Surface Pro 3 throttling

I just check every 4th gen i5 processor listed by Intel as a U. All of them have 15W TDP except for 4258U, 4278U, 4288U,  and 4308U.  There is no 4300 with a higher TDP for desktops. That said the 4300 I am sure companies are putting U processors with limited TDP into desktops.

I don't see any evidence that a desktop processor was used in any Surface model, except that desktops could use U processors.

Alright, I just finished my slew of tests and research and all I bring is bad news. The surface isn't throttling because of temperature directly, as in it being cut when it hits a certain degree C. Whats happening here is TDP throttling. The i7 Surface pro has a 15W TDP rated processor, but all that means is that it'll use around 15W TDP in average situations. However, Intel gives manufactures the ability to add their own TDP "levels" to the hardware to enforce temperature control. Thats exactly what we're seeing here. The Surface Pro 3 i7 has two TDP levels, level 1 at 11W and level 2 at 25W. Under "average situations" the surface will be fine with its 25W TDP as it will never have to jump above 10 or so for more than a few seconds, again in "average situations". Even stressing the processor to 100%, bringing it to 2.3GHz uses around 13W TDP, not enough to worry the Surface into any throttling yet.

However, you add the GPU into that and it jumps up to 17.5W TDP. The surface will maintain this for only a short while before starting to cut down on that number, even if the device is within acceptable temperatures. It doesn't care, it just wants to lower the TDP. I've noticed it isn't so simple as cutting to that 11W instantly, sometimes in my testing it would retreat to around 13W, leaving enough power for the processor to run around 1.3GHz and the GPU at 550-600MHz or so. Other times, especially after testing continuously, it would lock to that level 1 TDP and just drop to 8-9W TDP which is not enough to do much of anything when the CPU and GPU are sharing it. This is what we are seeing when framerates just drop and things become unusable until we let the surface take a break.

The best I was able to do with my processor tweaks is maintain that 13W TDP for about 15 minutes, after which I quit that test and went on to other tests.  

I wish I still had my Surface pro 2 to run the same monitors on, I'll have to see if I can borrow a friend's to do this at some point. Anyway, this TDP cap is something that Microsoft has decided on to make the Surface as stable as possible and is baked in fairly deep, it would be possible to release an update to change it but I really doubt that that happens.

Finally, I believe my incessant testing has royal confused my Surface. It now sees its stock running speed at 2.3GHz instead of 1.7, even showing so in the system panel of the control panel, where it says

i7-4650U CPU @ 1.7GHz    2.3GHz

The first bit is just a string, the number to the right is the processor's reported frequency. I'm doing a factory restore on my surface now, but if anyone else with an i7 surface can confirm that their processor info under control panel->system doesn't look like this it would help my tests.

Edit: The factory restore didn't change anything, so it looks like that was just a big waste of my time. I'm starting to doubt myself, but I still think that processor line in system is not right. If anyone with an i7 surface can confirm one way or another I would appreciate it. All I know is that my processor controlling application started out saying that 1.7GHz was the non-turbo frequency and seemed to be able to control the processor accurately, but now displays 2.3GHz as the stock frequency and generally doesn't work correctly anymore.

I've finished more testing too with a plasma simulation that pegs the CPU at 100%. I'll post some plots but I think I have a guess as to what is going on. I noticed that under very heavy load the computer can maintain a single thread at 100% at full 2.6-2.9GHz indefinitely at 70C using 9W. The back of the computer doesn't get too hot. This is important for later.

I did more testing with both cores at 100%. After the first 5 minute test everything was the same, except for using 9W it used 12W and the temperature rose to 76C.

I repeated the tests immediately, and noticed the computer throttling toward the end of my simulation. The CPU temperature wasn't too hot, but much more of the back of the computer was much warmer than the first round. The heat capacity of the body would have delayed this. This made me wonder if there wasn't something else the throttling is protecting. Then I remembered my Asus EP121. That computer had multiple temperature sensors inside the body of the device as well as what was reported by the processor. This makes sense if you are holding the device or if you are protecting other components.

I think what is going on is the body temperature is rising too high for other temperature sensors and the throttling is intended to cool the body of the computer. That would explain while decreasing the body temperature with an ice pack allows it to maintain a higher load even, while without such extra cooling even though the CPU isn't hot the computer slows down.

I think that extra 3W is the limit for how much cooling without heating the body.

I've finished more testing too with a plasma simulation that pegs the CPU at 100%. I'll post some plots but I think I have a guess as to what is going on. I noticed that under very heavy load the computer can maintain a single thread at 100% at full 2.6-2.9GHz indefinitely at 70C. The back of the computer doesn't get too hot. This is important for later.

I did more testing with both cores at 100%. After the first 5 minute test everything was the same, except for using 9W it used 12W and the temperature rose to 76C.

I repeated the tests immediately, and noticed the computer throttling toward the end of my simulation. The CPU temperature wasn't too hot, but much more of the back of the computer was much warmer than the first round. The heat capacity of the body would have delayed this. This made me wonder if there wasn't something else the throttling is protecting. Then I remembered my Asus EP121. That computer had multiple temperature sensors inside the body of the device as well as what was reported by the processor. This makes sense if you are holding the device or if you are protecting other components.

I think what is going on is the body temperature is rising too high for other temperature sensors and the throttling is intended to cool the body of the computer.

That's what this other user found

http://www.surfaceforums.net/threads/slightly-different-spin-on-the-reason-for-throttling.10266/#post-74611

I think what is going on is the body temperature is rising too high for other temperature sensors and the throttling is intended to cool the body of the computer. That would explain while decreasing the body temperature with an ice pack allows it to maintain a higher load even, while without such extra cooling even though the CPU isn't hot the computer slows down.

I think that extra 3W is the limit for how much cooling without heating the body.

I second this hypothesis, the data falls in line with my tests as well. I believe it is possible to maintain resonable TDP levels using the graphics, but only in a relatively cool ambient temperature along with starting with a cold surface. Adding in the idea that the Surface is watching the casing temperature and people "fixing it" by cooling down the body of the Surface seems in line with this.