Skip to content

Commit

Permalink
HFI warning, minor cosmetic fixes
Browse files Browse the repository at this point in the history
  • Loading branch information
@ThomasKaiser
ThomasKaiser committed Oct 16, 2022
1 parent a108283 commit 6174c35
Show file tree
Hide file tree
Showing 5 changed files with 661 additions and 3 deletions.
1 change: 1 addition & 0 deletions Results.md
View file
Original file line number Diff line number Diff line change
Expand Up @@ -171,6 +171,7 @@ So do **not** rely on collected numbers unless you carefully read through all th
| [x5-Z8350](http://ix.io/1HnC) | 1920/1680 MHz | 4.15 | Bionic amd64 | 4710 | 120300 | 207640 | 2740 | 3140 | - |
| [x5-Z8350](http://ix.io/2Jdb) | 1920/1680 MHz | 5.4 | Focal amd64 | 4790 | 129810 | 237230 | 3170 | 2960 | 9.38 |
| [Celeron J1900](http://ix.io/41Kb) | 2000/1333 MHz | 5.4 | Focal amd64 | 5530 | 34060 | 28860 | 3550 | 3400 | - |
| [Atom E3950](http://ix.io/4dd5) | 2000 MHz | 5.15 | Jammy amd64 | 6440 | 266710 | 374800 | 4400 | 4840 | - |
| [Celeron J3455](http://ix.io/1m5p) | 2300/1500 MHz | 4.17 | Stretch amd64 | 7000 | 316480 | 429660 | 4090 | 4050 | 17.26 |
| [Pentium N4200](http://ix.io/1ngq) | 2560/1100 MHz | 4.14 | Bionic amd64 | 7469 | 354328 | 468008 | 4682 | 4997 | 18.75 |
| [Pentium J4205](http://ix.io/1m5t) | 2560/1500 MHz | 4.17 | Stretch amd64 | 7570 | 355540 | 480640 | 5070 | 5170 | 18.82 |
Expand Down
Loading

1 comment on commit 6174c35

@ThomasKaiser
Copy link
Owner Author

@ThomasKaiser ThomasKaiser commented on 6174c35 Oct 17, 2022 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

As for the HFI warning... it seems neither taskset nor cgroups are sufficient to pin tasks to the E cores. At least my tries were to no avail: this is an i9-12900T in extensive mode: http://ix.io/4doc

7-zip MIPS when pinned to cpu0 is 6356 while pinned to cpu16 (supposedly an E core) is 4401 which is simply impossible with an E core. Different cpufreq policies do apply (cpu0 = 4900 MHz, cpu16 = 3600 MHz) and most probably this makes the difference while also the benchmark run pinned to an E core has been executed on a P core in reality with reduced clockspeed.

Same with multi-threaded numbers: CPUs 0-15 score 35600 7-zip MIPS (P cores with SMT enabled) while when pinned to CPUs 16-23 the score is 31800 which is impossible with E cores and again hints at the benchmark in reality running on the P cores.

The only somewhat useful sbc-bench mode with these hybrid Intel thingies is MODE=extensive sbc-bench -G to generate Geekbench scores and compare there 'all cores' vs. 'only P cores' to get an idea about the impact the (missing) E cores have: http://ix.io/4do3 or to be more precise: https://browser.geekbench.com/v5/cpu/compare/18004391?baseline=18004480

This works for the sole reason sbc-bench sending CPUs 16-23 offline prior to calling Geekbench. The other variant to only test on E cores doesn't work since cpu0 can't be sent offline.

Further reading: https://www.agner.org/forum/viewtopic.php?f=1&t=79

Please sign in to comment.