dreadnought (Main workstation)

Motherboard didn't work out for mirage because AMD dropped support for older CPU in their A520 chipset. Ended up holding on to the board for what became dreadnought.

Started collecting parts with the intention of building a video capture machine. Something with reasonable power and relatively compact. Case size became compromised a bit to fit a full size BD drive I had for use with UHD discs.

Settled on a Ryzen 3 Pro 4350G because it is cheap, it has enough cores, and limited compatibility of the A520 with 2nd/3rd gen iGPU Ryzens. GPU prices were sky high from market conditions, and even finding the WX 2100 I wanted was prohibitive. It could be a bit more powerful machine, but this also frees up the PCIe slot for firewire or internal capture devices.

Build went off without a hitch. The case is a little cramped for cable routing, but I only had to sacrifice the drive area intake airflow path.

Internals - Top View

Internals - Side View

Rear IO

Internals - cabling, drives, pcie

Drives installed

Attempts at running four iSight cameras over firewire ran out of isochronous resources. I can get away with reduced resolution or framerate, but they're already pretty limited so I went for the dual FW643 firewire card from Point Grey (FWB-PCIE-02) that should handle 3 per port.

I had trouble with obs midi segfaulting. It seemed to be in the libremidi library, but I didn't look into the backtrace much. I grabbed their alpha release, and a different repo of libremidi. Aside from following the instructions in build.md under ## Linux; I manually cloned libremidi into src/libremidi, and I changed CMakeLists.txt as below.

@@ -88,6 +88,7 @@ add_library(obs-midi MODULE
        src/forms/settings-dialog.ui)
 include_directories(
        "${LIBOBS_INCLUDE_DIR}/../UI/obs-frontend-api"
+       ${LIBOBS_INCLUDE_DIR}
        ${obs_midi_UI_HEADERS}
        ${Qt5Core_INCLUDES}
        ${Qt5Widgets_INCLUDES}

I also had to modify libremidi's CMakeLists.txt as follows.

@@ -49,6 +49,7 @@ endif()

 target_compile_features(libremidi ${_public} cxx_std_17)

+find_package(Qt5 COMPONENTS Widgets)
 find_package(Threads)
 target_link_libraries(libremidi ${_public} ${CMAKE_THREAD_LIBS_INIT})

I did a fresh install of Debian 11 with manual partitioning, and encrypted root. The installer is frustrating in that if you set up an encrypted partition there is no way to delete it without restarting the installer.

I also installed gnome because I forgot gnome defaults to Wayland which isn't suitable for my use (barrier, possibly local obs screen capture). Gnome is also annoying because despite it's ability to rotate the screen it doesn't handle input device (touchscreen, stylus) rotation so you're left with broken inputs. If I'm going to have to manually configure everything anyway I decided I'm better off in Xfce with fully configurable panels and other things that aren't built to hide every conceivable power user feature away.

Xfce is disappointing after using i3 for a long time. I gave up.

dreadnought has a semi-remote usb hub, and I want it to boot without interaction unless but still be encrypted root. I figured putting a key on a usb stick in the remote hub is effective enough to prevent easy access if the dreadnought is removed from the premises.

Based on https://www.howtoforge.com/tutorial/passwordless-encryption-of-linux-root-partition/ and https://gist.github.com/da-n/4c77d09720f3e5989dd0f6de5fe3cbfb

1. Put random data on the usb stick It's not clear where these tutorials are producing the random data that is used as a key. Probably something they did during install that I skipped. I just overwrote my whole usb stick with /dev/random to start with. It is slow (~1-2 MB/s).

   dd if=/dev/random of=/dev/disk/by-id/usb-General_USB_Flash_Disk_0000000000000000-0\:0 status=progress
   

2. Save a copy of the key

   dd if=/dev/disk/by-id/usb-General_USB_Flash_Disk_0000000000000000-0\:0 of=/root/cryptsetup.key bs=512 skip=4 count=16
   

3. Add the key to a key slot (keep slot 0 for password). You need to enter the existing passphrase to add the new key.

   cryptsetup luksAddKey /dev/nvme0n1p2 /root/cryptsetup.key --key-slot 1
   

4. Add a udev rule to symlink and enumerate usb devices

   Hide

   Listing 5: /etc/udev/rules.d/99-custom-usb.rules

   SUBSYSTEMS=="usb", DRIVERS=="usb",SYMLINK+="usbdevice%n"
   

   udevadm control --reload-rules
   

   Replug the device and check for those symlinks in /dev/usb

5. Add a script to read the key from the usb device

   Hide

   Listing 6: /usr/local/sbin/openluksdevices.sh

   #!/bin/sh
   ############taken from following link#########
   ###http://www.oxygenimpaired.com/debian-lenny-luks-encrypted-root-hidden-usb-keyfile
   
   TRUE=0
   FALSE=1
   
   # flag tracking key-file availability
   OPENED=$FALSE
   
   if [ -b /dev/usbdevice ]; then
   # if device exists then output the keyfile from the usb key
   dd if=/dev/usbdevice bs=512 skip=4 count=16 | cat
   OPENED=$TRUE
   fi
   
   if [ $OPENED -ne $TRUE ]; then
   echo "FAILED to get USB key file ..." >&2
   /lib/cryptsetup/askpass "Try LUKS password: "
   else
   echo "Success loading key file for Root . Moving on." >&2
   fi
   
   sleep 2
   

   Hide

   Make it executable:

   chmod a+x /usr/local/sbin/openluksdevices.sh
   

6. Add the keyscript to crypttab

   Hide

   Get device name

   lsblk|grep crypt
   

   Hide

   Get UUID

   blkid|grep crypto_LUKS|cut -d \" -f 2
   

   Hide

   Listing 7: /etc/crypttab (excerpt)

   nvme0n1p2_crypt UUID=11111111-1111-1111-1111-111111111111 none luks,discard,keyscript=/usr/local/sbin/openluksdevices.sh
   

7. Add cryptroot info to your initramfs config

   Hide

   Listing 8: /etc/initramfs-tools/conf.d/cryptroot

   CRYPTROOT=target=nvme0n1p2_crypt,source=/dev/disk/by-uuid/11111111-1111-1111-1111-111111111111
   

8. Add usb_storage module to initramfs

   Hide

   Listing 9: /etc/initramffs-tools/modules (excerpt)

   usb_storage
   

9. Add a script to add a udev rule to the initrd

   Hide

   Listing 10: /etc/initramfs-tools/hooks/udevusbkey.sh

   #!/bin/sh
   # udev-usbkey script
   ###taken from
   ###http://www.oxygenimpaired.com/ubuntu-with-grub2-luks-encrypted-lvm-root-hidden-usb-keyfile
   PREREQ="udev"
   prereqs()
   {
   echo "$PREREQ"
   }
   
   case $1 in
   prereqs)
   prereqs
   exit 0
   ;;
   esac
   
   . /usr/share/initramfs-tools/hook-functions
   
   # Copy across relevant rules
   
   cp /etc/udev/rules.d/99-custom-usb.rules ${DESTDIR}/lib/udev/rules.d/
   
   exit 0
   

   Hide

   Make it executable:

   chmod a+x /etc/initramfs-tools/hooks/udevusbkey.sh
   

10. Modify grub config (ignore fb/fbcon if you don't want those)

    Hide

    Listing 11: /etc/default/grub (excerpt)

    GRUB_CMDLINE_LINUX_DEFAULT="quiet fb=false fbcon=rotate:1 rootdelay=20 cryptopts=target=nvme0n1p2_crypt,source=/dev/disk/by-uuid/11111111-1111-1111-1111-111111111111,keyscript=/usr/local/sbin/openluksdevices.sh"
    

11. Update initramfs and grub

    update-initramfs -u
    update-grub
    

    Follow the linked guide if you want to unpack the initramfs and verify contents, or just try it. You can also use lsinitramfs for a quick look. What could go wrong?

Dreadnought has replaced chaos as my desktop. This mostly entailed adding a DisplayPort MST hub to allow four displays, and rearranging my desk.

It is a spec boost in single core CPU speed, and GPU speed. It has fewer CPU cores, and half the RAM though.

Some notable things that happened since then:

• My eSATA 4TB drive went bad, and the storage mirror became a single disk that I'm using ZFS snapshots/send to maintain copies of.
• Swapped the WiFi card for a 2.5gbit ethernet adapter (M.2 A+E key format)
• Linux MST crashes which have hopefully been resolved by building a kernel with patch from the thread linked here:Redhat bug

That's all I can recall at the moment. Overall it's doing a good job as a desktop, but the smaller fans can get a bit whiny under load compared to the 120mm fans that just whoosh a bit louder.

Dreadnought got moved into a Full ATX case, the 'Zalman i4'. Before it was in a 'Inwin BP691' ITX case with 'ASRock A520M-ITX​/ac' motherboard. Initially I intended to use a 'Asus B350-F Gaming' motherboard that I had spare, but decided to swap with chaos' 'Asus Prime X470-PRO' since it has better PCI-E expansion.

Cable routing in these segregated bottom cases is a pain. Constant flipping the case back and forth. Poke the cable through, plug it in, manage any excess. The end result looks alright.

In this upgrade I gained:

• 32GB memory (from spare 16GB set I had)
• External ZFS mirror became internal
• 4-port Firewire 800 card (dual controller)
• Extra graphics card (Radeon WX Pro 2100). With the onboard it gives theoretical 9 displays.
• PCIE DisplayPort to USB C (DisplayPort Alternate) adapter
• Extra M.2 slot. I stuck an SSD in there, and might move my 2.5" drive contents to it.
• A bigger heatsink. I had space, so I installed CM Hyper 212 tower heatsink. Before it was Noctua NH-L9a.

I lost:

• DDR4-3600 speed memory (down to 3200), memory went to justice (gaming pc)
• 2.5GbE adapter, I was using the M.2 Wifi slot for this. Substituted a 1G adapter for now.
• Internal Blu-ray drive

Just the motherboard

Board headers close up

Felt tape to prevent drive rattle

Cable management

Everything in place

Expansion cards

Back of expansion cards

Another neat thing for later is the case's vertical GPU mount. I happen to have a PCIE to PCI bridge with two PCI slots that could be mounted there to provide legacy expansion. Couldn't think of a use for now, but I mocked it up a little bit for fun.

PCI bridge mock-up

The Fractal Silent R2 fan I had on the tower heatsink began ticking. It's older and I had it facing downward. I replaced it with a PCCooler F5 R120 which is surprising heavy (190g vs 100g for the previous), and claims to have fluid dynamic bearings.

There is a slight issue when mixing this PWM CPU fan with a DC fan on CPU_OPT on this Asus board. Now the CPU_OPT fan always runs at full speed, so I guess I need to get another PWM controlled fan.

I also modded the other daisy-chained fans to disconnect the tachometer line after the first fan. I thought this would fix the sensor reading jumping up randomly. It probably improved it a bit, but now I notice asus_wmi sensor just randomly jumps up even on ports with just a single fan.

Fan connections for reference:

Since I moved to using two cards I found I could crash X if I enabled a monitor behind a DisplayPort MST hub on the second card.

I was able to verify that it didn't crash if I put the MST hub on the primary card so I've gone with that solution.

Looking back it may have been an issue with autorandr setting the incorrect CRTC, but I will need to test that another time.

This seems to have created a new issue. With my sideways L-shaped monitor config Xorg seems to think my monitors are in a non-contiguous configuration. Now it lets my cursor go into the non-visible areas which is really annoying.

It turns out Xorg fails to properly test for continuous monitors when multiple cards are involved. If the connector between two monitors is a monitor on a different card Xorg will assume there is no continuity, and let your cursor go into the non-visible space.

This machine has been running fine, but I decided to install a serial port on the motherboard header. During the install I:

• Removed the firewire card
• Plugged in the serial port header
• Reinstalled the firewire card
• Attached a null modem cable to a second PC
• Swapped some Displayport cables around

Then the machine refused to post. It also seemed to refuse to reset the BIOS, as normally that would turn the RGB back on. I started pulling cards, unplugging everything, and said lots of curses. No POST, no beeps, nothing.

Finally I got the machine out, pulled the heatsink (because it blocks a memory slot), and removed a pair of RAM sticks. It's back alive. Put the sticks back, and dead. Disable DOCP (memory back at 2133), rearrange the pairs so they're opposite the previous config. Boot and enable DOCP again, and everything works normally. All the devices go back in, and it is working like normal.

I attached the serial cable to the other PC, and it died again. I determined I could at least pull the other pair of RAM sticks to resolve it (saves removing the heatsink). I decided it was some weird voltage or ground loop issue so I made some changes to the setup. They were:

• Swapped my USB switch for another one I had been meaning to try
• Added an internal USB 2.0 hub that gets power from the power supply
• Attached most of my USB 2.0 stuff to that hub

This is way too many hubs, but it seems to have resolved the issue for now. I booted with the serial cable attached to my second PC without issue.

21

I got some new fans (received in exchange for reviews).

First there is this weird triple stacker LTC CF-121D. It's 3 RGB fans, but they only connect together in a line, and have a single cable. I replaced my 3 front fans with them. They PWM reasonably low (800 rpm, I've got them around 980), and are pretty quiet at low speeds. I was going to try the RGB, but it's the ARGB 3-pin 5V variety, and my board only has the 4-pin 12V RGB.

I also swapped the rear stock Zalman fan for a PWM Thermalright TL-P12. Now I can keep it on the CPU OPT header without it running full speed.

The new connection reference:

Things look like this now:

Front fan array

Internal view of front fans

Rear fan

New fan speeds

Not sure the cause, or if I even made an actual fix. My machine was running Debian 12, Linux 6.1.27, and amdgpu on Ryzen 3 PRO 4350G (Vega 6 iGPU). It got rebooted. Accelerated video playback would play a couple seconds then stop. Nothing obvious in the software stack had changed.

I rebooted. No change. I poked around my dpkg logs, nothing obvious.

I tried another machine with Intel GPU, same Debian setup, updated everything, and rebooted. No issues there.

I grabbed the linux-firmware.git repo, and copied amdgpu/renoir files into my firmware directory. Previous firmware files date from March 2024, so those haven't changed, but it is still worth a try. Reboot, and things are working again.

More poking in the dpkg logs, but I am not seeing any relevant packages. Did I just hit a fluke, or did something change? Dunno.

I traded my Ryzen 3 Pro 4350G for a Ryzen 5 5500. More cores, and faster. I put in a Radeon Pro WX 2100 for now, but I have something better on the way.

I was going to put the 4350G in mirage since it can't fit a GPU, and it would be a little upgrade. However, when I went to upgrade stargazer, and realized I needed at least a 3rd gen CPU to work with the A520 chipset board.

The Radeon Pro W5500 I got was used. It feels very solid. Lots of metal. It came with an extension bracket that I did not need. I ended up zip tieing it to the case, and it was useful for managing the PCIe power cable I forgot to install.

Swapping out the temporary Pro WX 2100 for the W5500 almost went flawlessly. I forgot to run the PCIe power cable, and it took a couple tries before I realized why the boot was failing with GPU error beep codes.

Big single slot Radeon Pro W5500

Back of the W5500

Just the PCB back of the W5500

Size comparison Pro W5500 vs Pro WX 2100

Size comparison Pro W5500 vs Pro WX 2100

While I was in the case, I added a baffle to force intake air through the CPU/GPU area. Before it had a direct route from the front fans to the top fans. I carved up a sheet of 1-inch foam board insulation. I taped the cut edges to keep it clean, and used a zip tie to secure it.

GPU installed and airflow baffle in place

Airflow baffle from below

Profile of the top section with baffle

The 4 full size DisplayPorts make my setup a lot simpler. Currently running my little touchscreen USB-C monitor through the SUNIX UPD2018 PCIe card. Then the two Dells get their own ports.

This makes my workstation a step up from my Windows gaming rig. For comparison:

This case (Zalman i4) has an easy to press power button. I often press it when working behind the machine. It is also near the USB ports so I have to be careful there. I got a plastic cover for it.

Button covers

Closed on the power button

Open on the power button

I was running into a few crashes with my quad DIMM 32GB setup. I downgraded to 2 sticks, but still had to play with swapping them to get a set that would pass Memtest86+.

I still wanted 32GB so I decided to go for a pair of 16GB OWC ECC DIMMs. These are 1.2V, DDR4-3200, 1RX8, ECC Unbuffered. They came labeled as part number: OWC3E2D41R816GB. They have Micron chips with FBGA code D8CJV, which corresponds to MT40A2G8SA-062E:F.

New OWC ECC DIMMs

Micron D8CJV chips up close

I ran them on chaos to verify they passed a Memtest before I took my desktop down.

Some Memtesting on chaos computer

While installing them in dreadnought, I took the time to also add an extra drive mount to turn my storage mirror into a 3 drive mirror. These DIMMs don't seem to have XMP/DOCP settings stored on them, but setting everything to auto in the BIOS got the correct values.

HGST drive installed on a 120mm fan mount

Quick Memtest on dreadnought

First boot, and not much has changed in my memory config. Mostly I just see that the total memory width has changed from 64 bits to 72 bits:

 Memory Device
 	Array Handle: 0x0030
 	Error Information Handle: 0x0038
-	Total Width: 64 bits
+	Total Width: 72 bits
 	Data Width: 64 bits
-	Size: 8 GB
+	Size: 16 GB
 	Form Factor: DIMM
 	Set: None
 	Locator: DIMM_A2

Internet places indicate Asus PRIME X470-PRO supports ECC, and so should the Ryzen 5000 series CPUs. I dove into the UEFI configuration, and found a ECC Enable setting which I switched from Auto to Enabled, but it had no effect.

dmesg has no mention of ECC, and edac-util -s reports 'No memory controllers found'.

After some more research I found the source of the problem. Ryzen 5500 is Cezanne based (Zen 3 APU with iGPU disabled), and aside from lacking PCIe 4.0 of normal 5000 series, it also to lacks ECC support. Maybe later I can upgrade, but for now I have some overpriced RAM that at least won't crash my machine.

Back to the drive. I tried adding it to the mirror with zpool attach, but it complained because the drive already had a filesystem. I formatted the drive blank with gnome-disks. Then I attached the new drive (HGST) to an existing drive from the mirror (WD80EMAZ):

sudo zpool attach zstorage /dev/disk/by-id/ata-WDC_WD80EMAZ /dev/disk/by-id/ata-HGST_HUH728080

ZFS went to work resilvering the array, and estimated about 7 hours for the 3.76TB of data.

  pool: zstorage
 state: ONLINE
status: One or more devices is currently being resilvered.  The pool will
        continue to function, possibly in a degraded state.
action: Wait for the resilver to complete.
  scan: resilver in progress since Sun Mar  9 00:30:14 2025
        1.57T / 3.76T scanned at 3.33G/s, 75.4G / 3.76T issued at 160M/s
        75.4G resilvered, 1.96% done, 06:42:03 to go
config:

        NAME                                   STATE     READ WRITE CKSUM
        zstorage                               ONLINE       0     0     0
          mirror-0                             ONLINE       0     0     0
            ata-WDC_WD80EMAZ                   ONLINE       0     0     0
            ata-WDC_WD80EMAZ                   ONLINE       0     0     0
            ata-HGST_HUH728080                 ONLINE       0     0 2.30K  (resilvering)