Re: Disabling CPUs

On Fri, Oct 7, 2022 at 9:56 AM Ian Smith <smithi@nimnet.asn.au> wrote:

> On 7 October 2022 2:05:32 pm AEDT, Kevin Oberman <rkoberman@gmail.com>
> wrote:
>
> [...]
>
>  > I've been silent while waiting for the additional memory since I
>  > could not
>  > do any real work with only 8G. It finally arrived and I installed it
>  > yesterday. I'm very happy to say that my VirtualBox VMs seem to now
>  > be
>  > working correctly. I've done limited testing, but seen no problems to
>  > this
>  > point. Much better than an instant system panic when a VM finished
>  > loading.
>  >
>  > On the plus side, with "cpuset -s 1 -l 0-3", things seem reasonably
>  > stable.
>  > That means no crashes with many hours of use and a lot of
>  > compilations.
>  > With only 4 cores, it's not a fast as I had hoped, but I will be
>  > benchmarking soon.
>
> Hi Kevin, three questions if I may:
>
> I may have missed it, but did you wind up testing sched_ULE vs 4BSD?
>

cpuset(8) only works with ULE, so that is what I am running. So far I have
not seen the interactivity issues that caused me to move to 4BSD over a
decade ago.


> Is that 4 cores including HT, or not?
>

Yes, 2 cores and 4 threads. Sorry to have failed to make that clear. My CPU
has 2 P cores and 8 E cores. It has never crashed  when running the 2 P
cores, but I had one crash running the 8 E cores. Now that I have a stable
system, I may try switching back to just the E cores as they  seem to
provide almost double the performance of the 2 P cores. Oddly, while using
rsync to move a big batch of files from my backup disk to the new system
after cpuset, I saw 0-3 busy, as I would expect, but one E core was running
at about 30% the entire time, as well. I have no idea why.

>
> And how much power does your new Thinkpad T16 use on battery, say, at idle
> and 'fairly busy' (acpiconf -i0) ?
>

With  the 2 P cores I see about 1.1W idle. Wow! While compiling Firefox
(all  threads running at over 95%), I see about 25.3W. With a 52500 mWh
battery, it will last a long time.
With the 9 E cores I see about 10.4W idle. Wow, the other way. While
compiling Firefox, I saw a momentary spike of 35W, but it varied between
20.5 and 27.8 W with occasional hits up to 31 W. The E core power
consumption seems far more inconsistent thaa the P cores. compile time with
P cores was 1:29 while it was only :50 with the 8 E cores.

In both cases I had my Mate desktop running.

The L15 was a piece of junk where I learned my lesson about budget
systems.  The ABS case started cracking at both hinges after about a year
of use. That was after about 9 months of trying to get it working properly
after Lenovo had replaced the motherboard and the disk. and, finally, the
keyboard. I think the MB was fine, but was the obvious suspect for system
lock-ups in POST. Took my a while to track it down to the keyboard and the
tech who swapped it out didn't believe that the keyboard was the problem
but agreed to humor me. As a result, by the time the case started cracking,
the warranty had expired. If I had been willing to give up the TrackPoint,
I would have bought a non-Lenovo/IBM for the first time ever.

The T16 is built like a T series with a metal frame and feels solid. My T16
is all SSD. Yes, it's heavier. The L15 flexed noticeably with no frame,
just the MB, battery, and disk attached to the ABS case.

cheers, Ian
>
>
>

-- 
Kevin Oberman, Part time kid herder and retired Network Engineer
E-mail: rkoberman@gmail.com
PGP Fingerprint: D03FB98AFA78E3B78C1694B318AB39EF1B055683