexy" agenda.

Click to expand...

can you provide some graphs/test results? mine happens before hitting the power limit or w/o the adapter (on sole battery power).

exnihilodub said:

can you provide some graphs/test results? mine happens before hitting the power limit.

Click to expand...

Since you don't appear to be hitting a power or cpu/gpu thermal limit I again have to go back to your power brick.

It has been observed that when your powerbrick itself goes above a certain temperature, the system will respond by throttling components in order to reduce power consumption, which in turn will reduce the stress on the power supply and allow it to cool off.

Make sure your brick is in the open and has access to some form of airflow if possible. You can test by maybe wrapping it in a flexible icepack while under load and see if you can get any change.

I know you mentioned that you could get the clocks to go down by strangling the laptop of fresh air, but that may be a separate response and not the one causing your throttling under load.

Who knows!

x3n0n1c said:

Since you don't appear to be hitting a power or cpu/gpu thermal limit I again have to go back to your power brick.

It has been observed that when your powerbrick itself goes above a certain temperature, the system will respond by throttling components in order to reduce power consumption, which in turn will reduce the stress on the power supply and allow it to cool off.

Make sure your brick is in the open and has access to some form of airflow if possible. You can test by maybe wrapping it in a flexible icepack while under load and see if you can get any change.

I know you mentioned that you could get the clocks to go down by strangling the laptop of fresh air, but that may be a separate response and not the one causing your throttling under load.

Who knows!

Click to expand...

If what you mean is the AC adapter by saying "power brick" I've already tried it by plugging it out and it was still throttling. If what you mean is the PSU inside the macbook which is the energy carrier from the battery to logic board, I was worried that it may be the cause for downclocking. I'll make sure which one is the culprit and report back. First I'll need to go somewhere cold or wait until the weather turns back to normal again. IT'S SCORCHING OVER HERE.

To see how my laptop performed, I played diablo 3 for half an hour with the power limit set to 95% in power options, which left clocks somewhere around 2Ghz.

I was also using an external monitor and not the built in display. Here is my chart from Nvidia Inspector. I have my AC adapter plugged into a Watt Meter and usage during this test was around 65-75 watts at the wall.



The GPU clock average position was around 800MHz, with the spikes being close to 900.

x3n0n1c said:

To see how my laptop performed, I played diablo 3 for half an hour with the power limit set to 95% in power options, which left clocks somewhere around 2Ghz.

I was also using an external monitor and not the built in display. Here is my chart from Nvidia Inspector. I have my AC adapter plugged into a Watt Meter and usage during this test was around 65-75 watts at the wall.

Image

The GPU clock average position was around 800MHz, with the spikes being close to 900.

Click to expand...

I'll also need to see the CPU clocks. GPU throttling begins as soon as the CPU is downclocked and hits the lower floor of 1190~(i don't remember the exact frequency) and GPU is the only remaining peripheral to clock down in order to reduce temps.

i've read all the posts, thank you to everyone, especialy to exnihilodub.

I've been 4 days in a row trying to fix this problem (constant spikes from 724 to 270 mhz during 0,1s or so). i've tried everything on this forum and others, re-format, oc,uv,fixed volt, turboboost, c1e, prochot, NVinspector, throttlestop..cheking temps on both gpu cpu... nothing did it.

You said something can't go to 50C, so i've decided to open my mac and point two fans at the mainboard without any software on ecxept gpuz for monitoring.
IT WORKED!!! the core and mem frequencies kept at 900/2500 all the time when playing payday 2 at 70-75fps,about 20-30min. i took the fans out and in about 10-30 secs it started to go 724 and a few seconds latter the f*$«%*# throttling beggan even with gpu at less than 70C and cpu less than 75. (then i tested with realtemp and cpu was fixed at 2ghz, from 95% in win energy preferences i think)

I can test everything you want to take out some theories, like keeping gpu and cpu at higher temps(80-90) and keeping 900/2500 with extra fans on. ask for anything you want. tomorow ill do more tests.

I'll try to get some friends infrared distance thermometer to find witch component needs some mini fan or passive dissipator.
http://www.clickbd.com/global/classified/item_img/1090077_0_original.jpg

nightgost said:

i've read all the posts, thank you to everyone, especialy to exnihilodub.

I've been 4 days in a row trying to fix this problem (constant spikes from 724 to 270 mhz during 0,1s or so). i've tried everything on this forum and others, re-format, oc,uv,fixed volt, turboboost, c1e, prochot, NVinspector, throttlestop..cheking temps on both gpu cpu... nothing did it.

You said something can't go to 50C, so i've decided to open my mac and point two fans at the mainboard without any software on ecxept gpuz for monitoring.
IT WORKED!!! the core and mem frequencies kept at 900/2500 all the time when playing payday 2 at 70-75fps,about 20-30min. i took the fans out and in about 10-30 secs it started to go 724 and a few seconds latter the f*$«%*# throttling beggan even with gpu at less than 70C and cpu less than 75. (then i tested with realtemp and cpu was fixed at 2ghz, from 95% in win energy preferences i think)

I can test everything you want to take out some theories, like keeping gpu and cpu at higher temps(80-90) and keeping 900/2500 with extra fans on. ask for anything you want. tomorow ill do more tests.

I'll try to get some friends infrared distance thermometer to find witch component needs some mini fan or passive dissipator.
http://www.clickbd.com/global/classified/item_img/1090077_0_original.jpg

Click to expand...

Thanks man! that was really helpful. I had some questions:

What model of Mac you have? Because on rmbp's, only the bottom cover is removable and it's hard to play like that. I assume you have another mac but your max clocks are the same with me.

Could you actually pinpoint where on the mainboard delays the throttling when you hold the fans up? (might take a lot of time, cooling the system down and trying over and over again)

Were the CPU/GPU temps actually able to go higher before you removed the fans and the throttling began?

-----------

Btw it became 35C here and %78 humidity, before that i could play with downclocking the GPU/CPU intentionally at the expense of FPS, but now it seems impossible even with extreme downclocking made with NVI&Throttlestop. I'm like 90% sure now it's a problem of temperature.

I thank you again for taking this extreme measure and testing it like this You've been a great help..

Mid-2012 15" - MD103LL/A - MacBookPro9,1 - A1286 2.3Ghz processor

i've mounted it on two wooden bars with the fans at mid air pointed up to the mac.

I didn't understand what to pinpoint. but there is no slow process here.
after playing with extra fans i remove the fans and about 20-30 secs(sometimes less) it start to downclock, 10 sec more and the spikes to 270 begin. at this point there is no need to restart, just turn on the extra fans and 5 secs latter its back to 900/2500.

sorry for my english, not my native language. now i'm going to try turboboost enabled and see if thats a problem. then high temp gpu e cpu, above 80C.
PS:all tests were performed with charger pluged in and 100% battery.
the bottom cover has iman, so no screws for me, my mac is free=D

first test: turboboost playing payday 2 (maybe i need to change the testing software to have max loads)
it worked(the clocks on the gpu didnt throttle down)

gpu - load 98% (max) - max temp 68C
gpu mem - load 47%
cpu at 3050 with throttlestop - load 25% (max is 3,1Ghz when using 4 cores) - max temp 81C

im using arctic silver 5 thermal paste.

second test: keeping high temps arround 90C

result: lunch timeXD

i was able to get throttling before temp 7 or 8 hits 50C. but temp 10 hiting 60C was at the same time of throttling. this was achieved by turning off and on the extra fans.
see for yourself(temp 7 and 8 not in the graph):

nightgost said:

i was able to get throttling before temp 7 or 8 hits 50C. but temp 10 hiting 60C was at the same time of throttling. this was achieved by turning off and on the extra fans.
see for yourself(temp 7 and 8 not in the graph):

Click to expand...

and i was able to delay thottling until it hit 54C. so it's not that sensor's problem. Our labels for the Temp#'s are different, so i cannot compare these results (my temp #10 never exceeds 44C even under CPU&GPU stress test)

there are still a few temperature lines moving inversely proportional to the CPU/GPU temps. I'll chill the macbook next week and try it again then.

EDIT: and nightgost, could you please provide me with your PCH Diode temps under stress test and where it throttles - if that diode is available on your graphs? Because that diode still keeps heating up even when the CPU&GPU throttled itself down and cooled themselves.


***If Apple did the intensive testing we did, they'd have released a better machine to the market -just joking- ***

well guys, placing the macbook in front of an air conditioner worked. I've opened a game, the clock rates dropped after a while and as soon as i held my macbook against the AC, the clock rates went up again. I couldn't make out which peripheral is the cause. the temps only raised when I held the macbook against the AC- that's because the CPU/GPU had more thermal headroom when I cooled the whole macbook-. I don't know how to implement this technique in everyday use.. It may be a broken sensor/hidden sensor/dust-nicotine or a thermal paste went bad (in 1 year??) in my case. I'll try to clean the innards of the macbook as soon as i get my hands on some torx and pentalobes. Will report back shortly

Never blow sub ambient air on your laptop. This can easily cause condensation and short something out.

x3n0n1c said:

Never blow sub ambient air on your laptop. This can easily cause condensation and short something out.

Click to expand...

yeah that's not something I'd use forever for managing this problem. I just took the risk and did it for a very short time to make sure that the problem was temp related.

Have any Yosemite beta testers encountered this problem?

If the power adapter is hitting 75 watts then I'm sure that is it's max. There are two separate voltages coming out of the wall adapter and I'm sure each have their own limit.

just thought I should update from before,
I bought a new right fan from iFixIt (my last one completely died) and now I'm getting no throttling at all, GPU isn't going past 70 degrees (with 1035MHz OC) and CPU hasn't gone past 90 degrees (throttlestop with modifier locked to 24).

So, that was my solution.