- Posts: 381
- Thank you received: 0
Results of latency test, list of computers tested for use with LinuxCNC
i7s usually have more cache (i7 3770: 8MB, i5 3570: 6 MB), which should/could result in better latency. So, it was a wise decision to take an i7.faeluke wrote: ...I opted for the I7, although I am not sure if it will make difference with everything turned off in the BIOS.
So: congrats! And if you ever were, you are certainly no noob anymore... so getting this all up and running with Mint on a new PC and posting the results was for me, a Christmas miracle.
In Bios, the most important part is to disable evrything that smells like power saving and hyperthreading....I had set multiprocessor -> disabled in the BIOS, Is this the same as running with Isolcpus?
while playing with bios version, note: hp says you cannot roll back to an earlier version. Usually because ther seemingly are "security" reasons. Isolcpus does disable cpus, but lets you chose which ones. If you read more, you will see posts where someone got best results isolating the lower cpus and others with isolation the upper cpus. Therefore, I prefore leaving multi core on and use isolcpus. There are machines out there, where isolcpus is of zero effect.On your machine, BIOS 2.99 appears to have the best numbers. I will update to 2.99 and try that out.
I will also take at modifying the Grub command line to see if I can improve the results further.
I tried to get a clearcut overview here . That's for a dell optoplex 980 though, but I hope, it is inspiring enough to at least serve as a hint for other pcs.Are you asking me to post the BIOS settings up to this point or to post the additional changes once I complete additional testing?Kindly also state, what options you altered.
Greez
chris
Please Log in to join the conversation.
tommylight wrote: Set the base period higher than the latency result shows, at least 20 to 30% higher.
Tommy
I will add a hint stating exactly that when a basethread is being used.
greez
chris
Please Log in to join the conversation.
macsddau@linuxcnc wrote: How Latency test are reliable ?
As you can see - and that was the very purpose of leaving/protocolling the very call for the latency test in the script's output - the very command is there in every output (quote taken from your post):
I myself do have, say a certain doubt too. If you run the test without a base thread, then the servo test results seem to return about double the values as with a base thread. I cannot see areason for that, raised my doubts, but so far, noone jumped in with an explanation.
So, maccsddau, I am with you raising at least an eye brow.
The script does nothing but logging hardware, prepare automatically glxgears, webbrowser, the commandline for latency-test and log the temperature, latency data and later more. I have written lcnc-hw to provide say a "standardized" way to occupy the pc while running the latency test and collect info of the pc. The main goal still is to provide repeatable and comparable results. I did not and will not mess with the core programme: "latency-test". I might later look into the code of latency-test to see or understand what causes double times for servo thread when running without base thread compared to with it.
greez
chris
Please Log in to join the conversation.
- Project_Hopeless
- Offline
- Elite Member
-
- Posts: 198
- Thank you received: 0
Linux Buster:
Linux Mint:
What is a good range? Even with Buster my numbers seem quite a bit higher relatively speaking than other posts here. I am using the onboard GPU, how much of an impact would that be?
Please Log in to join the conversation.
In my testing, I found that that was better than turning them with isolcpus= statement. I took extensive data over the weekend but did not have time to post.
Please Log in to join the conversation.
I am running servos, so I am not concerned about the PC... Data uploaded for information purposes.
su luke -c "latency-histogram --nobase --base 25000 --sbinsize 1000 --nox 2>&1 | tee -a 1609028912.lat &"
Does the script report the actual line run? Is there an issue with the --nobase switch appearing in the command line?
Common to All
================================================================================
General info:
Hewlett-Packard, HP Compaq Pro 6300 SFF
model name : Intel(R) Core(TM) i7-3770 CPU @ 3.40GHz
Kernel 4.9.0-13-rt-amd64
================================================================================
Bios changed with Base Thread
Bios version K01 v02.99, dated 09/15/2015
================================================================================
CPU related data:
cpu cores : 1
================================================================================
GRUB_CMDLINE_LINUX_DEFAULT="quiet splash"
================================================================================
Command for latency test is:
su luke -c "latency-histogram --nobase --base 25000 --sbinsize 1000 --nox 2>&1 | tee -a 1609093699.lat &"
================================================================================
last latency data is as follows:.
7200 secs servo min: -39.200 uS max: 42.000 uS sdev: 2.200 uS
7200 secs servo min: -39.200 uS max: 42.000 uS sdev: 2.200 uS
No Base Thread
********************************************************************************
Bios version K01 v02.90, dated 07/16/2013
su luke -c "latency-histogram --nobase --sbinsize 1000 --nox 2>&1 | tee -a 1609083174.lat &"
================================================================================
last latency data is as follows:.
7201 secs servo min: -19.600 uS max: 21.300 uS sdev: 2.100 uS
7202 secs servo min: -19.600 uS max: 21.300 uS sdev: 2.100 uS
With Base Thread 4 cores Isolcpus=2,3
********************************************************************************
Bios version K01 v02.90, dated 07/16/2013
cpu cores : 4
GRUB_CMDLINE_LINUX_DEFAULT="quiet splash isolcpus=2,3"
================================================================================
Command for latency test is:
su luke -c "latency-histogram --nobase --base 25000 --sbinsize 1000 --nox 2>&1 | tee -a 1609038229.lat &"
================================================================================
last latency data is as follows:.
7203 secs servo min: -48.500 uS max: 45.100 uS sdev: 2.300 uS
7204 secs servo min: -48.500 uS max: 45.100 uS sdev: 2.300 uS
With Base Thread 4 cores Isolcpus=3
********************************************************************************
Bios version K01 v02.90, dated 07/16/2013
cpu cores : 4
GRUB_CMDLINE_LINUX_DEFAULT="quiet splash isolcpus=3"
================================================================================
Command for latency test is:
su luke -c "latency-histogram --nobase --base 25000 --sbinsize 1000 --nox 2>&1 | tee -a 1609028912.lat &"
================================================================================
last latency data is as follows:.
7200 secs servo min: -41.100 uS max: 38.800 uS sdev: 1.900 uS
7201 secs servo min: -41.100 uS max: 38.800 uS sdev: 1.900 uS
Please Log in to join the conversation.
clifo wrote:
Code:LinuxCNC pc tests, version 20200930, started 09.12.2020 06:59:30 ******************************************************************************** General info: , Bios version , dated Chipset: Harddisk is non-rotational, ie. SSD The amount of swap currently used is 0 ================================================================================ CPU related data: model name : Intel(R) Celeron(R) CPU J3455 @ 1.50GHz cpu cores : 4 stepping : 9 cache size : 1024 KB ... ================================================================================ Kernel related data: Kernel 4.19.152-rtai-amd64 GRUB_CMDLINE_LINUX_DEFAULT="quiet isolcpus=2,3 idle=poll acpi_irq_nobalance noirqbalance lapic intel_idle.maxcstate=0 processor.max_cstate=0 ipv6.disable=1 pci=routeirq " Cpu idle driver: ================================================================================ LinuxCNC related data: LinuxCNC version is: =============================================================================== Swap useage: prerun: 0, running: 0 ================================================================================ CPU core temperatures: Core 0: Tmin: 36.0°C, Tmax: 42.0°C, (high = +105.0°C, crit = +105.0°C) Core 1: Tmin: 36.0°C, Tmax: 42.0°C, (high = +105.0°C, crit = +105.0°C) Core 2: Tmin: 33.0°C, Tmax: 39.0°C, (high = +105.0°C, crit = +105.0°C) Core 3: Tmin: 33.0°C, Tmax: 39.0°C, (high = +105.0°C, crit = +105.0°C) ================================================================================ last latency data is as follows:. 7204 secs base min: -4.700 uS max: 4.600 uS sdev: 0.100 uS 7204 secs servo min: -3.700 uS max: 3.800 uS sdev: 0.100 uS
clifo
very low values, especially in regard with the smal cache! This one should run very well for software stepping (par port). Can you shed a light on the mobo used and the ram size you have installed? I also wonder why ny script is not able to detect the rams. Could you run some commands for me (I'll post a question over in the scripts thread , once I can resume working on it {might be late January only I am afraid})?
Your "GRUB_CMDLINE_LINUX_DEFAULT" uses some parameters, that I did not consider so far. So you got me inspired, thanks a ton!
Tia
chris
Please Log in to join the conversation.
faeluke wrote: Additional Data for HP Compaq Pro 6300 SFF all data taken with Linux Mint V20
I am running servos, so I am not concerned about the PC... Data uploaded for information purposes.
su luke -c "latency-histogram --nobase --base 25000 --sbinsize 1000 --nox 2>&1 | tee -a 1609028912.lat &"
Does the script report the actual line run? Is there an issue with the --nobase switch appearing in the command line?
Common to All
================================================================================
General info:
Hewlett-Packard, HP Compaq Pro 6300 SFF
model name : Intel(R) Core(TM) i7-3770 CPU @ 3.40GHz
Kernel 4.9.0-13-rt-amd64
Faeluke
My 8300 offers the option to disable the parport. Sinceyou're looking for servo performance only, I'suggest you disable it there, if possible.
greez
chris
Please Log in to join the conversation.
Please Log in to join the conversation.
Below I've added the changes from the above entry with that Video Card/GPU installed -
From Dell Optiplex 9020...
Please Log in to join the conversation.
