BOINC does some really strange things when running multiple projects. For no apparent reason to humans it will stop running task, other times it will start a task on a project, stop that one and start a new task from the same project.... All very frustrating.

I would guess that actually measuring CPU usage with a mixed-core CPU like yours is going to be a bit of a challenge. Does the Windows Task Manager give any indication of which tasks are running on which core type?

Fingers crossed is about all we can do until there is a better understanding of how BOINC & Windows are working with your CPU - let it run for a day or so and see how (if) it settles down.

---

Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?

Back to 25-45%. Still running 7 MilkyWay WUs at 4 CPUs each. HWiNFO64 still shows 8 P-cores and 12 E-cores in use. Maybe Task Manager isn't getting it right, and the computer was always running at higher capacity. Let's see if the statistics change.

Once again, thanks for your help.

Back to 25-45%. Still running 7 MilkyWay WUs at 4 CPUs each. HWiNFO64 still shows 8 P-cores and 12 E-cores in use. Maybe Task Manager isn't getting it right

Task Manager is displaying the information in a bit weird way, but it's not completely wrong what it is showing. Your CPU has a 28 seconds PL2 Tau Limit, so it can run at 100% max. 28 seconds and only if that's possible within the 219W limit, after it will limit itself to 65W. And that is of course only possible far away from the 5.4GHz turbo clock, which the Task Manager considers as 100%.

I have that also on my laptop, there however not because of some power limits, but because of thermal throttling. As long as the CPU runs at full speed, Task Manager shows ~99%, if the CPU downclocks itself it will show 89% or less, depending on how much it has to throttle, even if actually it's still 100% in use, just at lower clock speed.

HWiNFO64 is different, it considers the base clock as 100%, so it might even show values over 100%, I've seen up to 140% on my system when I've let it boost up to 4.65GHz with all 16 threads in use, however not with Milkyway in MT-mode as that never uses 100%, so don't expect that.

Long story short, what you want to see is that all cores are in use and that's apparently the case now. If you want to further optimize your setup you might want to experiment with less cores in use considering the quite low power limit for all those cores your CPU has, but I'd recommend doing it with some project with constant WU runtimes like for example WGC's MCM1, eventually while running 1-3 MW WUs at the same time if that's some realistic mix in your case. That's a lot easier as you see the changes directly, here you can only guess unless you want to run benchmarks with a set of selected WUs. But that's not worth the effort IMHO when you run several projects anyway.

---

has a 28 seconds PL2 Tau Limit

Ouch - that's quite a serious limit to be hitting :-(
What is not obvious is the "cool down period" - how long one has to wait until being able to use the full turbo speed on all cores, but with the PL1 being about a third of the PL2 I would guess it's quite long - very disappointing :-(

From reading around it would appear that when PL2 is hit the processor drops to its base frequency, and maybe dropping -E cores (some disagreement between articles) to get down to the lower power limit of 65W.

AS Link says, the most probable solution is to limit, via BOINC, to the number of cores in use by Milkway (and other BOINC-based tasks), start low (e.g. 4) and see what happens. Once you've seen what happens increase the "BOINC" cores to 8. This should "fill" the basic E-cores without hyper-threading, but may pull in some -P cores instead. Does HWINFO64 let you see the power being used by the CPU?

-----------
note - where I say "via BOINC" you can either use the server settings, or the local settings, but remember not to expect running tasks to change number of threads as this an cause, indeed it might be worth setting "no new tasks" and running any tasks ready to start or waiting to run.

---

Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?

Right now, with the way I have my current four projects and their resource shares set up, there is not one with enough constant run times. However, I will be changing the resource share distribution in mid-December and I will try what you suggest with Einstein. WCG is currently not providing very many WUs.

Thanks once again.

What is not obvious is the "cool down period" - how long one has to wait until being able to use the full turbo speed on all cores

AFAIK you must remove enough load so that the CPU drops below the 65W. No idea, how long you need to stay below, but it's not enough to remove just some of the threads, it will than boost the cores for the remaining threads and stay at 65W. So for BOINC it means you have 65W.

Does HWINFO64 let you see the power being used by the CPU?

It let's you even see what each core uses, at least if the hardware supports it.

In theory running on all cores with lower clocks should be more efficient than running on fewer cores and higher clocks, but of course that depends on the cache size, RAM speed and also the applications you run, so as usual, YMMV. Some people simply turn off E-cores in BIOS and only use the P-cores.

---

I don't see the power being used by the CPU, or by the cores.

That's a shame :-(
What about core temperatures and core clock speeds?

---

Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?

You must start in the sensors view, not summary.

---

CPU Package Power: Average 65.317W
IA Cores Power: Average 58.216W
GT Cores Power: Average 0.008W
System Agent Power: Average 6.100W
Rest-of-Chip Power: Average 0.327W

CPU Package Power: Average 65.317W

There's your 65W limit.

---

So, I guess I'll have to be happy with the current (no pun intended) situation.
Once again, thanks for your help.

Well, AFAIK you can't get that CPU to use more power, but that does not mean, that there's nothing you can do to get more work done per day. Like I said before, some people crunch only on the P-cores or even disable the E-cores completely in the BIOS to get all the power to the P-cores. So there are few things you can try, but like I said, better do that with a project with constant (and not too long) WU runtimes.

If I had a CPU like this, I'd first try to run project applications on the P-cores only (with process priority 4 they should stay there) and leave the E-cores for the system and other stuff. The next step would depend very much on what I'd see during those runs, in particular clocks, wether the tasks actually stay on the P-cores and of course how the system behaves with all P-cores used by high priority BOINC tasks.

---

I will be rebalancing the resource shares on the three non-Rosetta projects in mid-December, and it will take some time for them to settle down to near constant runtimes. Hopefully there will be one with not-too-long ones. I'll take a look at running a project only on P-cores in January. Thanks again.