I would imagine that PrimeGrid is performing almost exclusively integer mathematics, allowing the CPU's floating point unit to draw close to zero power.

PrimeGrid task is running on the GPU and two other tasks (Asteroids@home) are running on the CPU.

So you mean to say that collectively CPU and GPU draw much more power on full load than they are designed for? Aren't these chips designed and tested to be on full load at all times?

I would expect that the chip is tested under full load, if only to see that its power management works and the chip doesn't burn down or that it doesn't have any weird bugs that show up only during full load. What it is designed for is totally different question. People like us who try to run every component of the system to their full potential must be in minority. A more common scenario would be where the load goes from CPU to GPU and back.

Also this doesn't explain the fact that I was able to run PrimeGrid task and two CPU tasks at the same time with CPU and GPU both under 100% utilization without CPU being "power throttled".

Different apps behave differently. For example, the data for one app might fit nicely in cache while the other one often needs to fetch data from RAM. While CPU is waiting for data from RAM it can't do anything and consumes less power. Or a fancy thermal/power manager might notice that CPU is constantly waiting for RAM so the CPU could be run a bit slower allowing some other part to run a bit faster.

The thermal/power manager may use a lot more inputs than what is made visible for software and it is most certainly running some proprietary algorithm revealed only to AMD engineers. While trying to reverse engineer it might be some sort of fun I would recommend you to just set a manual overclock if your system allows that.

Here's some reading for you:
AMD Turbo Core
CPU + GPU = APU: East Meets West (Yours is Richland, Llano is two generations earlier.)
New Details Revealed on AMD's Upcoming Richland Chips

As you suggested, I did try overclocking it manually by disabling the turbo core and adjusting the clock multiplier by AMD's OverDrive. It seems to have no effect on the earlier issue as the clock speed is still 3Ghz while the multiplier is set to 41x (4.1Ghz).
Read the resources provided by you and did a bit of searching myself and found out this to be a common issue among APU users.

One thing that I noticed is that my WUs from some projects take exceptionally long to complete and the credit granted is not at par with the CPU time provided. Could be due to clock speed instability?

Well bummer.

Clock speed dropping doesn't help with CPU tasks' run time but the real problem could be lack of RAM bandwidth. According to one of the articles GPU's RAM access is prioritised. It could be that CPU is spending time waiting for RAM.

On my system running tasks on Intel GPU triples run times of CPU tasks. So not worth it.

Well I had different opinion on running GPU along with CPU. I thought it would be worthwhile to have 3 tasks running simultaneously but now it seems the GPU is compromising the output of CPU to such an extent that the output of GPU is not really worth it. Never knew that RAM bandwidth could be a problem.

So do you suggest to suspend the iGPU completely in order to get most out of my CPUs or should I let it run as it is and keep with the issues? All I want is to contribute as much possible from my computer as it can provide.

The Intel GPU equals to about 1.5 CPU cores so for this system it's not worth it. At some point I'll add another memory module which should put memory access to dual channel mode. That may make Intel GPU more interesting option.

You'll need to measure what's best for your system. It could be that limiting CPU or GPU to only one or few projects gives best throughput. My experience is with Seti only.

I observed some strange things:
1) Whenever I reboot my computer the clock speed is stuck at 1.7Ghz and doesn't go up even under full load. Even the GPU runs at 300MHz instead of 800Mhz.
2) Shutting down the computer and turning of the mains supply and then let the capacitors discharge and then power on the system the clock speed stays at 4.1GHz (Even on 100% CPU-GPU load regardless of the tasks running) and sometimes repeating the same procedure yields the other issue of throttling to 3GHz under GPU load. GPU is running at 800MHz always in this scenario.

I suspect that the PSU may be bad or the motherboard itself may be faulty. Need some insight on this strange behavior!

If your motherboard provides sensor output to software you could see how stable the various voltage lines are. They have rather wide tolerances but if there are large swings between idle and fully loaded that hints that the PSU is going bad. Alternatively you could use a multimeter but you can easily access only some of the voltage lines.

Motherboard you could inspect visually if there's any bulged or leaked capacitors.

If you have available or can borrow easiest is to swap in a known good part.

As per HWMonitor the voltages seem quite stable but the power seems to fluctuate. To be sure I used multi-meter to check the CPU Core supply( 4 pin connector) and it was a stable 12 volts.
Visual Inspection of motherboard yields no abnormalities either.

If you have available or can borrow easiest is to swap in a known good part.

I think I don't have any compatible hardware to test these parts.

If you are really lucky firmware might have generated an event and Windows might have logged it which might give some more insight into the problem.

Have you checked if there are BIOS updates?

Other than that I don't have any better ideas but to run the system until it fails for good.

For those who attempt to configure APU for maximum performance with SETI BOINC this could be of some interest:
http://lunatics.kwsn.info/index.php/topic,1735.0.html

---