slow benchmarks: Optimus V, 4710MQ

S

sjpt

Member
I've recently bought a pcspecialist Optimus V laptop with i7-4710MQ processor.
I'm very pleased on the whole, but finding the cpu running much slower than I expect.

This is reflected in slow benchmarks. My Passmark CPU score is 4084, similar Passmark scores suggest it should be around 8000.
For example #273827 gave 7706 on an Optimus V with the same processor.

This was run on Win 8.1 High Performance Plan.
Are there any settings I am likely to have missed?
Can anyone suggest a reason why?
 
S

sjpt

Member
Seems to have been a curiosity of Speedstep? When I was getting the poor results the CPU clock speed never went above 1.5GHz. (as shown by Open Hardware Monitor and other testing programs). This was NOT due to high temperatures.

After a reboot (including a quick check in the BIOS where I didn't change anything) the clock was allowed to go to a full turbo speed of 3.3GHz and I got exactly the sort of results I would expect. Also, the application that had been giving problems started behaving too.

So, now all is well; still not sure what was happening in the first place.
 
S

sjpt

Member
I'm adding a few more details here that may be relevant to some people trying to get the best out of their laptop, especially if they are also performance testing software.

Summary: Speedstep can significantly effect attempts to analyse application cpu usage.
It appears to be smart enough that it does not hit real world performance.

There is an oddity with SunSpider on Chrome when running Avast!

~~~~~~~~~~~~~~~~~

I have got a new laptop with i7-4710MQ (and GTX860M, not relevant to this note).
This has very aggressive Speedstep, with a nominal core CPU frequency of 2.5GHz,
but a resting frequency of 0.8GHz and a turbo of 3.5MHz.

Most reported cpu load figures show a proportion of the CPU relative to the current core frequency;
for example ProcessHacker and the Chrome javascript profiler.
Thus it can appear that you are almost saturating the cpu when in fact you are not.
Open Hardware Monitor is a good way to keep track of various things including core frequencies,
and Resource Monitor shows the compensated cpu time in the graph.

Speedstep is affected by Power Management/Processor Power Management settings.

Changing to lower power management has immediate effect on lowering core frequencies.
However, raising power management level does not always have immediate effect.
Core frequencies are not raised if the overall cpu requirement is ok at the lower frequencies.
However, once the core frequencies have been raised they stay high.
So, to get consistent speed tests and analysis I have had to

* raise the power plan to High Performance
* give a burst of cpu hitting, for example running SunSpider
https://www.webkit.org/perf/SunSpider-1.0.2/SunSpider-1.0.2/driver.html
* run the test.
It doesn't matter if the test program was already running, as long as timings are not taken till after steps 1 and 2.

Evidence from the app I was testing (Webgl, low cpu, high gpu) is that Speedstep does the job remarkably well and that its framerate did not lower with Balanced or even Power Saver. It is only necessary to force full speed in order to get consistent cpu timings for analysis.

~~~~~~~
There is also an oddity in SunSpider tests on some versions of Chrome; made a lot worse if running Avast!
One particular test (date-format-tofte https://www.webkit.org/perf/sunspider-0.9/date-format-tofte.html) gives absurdly high results (thousands rather than tens) and completely skews the overall result.

I am not sure if this also affects real javascript performance with Avast! running. It certainly caused me extra confusion while checking out my laptop performance.
 
Last edited:
You must log in or register to reply here.
Top