What's new

Solved Sustained Core i7 performance (no throttle) using additional USB fan

RémiM

Active Member

Capture.PNG Capture2.PNG
 

Cothek

Active Member
I ran a 15-minutes CPU stress test from Intel Extreme Tuning Utility. Max 100% processor frequency is just above 2.8GHz (that's Turbo Boost, the Core i7 4650U being rated at 1.7 GHz).

The horizontal red line at the bottom of the graph shows how sustained and constant the CPU speed is. The two peaks in the green line at the bottom are minor throttling events that effectively resulted in the internal fan kicking-in, thus why the CPU and GPU temperatures gradually decrease afterwards (blue and green-ish line at the top). During 15 minutes of high load, this kind of (non-disruptive) temporary throttling occurred three times only. Meanwhile, the CPU was running a near-full-frequency.

Surface_Pro3_Corei7_256_GB_Intel_Extreme_Tuning_Utili.png




When I run Intel XTU's GPU stress test, the CPU stays slow, and the combination of internal + external cooling works as expected (no noticeable degradation in performance). I also tried 3D Mark and PassMark Performance Test ... but to be honest I am not familiar with how relevant these synthetic benchmarks are (all I care about is real-world usage, which in my case does not involve gaming ;) ).

Anyways, 2892 score at 3DMark Sky Diver (stock CPU settings):
http://www.3dmark.com/3dm/5630235
That's with the latest driver for Intel HD Graphics 5000 ==> 10.18.14.4080 (manually installed over Windows' own driver).
Sorry to be nit-picky but we really need to see the CPU Total TDP and power limit throttling in the graph as well. The limiting factor of the SP3 is actually not the thermal throttling shown in this graph (which is based on the CPU) but a thermal limit inside the case somewhere. The reason you are seeing very little thermal throttling is because the CPU will power throttle once the case hits a certain temp which in turn drops the TDP. Also the mountains and valleys in the temp show that the CPU is slowing down (or dropping power) allowing the temp to drop on the CPU and case then after the external thermal sensor gets low enough the CPU ramps up again. Performing the GPU stress test would actually provide a more in-depth and worst case look at what those fans provide in terms of cooling.

Also, thanks for at least sharing this on the forum here.
 

ptrkhh

Active Member
the mountains and valleys in the temp show that the CPU is slowing down (or dropping power) allowing the temp to drop on the CPU and case then after the external thermal sensor gets low enough the CPU ramps up again.
I wonder though, why does the CPU clock stay when the temp gets lower?
 
OP
D

Daniel Weck

New Member
Thank you for the feedback. I appreciate your technical insights, as I want to be able to correctly interpret the statistics generated by Intel Extreme Tuning Utility.

Basically, are you saying that even though I experience sustained CPU clock speeds (> 2GHz on average), the performance is not actually what it seems? My benchmark results seem to indicate that computational power remains high thanks to the additional cooling, which is exactly what I am looking for.

I am not interested in gaming performance, but I can see that when the GPU is solicited, the heat+throttling pattern is different (using the stress test in Intel XTU, the CPU frequency actually remain slow, as only the GPU seems to be under load). I also posted 3DMark results, but I am not sure how relevant this is?

Please let me know how I should test, which software, etc. and what performance criteria I should be watching for.


Thanks, Dan
 

Justing6

New Member
Thank you for the feedback. I appreciate your technical insights, as I want to be able to correctly interpret the statistics generated by Intel Extreme Tuning Utility.

Basically, are you saying that even though I experience sustained CPU clock speeds (> 2GHz on average), the performance is not actually what it seems? My benchmark results seem to indicate that computational power remains high thanks to the additional cooling, which is exactly what I am looking for.

I am not interested in gaming performance, but I can see that when the GPU is solicited, the heat+throttling pattern is different (using the stress test in Intel XTU, the CPU frequency actually remain slow, as only the GPU seems to be under load). I also posted 3DMark results, but I am not sure how relevant this is?

Please let me know how I should test, which software, etc. and what performance criteria I should be watching for.


Thanks, Dan

You can still use the Intel XTU, just make sure you run the GPU test. Yes, the CPU will run considerably slower than the turbo boost speeds, but that is because the graphics is getting the majority of the TDP. When only running the CPU test, the CPU gets around 12W TDP to run at 2.8GHz. 10W TDP which is around 2.3GHz, not too much slower, is actually maintainable with the internal cooling. When running a GPU test, the GPU gets 9W and the CPU only 5 or 6, this means the CPU has to run at a lower clock speed but it is using more TDP overall, close to the max of 16, which puts much more heat strain on the processor and causes the throttling. Just run a GPU stress test in XTU and show the same stats on the graph as this one from Cothek:

08-19-gpu-test-balance-fan-png.3498
 

Cothek

Active Member
Thank you for the feedback. I appreciate your technical insights, as I want to be able to correctly interpret the statistics generated by Intel Extreme Tuning Utility.

Basically, are you saying that even though I experience sustained CPU clock speeds (> 2GHz on average), the performance is not actually what it seems? My benchmark results seem to indicate that computational power remains high thanks to the additional cooling, which is exactly what I am looking for.

I am not interested in gaming performance, but I can see that when the GPU is solicited, the heat+throttling pattern is different (using the stress test in Intel XTU, the CPU frequency actually remain slow, as only the GPU seems to be under load). I also posted 3DMark results, but I am not sure how relevant this is?

Please let me know how I should test, which software, etc. and what performance criteria I should be watching for.


Thanks, Dan
The performance is what it seems but it doesn't look like it's "throttle-free". I've annotated your graph below with my thoughts:
upload_2015-1-29_21-7-4.png


True throttle-free would never see the temp hit 90C. Thermal throttling should only really kick in once the CPU internal temp hits 90C. Also, what's the ambient temp you are running your test in?

I've been working on a brand new duct using a radial notebook fan and so far the results are very promising. Running the tests at my fairly well airconditioned work gives me results of no thermal throttling! As you can see below my CPU never hits the 90 mark.
upload_2015-1-29_21-15-56.png

I ran the test for 45min and kept a rock steady 17W Total TDP and mostly constant 11W GPU TDP.

For comparison the CPU test with the same fan maintained a much lower temp of around 80-81:
upload_2015-1-29_21-24-27.png
 

Attachments

  • upload_2015-1-29_21-6-26.png
    upload_2015-1-29_21-6-26.png
    106.6 KB · Views: 844
Top