So, it seems (OzzFan)- a potential optimization of the free-core concept, might be to free a Virtual Core/Thread(if I can figure that out) for the GPU apps?
If that makes sense, would I do that in P Lasso Affinity setings?
Or just reduce cores to 6 or 75% in Boinc?
=
Re: PCIe thing, I did just discover that my new i7 4790K Quad/8-Thread I'm building has 16 PCI Lanes just like my i5 Quad/4-Thread; vs the i7 5960X 8 Core/16 Thread's 40 Lanes
My point actually is that the CPU apparently *is on PCIe Bus or it's part *of the older FSB I/we grew up with or mebe it's connected to that bus *because of the iGD
Smoke and Mirrors I tellya! ;)

---

So, it seems (OzzFan)- a potential optimization of the free-core concept, might be to free a Virtual Core/Thread(if I can figure that out) for the GPU apps?
If that makes sense, would I do that in P Lasso Affinity setings?
Or just reduce cores to 6 or 75% in Boinc?

A free virtual core is good enough. I don't use P Lasso myself, and in general I don't think it's wise to override the OS's ability to spread the load across CPUs (affinity locks to a single CPU), so I'm not sure how P Lasso fits into freeing a core. It doesn't have to be the same core; letting the OS handle it is sufficient enough.

And yes, I use BOINC's settings to free up a core. If you want to free up a single core on an 8 core machine in BOINC, set the CPUs to 87.5%.

My point actually is that the CPU apparently *is on PCIe Bus or it's part *of the older FSB I/we grew up with or mebe it's connected to that bus *because of the iGD
Smoke and Mirrors I tellya! ;)

No, not entirely correct. What you may have called the "northbridge" part of the chipset in older PC designs - many of those functions were moved into the CPU, including the memory controller and bus controller. Which means that the bus has a direct connection to the CPU, but the CPU doesn't sit on it or gain any speed from it. Only the reverse is true; devices sitting on the local bus gain a small speed benefit from a direct connection to the CPU, which would be why the memory and bus controller were moved into the CPU.

So yes, you will see references to how many PCIe lanes a CPU supports, but only because the bus controller is built into the CPU now.

Would be interesting to see if "virtual core" would be enough indeed.
It depends of why GPU driver need that CPU cycles in first hand.

It's possible to free CPU via affinity (to set every process on host to run on all CPUs but particular one). But that way quite "real" CPU will be freed, not "virtual". What's "virtual core" ?

Would be interesting to see if "virtual core" would be enough indeed.
It depends of why GPU driver need that CPU cycles in first hand.

It's possible to free CPU via affinity (to set every process on host to run on all CPUs but particular one). But that way quite "real" CPU will be freed, not "virtual". What's "virtual core" ?

===
I'm with *you Raistmer, "Why" indeed!?!
=
My *understanding of a "virtual core" is that of a Threaded one
eg = i5 2500 Quad core has 4 Threads and is *seen as 4 processors
vs = i7 4790K Quad core has 8 Threads and is *seen as 8 processors
(4 physical + 4 logical or "virtual" cores)
=
My *experiments with Affinity are a bit Thin yet.
I *have been wanting/thinking to assign CPU0 to Windows et al and the rest to Programs/Data = with another potential secondary division of assigning 1or2 cores to GPU apps(similar to core-lock switch effect)
Then maybe tell Boinc to just use 7 or 6 cores (leaving 1 or 2 core(s) "free" as a catch-all)
=
I do tend to over-complicate things, but it's my way of unraveling, then tweak-optimizing the inherently complex/complicated stuff(resources) we deal with.

---

My *understanding of a "virtual core" is that of a Threaded one
eg = i5 2500 Quad core has 4 Threads and is *seen as 4 processors
vs = i7 4790K Quad core has 8 Threads and is *seen as 8 processors
(4 physical + 4 logical or "virtual" cores)

All CPUs run "threads", therefore all CPUs are threaded. A "virtual core" is when a CPU supports Simultaneous Multithreading or SMT, which effectively virtualizes all the CPU registers and allows two or more threads to run on a single, physical core/CPU.

Ya, touche' there.

2 Threads per vs 1 per in my CPU comparison
Second Thread assumes the Logical/Virtual moniker.

Very much like Partitions and Logical Drives within one HDD
=
Edit= Surely, *needing a free core isn't necessary with a Hyperthreaded CPU
Wasn't this advice intended to *apply another Thread (not a whole core) in the case of a non-hyperthreaded CPU where the core-thread ratio is 1:1?

---