I will buy a new CPU and want to find out how much the new CPU will improve compile speed.

So I'm looking for any DBs for SBUs of each CPUs.
Is there any?

Or should I make one to use?

I've never heard of something like that. It sounds like a great project.
The only problem I see is that there is such a great variance between different distros, For example if you dual booted a bloated distro and a finely tuned minimal distro, you're going to get two different SBU.

To overcome this you'd have to implement some average features of some sort.

If you also think this is a good idea, then how can I open more broad conversations?

Well I think this is the right place, there's quite a few LFS users (including me ) on LQ. My guess it'll take some time before everybody sees this thread.

Good Luck!

I've done this in the google search bar, say something like
gets this
http://cpu.userbenchmark.com/Compare...M/m38171vsm949
and intel has a comparision on their page for their cpu's
https://www.intel.com/content/www/us...sors/core.html

but for compile time that is too selective and a lot of variables that no one could know about to make such a deceive comparison.

Probably there is not any.

I don't think it is a good idea. So many processor type.
Probably using same CPU, but different host would cause different result.

You should use same host distro (with same packages installed) for each CPU to make it sound fair.

Maybe all LFS user could post their SBU time (with their built environment information; CPU Type, Host, etc).
And then we need volunteer to collect all informations to create the database.

There are too many hardware configurations for this to be practical. It depends on the MB, socket type, CPU type, memory speed, DDR4 vs DDR3, overclocking, spin drive vs SSD vs M2, the background daemons, etc.

Even the SBU as defined in the book is somewhat elastic. They define 1 SBU as the time taken to compile binutils as per chapter 5 - Pass 1. In chapter 6, binutils is supposed to take 6 SBU. For me, it took 3.2 SBU.

On my desktop, 1 SBU has a wall clock time of 22s. This is an i7-7700k with 3000 GHz DDR4 with XMP on and a Mushkin SSD.

On my laptop, the wall clock is so long it is not even worth recording.

Dear, My Mentor

I also experienced what you said, such as different from what SBU expected.
But I mean it could act like a rough indicator to build LFS to NEWBIE.
Just an indicator, not something precisely correct with every package building process.

For example, if I'm first trying to build LFS, and don't even know how much it will take, I'll be exhausted.
(Can't expect anything and just walk for first SBU package, which is now Binutils - 1 pass, and record the time.
How about it turns out too long for him to go through, someone expected "OK, I hope it'll take 10~20 minutes, than I'll go through.
But turns out it took 39min. He will so much disappointed and reboot to some fancy GUI installer.)

How about just simple Database.

[Needed information] CPU model / Host distribution / Ram type and speed / root on SSD or HDD or others
[Expected information] MB / Daemons / Overclocked or not
[More Information] User input.

Well, you can always calculate the wall clock time for 1 SBU and then add up the total SBU for all the chapters to get a final estimate. But this is just a best case scenario.

For me, this would still not be an accurate guide for how long it took me to build LFS. My LFS build did not boot first time. I spent more time than the sum of all SBUs in the book to troubleshoot the problem.

I you read this forum, you will see a lot of OPs troubleshooting errors which take up more time than a sum of SBUs would represent.

OK I got it.
So I'll abandon that idea !
Thanks for your suggestion!