Appreciate some Thoughts on latency results

When you say you ran the machine it was a real machine, not a sim? Sims often trigger a latency fault for me but never on the real hardware.

Yep on a real machine - (cutting metal... )

I was surprised to see the error as everyone seemed confident in my numbers, so I restarted the session and continued to work for a while and it didn't reoccur, but I'm sure it would have reoccurred sooner or later...

Hopefully the SMI fix did the trick....

Ok after several days the performance is good. I have seen one more occurance of the real time error, it's occurred before I start maching and seems to have no impact on machine performance. Latency numbers still well within requirement for mesa card.

Might try a graphics card at some point but for now I'm just going to roll with it until I get my back up pc built

Just FYI,

I also have this repo:

github.com/NTULINUX/smi-killer

Disables SMI on Intel chipsets using standalone module, ported from RTAI:

github.com/NTULINUX/smi-killer

Default value passed is 0x1 but can be controlled via ./setsmi hal_smi_masked_bits=<variable here> (no <>

Script:

github.com/NTULINUX/smi-killer/blob/master/setsmi

I also have 5.4 RTAI kernel debs for Bullseye here:

github.com/NTULINUX/RTAI/releases/tag/v5.3.3-delta

Some boards preform better with PREEMPT_RT, others have lower latency with RTAI.

In either case, please refer to this document for a list of problematic BIOS settings:

github.com/NTULINUX/RTAI/blob/for-next/README.BIOS

Using this PREEMPT_RT kernel config btw, I had latency under 1 microsecond (yes, one) with my NIC disabled in BIOS, no background tasks:

github.com/NTULINUX/gentoo_backup/blob/m...fig-6.0.0-rc1-rt2-RT

Will only boot as-is on AMD platforms with AHCI or NVMe block devices (XFS and EXT4) Does not support SystemD either (i.e. will not work with Debian)