I'm using Intel GPA on a Surface Pro 2 to measure power usage on an application and I've spotted something that I can't explain. If I (using the same power profile) unplug the power cord when profiling the power usage *goes up*. How is this possible? I'd expect power to remain the same, or maybe go down (in case there's some HW limitation kicking in when not plugged in), but for the Processor Package power usage to go up when the power cord is unplugged is deeply mysterious.
Can anyone explain what's going on? What should I use for comparisons, plugged or unplugged?
Can you provide some more information about your specific configuration? First of all, right-click the Intel GPA Monitor in the notification tray and copy the "About..." info here. And please be sure that you have the latest version of Intel GPA, and the latest BIOS and drivers for your system configuration, and check your BIOS to be sure there are no power settings that might affect total power usage.
Also, what app or game are you using to monitor power usage? And you say the power usage increases, but you don't mention the specific power usage for plugged vs. unplugged (is it a 3% difference or 30%?). And how reproducible and consistent are these numbers?
Also, though you indicated that you're using the same power plan, I'm assuming that everything is the same for both the standard settings and the advanced settings (such as PCI express, system cooling policy, and min/max processor utilization)? Unfortunately there are quite a few power settings, and one or more of them could have an impact on power usage overall.
Sorry that I don't have a simple answer for you on this, but after you provide some additional info, either myself or others watching this forum can make some comments in response.
Hope this helps!
I just installed GPA the other day, here's the about stuff:
Windows 8.1 build 9600, 64-bit DEP enabled
Num Processors: 4
System BIOS: American Megatrends Inc. 2.03.0250 (09/06/2013)
Device: Intel(R) HD Graphics Family (Intel Corporation - WDDM 1.3)
Provider: Intel Corporation (Intel Corporation - WDDM 1.3)
Supports GPA Instrumentation
GPA install directory: C:\Program Files\Intel\GPA\2013 R4\
GPA version: 13.4.215202
Current user is in Administrators group: YES
Current GPA 2013 R4 (13.4.215202)
I've run through all the latest updates on the system (which is the only way I know how to update this thing since it's a tablet). The difference is huge, like 10W for plugged in to 15-20W when unplugged. It's very reproducible in this one particular app, but not in another app. The application is a high end game engine (not sure how much I can say since it relates to in-development products), so stresses both CPU and GPU.
I've used the unmodified "power saver plan" in both cases, so all the settings should be identical.
Thanks for the quick response!
First of all, I believe that you may have an old graphics driver -- please update this by going to the Intel Download Center and running the automatic driver update utility. I think there may have been a driver update in December for your platform.
Also, you might want to look at the following metrics at the same time as power: aggregated cpu load, gpu EUs active, GPU busy, and target app CPU load. And try seeing if there is any difference when you run "analyze application" locally, vs. running Intel GPA System Analyzer from a remote system connected to your target system. Another advantage of using a remote system is that you can display more metrics, and a key one might be processor frequency (for each of the processors) -- the reason for this is the latest Intel CPU's share power between the CPU and the GPU depending upon the load, and you might see a difference in the CPU frequency and the other metrics I mentioned above when running plugged in vs. on battery power. So try these and see what happens.
Also, it's strange that you see a big difference with one app, and no difference with another. Again, checking the other metrics besides power might show you something different with this one app. Might also be worthwhile to check other apps as well.
Hope this helps!
I will try with a remote app, but re: the driver I can say that I have tried this, but the driver I was pointed to failed to install with an error message saying it hasn't been validated for my platform. If you have a direct link to a suitable graphics driver for Surface Pro 2, I'd be happy to try it!
Update: I can't actually get per-app metrics (even with remote app) all I get is the system view.
When I unplug "GPU EUs active %" goes down (e.g. from ~30% to 24%.. relatively minor fluctuation in the two states). The overall CPU load seems roughly the same but it's very noisy. I have no idea why the GPU changes though. CPU clock frequency is not changed.
The specifics of those numbers change if I change what's on screen, but the pattern seems the same: GPU EUs active goes down when unplugging, everything else I've looked at looks roughly the same.
I'm checking with a few other people here on why this would occur.
Note that not being able to see "per-app" metrics implies that your app is not using DirectX.
Can you also look at GPU power and CPU power metrics, and see if something looks amiss when unplugging?
More data: the memory traffic graphs make a pretty huge jump when I unplug. With the power cord plugged in they're mostly at zero with the occasional jump to a few hundred. After I unplug it goes up to ~25K altogether. I'm not sure what the graph actually is (it's clearly not "bytes" like it says, because it's not monotonically increasing.. it must be bytes/some_unit_of_time...), but the difference is about two orders of magnitude.
Anyway, there's a big difference and as far as I understand memory traffic is a big power sink. That still doesn't explain *why* this happens. I'm rendering the same thing! Any ideas?
It is indeed using DirectX. I think there's a limitation in GPA for this particular configuration (I can't see the Intel GPA Sample either). I can't see separate CPU/GPU metrics, I only see "Package Power".
I sent you a private message with some additional thoughts on this issue. Please read that email, and reply via private message back to me.