Hardware latency tests, used PC's

More
13 Aug 2023 21:22 #277875 by seuchato
Eric
Your emphasis on bios is imho well appreciated.
But there are also the grub default parameters like "isolcpus" etc. to consider.
I cannot prove it like bullet proof: the more cache and (possibly) single vs multichannel ram a machine had, so far it proofed to be better suites for LCNC.

Besides, your statement "been buying Dell mini towers with i5 and the latency has been fine" looks a bit broad. I Guess you could pin down what what models and latency times you got?

greez
chris

Please Log in or Create an account to join the conversation.

More
07 Sep 2023 14:59 #280209 by tommylight

But there are also the grub default parameters like "isolcpus" etc. to consider.

I have never used those, tried it once several years back, it crippled the PC, everything got really sluggish.

Please Log in or Create an account to join the conversation.

More
19 Sep 2023 01:57 #281128 by Hendrixx
   I picked up a budget mini pc at the beginning of this build. I think I paid $129 or close to it and it worked fine for getting the machine set up and in working order but as soon as I started trying to do any real work with multiple tool changes and set ups watchdog.has_bit triggered within 2 minutes and servo-thread warning as soon as the file started to load in. Isolcpus helped a little and I could do small files but not a viable option.   I need this machine to be reliable and consistent so I chose to go with an older workstation and I got a HP Z440 for $185 and am very happy with it.

this is with no stress on the computer
   

This is with 50 glgears going
 

this servo thread
 

and this is how it compares to intels latest and greatest 
 

System:
  Kernel: 6.1.0-12-rt-amd64 arch: x86_64 bits: 64 compiler: gcc v: 12.2.0
    Desktop: Cinnamon v: 5.6.8 Distro: Debian GNU/Linux 12 (bookworm)
Machine:
  Type: Desktop System: Hewlett-Packard product: HP Z440 Workstation v: N/A
    serial: <superuser required>
  Mobo: Hewlett-Packard model: 212B v: 1.01 serial: <superuser required>
    UEFI: Hewlett-Packard v: M60 v02.58 date: 12/07/2021
Battery:
  Device-1: hidpp_battery_0 model: Logitech Wireless Touch Keyboard K400 Plus
    charge: 100% (should be ignored) status: discharging
CPU:
  Info: quad core model: Intel Xeon E5-1630 v4 bits: 64 type: MCP
    smt: disabled arch: Broadwell rev: 1 cache: L1: 256 KiB L2: 1024 KiB
    L3: 10 MiB
  Speed (MHz): avg: 2417 high: 3700 min/max: 1200/3700 cores: 1: 1320
    2: 1198 3: 3452 4: 3700 bogomips: 29530
  Flags: avx avx2 ht lm nx pae sse sse2 sse3 sse4_1 sse4_2 ssse3
Graphics:
  Device-1: AMD Oland [Radeon HD 8570 / R5 430 OEM R7 240/340 Radeon 520 OEM]
    vendor: Dell driver: radeon v: kernel arch: GCN-1 bus-ID: 02:00.0
    temp: 45.0 C
  Display: x11 server: X.Org v: 1.21.1.7 with: Xwayland v: 22.1.9 driver: X:
    loaded: radeon unloaded: fbdev,modesetting,vesa dri: radeonsi gpu: radeon
    resolution: 1920x1080~60Hz
  API: OpenGL v: 4.5 Mesa 22.3.6 renderer: OLAND ( LLVM 15.0.6 DRM 2.50
    6.1.0-12-rt-amd64) direct-render: Yes
Audio:
  Device-1: Intel C610/X99 series HD Audio vendor: Hewlett-Packard
    driver: snd_hda_intel v: kernel bus-ID: 00:1b.0
  Device-2: AMD Oland/Hainan/Cape Verde/Pitcairn HDMI Audio [Radeon HD 7000
    Series] vendor: Dell driver: snd_hda_intel v: kernel bus-ID: 02:00.1
  API: ALSA v: k6.1.0-12-rt-amd64 status: kernel-api
  Server-1: PipeWire v: 0.3.65 status: active
  Server-2: PulseAudio v: 16.1 status: off (using pipewire-pulse)
Network:
  Device-1: Intel Ethernet I218-LM vendor: Hewlett-Packard driver: e1000e
    v: kernel port: 3020 bus-ID: 00:19.0
  IF: eno1 state: up speed: 1000 Mbps duplex: full mac: <filter>
Drives:
  Local Storage: total: 1.94 TiB used: 15.5 GiB (0.8%)
  ID-1: /dev/sda vendor: Toshiba model: MG04ACA200NY size: 1.82 TiB
  ID-2: /dev/sdb vendor: Samsung model: SSD PM871 2.5 7mm 128GB
    size: 119.24 GiB
Partition:
  ID-1: / size: 115.38 GiB used: 15.49 GiB (13.4%) fs: ext4 dev: /dev/sdb2
  ID-2: /boot/efi size: 511 MiB used: 5.8 MiB (1.1%) fs: vfat dev: /dev/sdb1
Swap:
  ID-1: swap-1 type: partition size: 977 MiB used: 0 KiB (0.0%) dev: /dev/sdb3
Sensors:
  System Temperatures: cpu: 38.0 C mobo: N/A gpu: radeon temp: 45.0 C
  Fan Speeds (RPM): N/A
Info:
  Processes: 239 Uptime: 18m Memory: 31.26 GiB used: 2.37 GiB (7.6%)
  Init: systemd target: graphical (5) Compilers: gcc: 12.2.0 Packages: 3051
  Shell: Bash v: 5.2.15 inxi: 3.3.26

 
Attachments:
The following user(s) said Thank You: tommylight

Please Log in or Create an account to join the conversation.

More
24 Sep 2023 23:09 - 24 Sep 2023 23:12 #281565 by EricKeller
The most recent computers I have worked with are Optiplex 7050 and 7010. The supply of used computers is such a moving target I'm not sure that will help anyone. Lots of relatively recent computers are being sold off because they won't run win 11. That's why I was able to buy the Opitplex, the university usually keeps them for more than 4 years. Although they do somethings just get rid of computers when they go off of maintenance.

I usually set an lcnc pc up with isolcpus. Unfortunately I couldn't find any interesting interrupts to change affinity for, that seems like it could lead to a big improvement in latency.

I think the best latency on a computer that set up recently was a dual Xeon workstation with RTAI.  Can't use RTAI with ethernet mesa cards though.  I might set up my pci mesa card in that one, but it's such a large computer I am not sure what machine I would control with it.
 
Last edit: 24 Sep 2023 23:12 by EricKeller.

Please Log in or Create an account to join the conversation.

More
25 Sep 2023 19:50 #281610 by rodw

I usually set an lcnc pc up with isolcpus. Unfortunately I couldn't find any interesting interrupts to change affinity for, that seems like it could lead to a big improvement in latency.
 

This is only useful if you are running a Mesa card or similar ethernet card. Then you can use affinity of the NIC IRQ to be the same as the isolated core.

Please Log in or Create an account to join the conversation.

More
26 Sep 2023 21:25 #281666 by EricKeller
That's what I tried to do, but no joy. I figure you could move userland irqs off of the isolated cpus as well, but I didn't see any good candidates.
Apic is stuck to cpu 0, and that's where my nics both are.

Please Log in or Create an account to join the conversation.

More
26 Sep 2023 22:37 #281673 by rodw
This page may be useful
Find your nic name with ip a (mine ie eno1)
type cat/proc/interrupts and look for the CPU that the NIC is on. In my case here CPU 8
 
Now open linuxcnc and not the core its running on. To do this, open linuxcnc and repeat the test you should see a new thread callrd rtappoapp or similar, Note the CPU the same way, (it should be on an isolated core)
Then use afinity commands to move the  NIC to that CPU

 
Attachments:
The following user(s) said Thank You: JohnnyCNC

Please Log in or Create an account to join the conversation.

More
30 Sep 2023 17:34 - 30 Sep 2023 17:35 #281948 by EricKeller
It's been a while since I tried to change interrupt affinity, but I don't think I can do anything useful in the case of this particular computer. I forget where I got the idea, do you know of anyone that has improved performance by changing interrupt affinity?


Also, I didn't find anything in /proc/interrupts that starts with rtapp. Checked to see if I was running a preempt kernel, and I am.
Last edit: 30 Sep 2023 17:35 by EricKeller.

Please Log in or Create an account to join the conversation.

More
24 May 2024 15:06 #301315 by garja
Hey Tommy... you have posted a lot regarding computers and latency. I found this listing that dates back to 2022. I'm running Debian Bookworm 12.4, Linuxcnc using a Mesa 7i96s card. I'm currently using a Fujitsu Esprimo D757 DT Intel i5 8 GB RAM. The latency is terrible. I started running a graphic intensive vcarve and received latency errors. I think I'd be better off purchasing another computer. Would the Dell 980 you mentioned be a good option or is there a better choice?

Please Log in or Create an account to join the conversation.

More
24 May 2024 15:38 #301319 by tommylight
I had those Fujitsu, 4 versions of them, all worked perfectly latency wise, so
-disable hyperthreading
-disable TPM
-disable virtualization
-disable power savings, all
-disable turbo
Not sure what else, it was a few years back, but try those first.
Fujitsu uses very good thermal compound so if the processor has not been removed, it should be OK.

Please Log in or Create an account to join the conversation.

Time to create page: 0.360 seconds
Powered by Kunena Forum