- Thank you received: 0
Testers Needed for New RTAI Packages
- twoflowers
- New Member
-
Less
More
7 years 2 weeks ago #147699
by twoflowers
Replied by twoflowers on topic Testers Needed for New RTAI Packages
Ok, did some extensive testing today. i3 quadcore 3GHz, lateny < 3uS ... this is 1/10th of the best value that RT-PREEMPT could do!
Adding 20xglgears gives this result:
- Still 1/5th of RT-Preempt.
Only caveat: display resolution is 1024x786 ... and I don't know why it won't go higher.
Adding 20xglgears gives this result:
Attachment not found
Only caveat: display resolution is 1024x786 ... and I don't know why it won't go higher.
Please Log in to join the conversation.
7 years 2 weeks ago #147706
by Bari
Replied by Bari on topic Testers Needed for New RTAI Packages
RTAI is only for x86.
Please Log in to join the conversation.
- Todd Zuercher
- Offline
- Platinum Member
-
Less
More
- Posts: 4739
- Thank you received: 2
7 years 2 weeks ago #147713
by Todd Zuercher
Who says that?
It's for what ever kernel architecture you can build it for.
Replied by Todd Zuercher on topic Testers Needed for New RTAI Packages
Bari wrote: RTAI is only for x86.
Who says that?
It's for what ever kernel architecture you can build it for.
Please Log in to join the conversation.
7 years 2 weeks ago #147714
by Bari
Replied by Bari on topic Testers Needed for New RTAI Packages
It's currently only working for x86 to answer twoflowers question.
Wow how did you miss this point?
Wow how did you miss this point?
Please Log in to join the conversation.
7 years 2 weeks ago #148114
by DomCa
Replied by DomCa on topic Testers Needed for New RTAI Packages
Hi!
I just installed the kernel and 2.8pre linuxcnc packages on a fresh debian (amd64).
The PC is an old athlon dual core with an old 512MB Radeon card.
Without any tweeking I am getting below 10.000ns latency (copying, watching youtube and running some glxgears). For quick testing I used latency-test and as second test set the base_thread of the "axis_mm" to 10.000 and I am not getting any realtime error in linuxcnc. 5.000 is not possible on this PC.
I also tried the knwon "tricks":
isolcpus=1 acpi_irq_nobalance noirqbalance processor.max_cstate=1 intel_idle.max_cstate=0 radeon.modeset=0
However: Latency became totaly uncontrolable up to 100s of thousands.
Is the new RTAI different in this aspect?
Best regards
DomCa
I just installed the kernel and 2.8pre linuxcnc packages on a fresh debian (amd64).
The PC is an old athlon dual core with an old 512MB Radeon card.
Without any tweeking I am getting below 10.000ns latency (copying, watching youtube and running some glxgears). For quick testing I used latency-test and as second test set the base_thread of the "axis_mm" to 10.000 and I am not getting any realtime error in linuxcnc. 5.000 is not possible on this PC.
I also tried the knwon "tricks":
isolcpus=1 acpi_irq_nobalance noirqbalance processor.max_cstate=1 intel_idle.max_cstate=0 radeon.modeset=0
However: Latency became totaly uncontrolable up to 100s of thousands.
Is the new RTAI different in this aspect?
Best regards
DomCa
Please Log in to join the conversation.
7 years 2 weeks ago #148116
by DomCa
Replied by DomCa on topic Testers Needed for New RTAI Packages
twoflowers: Can you tell me how to get the histograms you are showing?
Thanks!
DomCa
Thanks!
DomCa
Please Log in to join the conversation.
- tommylight
-
- Away
- Administrator
-
Less
More
- Posts: 22012
- Thank you received: 4
7 years 2 weeks ago #148117
by tommylight
Replied by tommylight on topic Testers Needed for New RTAI Packages
That is a very good latency, like extremely good, so why are tweaking anything ?
If you have the base period set at 10.000 and no errors pop up, you can be very happy.
For the histogram, open terminal and type:
latency-histogram
If you intend to use Mesa boards, type:
latency-histogram --nobase
If you have the base period set at 10.000 and no errors pop up, you can be very happy.
For the histogram, open terminal and type:
latency-histogram
If you intend to use Mesa boards, type:
latency-histogram --nobase
Please Log in to join the conversation.
7 years 2 weeks ago #148126
by DomCa
Replied by DomCa on topic Testers Needed for New RTAI Packages
I know. The tuning is just out of curiosity... I can currently run my CNC at full recommended speed of 6m/min with 3200 microsteps with the official rtai linuxcnc download (and heavy tuning).
I never managed to come even near this values with the preemt version.
Will the rtai5.x and newer debian versions than currently available be an official version in the future?
With this version I had the low latencies almost instantly.
I never managed to come even near this values with the preemt version.
Will the rtai5.x and newer debian versions than currently available be an official version in the future?
With this version I had the low latencies almost instantly.
Please Log in to join the conversation.
7 years 2 weeks ago #148133
by Bari
This is also why we post the RTAI source and build instructions here: github.com/NTULINUX/RTAI so that you can build a working version of RTAI yourself.
Replied by Bari on topic Testers Needed for New RTAI Packages
That is entirely up to the dev that builds the next LCNC ISO's.DomCa wrote:
Will the rtai5.x and newer debian versions than currently available be an official version in the future?
This is also why we post the RTAI source and build instructions here: github.com/NTULINUX/RTAI so that you can build a working version of RTAI yourself.
Please Log in to join the conversation.
- twoflowers
- New Member
-
Less
More
- Thank you received: 0
7 years 1 week ago #148410
by twoflowers
Replied by twoflowers on topic Testers Needed for New RTAI Packages
@RPi & RTAI: Ok, I had an old RPi 2 some years back with a RTAI kernel (3.something). Latency was good, ~ 1/10 of the latest RT-PREEMPT kernel, so I was wonndering if anybody did get taht working on newer kernels, too.l Anyway, I have now a stock of 9 perfectly working PCs with i3, so no need to go the RPi road any more. In hindsight, I would not go the RPi road any more.
Please Log in to join the conversation.
Time to create page: 0.149 seconds
