On Thu, Jan 11, 2018 at 06:57:20PM +0100, Riccardo G Corsi wrote:
> Hello, I've updated firmware of my I3-4170 to 0x23 (LFS 8.1 kernel 4.14.12).
> 
> Some suggestions how to verify performance changes? Log files to look for?
> 
> Thanks
> 
> Riccardo
> 
What performance changes are you expecting ?

The updated microcodes are believed to mitigate Spectre attacks.  I
don't know if yours is one of the updated ones, but my i7-4790 is
also using 0x23 and that has a date of 2017-11-30 in dmesg.  If
yours is the same then yes, it is believed to include this
mitigation.

The revision numbers such as 0x23 are specific to the family - model
- stepping, others will have very different revision numbers.

If you are running a webserver, or perhaps a postgresql database, or
anything requiring it to process a lot of network packets, then you
might notice a slowdown.  But as everybody says - it depends on your
workload.

My i7 is mostly used as a desktop and for building LFS and BLFS.  At
the moment I'm running a series of LFS builds to try to determine if
there is any impact.  My impression is that the variations so far
are within normal tolerances, but this is the 'safe' set with the
new microcode and PTI, I'll do the 'unsafe' set with 'nopti' and
older microcode later.

The microcode is a black box, so apart from seeing the version
number and date, and the message
[    0.000000] Kernel/User page tables isolation: enabled
there will be nothing in the logs.

I suspect that the interesting thing will be the 'retpoline' kernel
mitigation - that is still under development / review and needs an
updated gcc.  I plan to test that eventually.  In the meantime,
although I don't regard the phoronix testsuite as generally useful,
there are some tests at
https://www.phoronix.com/scan.php?page=article&item=pre-pcid-kptiretpoline&num=1
But I think these are all from one run of each test, and my view is
that most changes are probably only the normal variability (e.g. try
building your own kernel config several times, from a clean start
and dropping the caches - I expect the times will differ by a few
seconds).

From the comments there, somebody reports a severe slowdown in
ffmpeg, maybe I should also test that.

ĸen
-- 
Truth, in front of her huge walk-in wardrobe, selected black leather
boots with stiletto heels for such a barefaced truth.
                                     - Unseen Academicals
-- 
http://lists.linuxfromscratch.org/listinfo/blfs-support
FAQ: http://www.linuxfromscratch.org/blfs/faq.html
Unsubscribe: See the above information page

Reply via email to