Now that the cooler has been modified to fit properly and I’ve confirmed via HWiNFO that extreme temperatures don’t occur – at least under light load, I turned my attention to the firmware:

The Marvell/Aquantia controllers have internal firmware that can be updated. Issues are still being fixed to this day, and it’s highly recommended to flash the latest firmware version, even if that entails certain risks. The update is (only?) possible under Windows, so it makes sense for me to do this now, before Proxmox and thus Debian is running on the mini-PCs.

There are good resources for the Aquantia controllers at station-drivers.com

Update Process

I was a little worried, after all, you can easily render the module unusable if something goes wrong. It’s similar to a BIOS or UEFI update during a power outage.

That’s why I updated in two steps: first to version 1.5.38 using the official Marvell firmware updater and then a small step up to 1.5.45 using the modified updater from station-drivers.com. Both are available for download there, along with a detailed forum thread:

Download v1.5.38 (Official)
Download v1.5.45 (Mod)
Forum thread

The release notes are also interesting; there have been a lot of improvements over the years, thanks to Marvell / Aquantia:

FW version 1.5.45 for Marvell AQC113 Family of Controllers
==============================================

1.5.45
?
-------
* Updates the PCIe Gen3 x4 advertisement information.
* Tunes the PCIe memory timing for 0.7 V power supply.
* Sets hysteresis to 50 mV.
* Provides SRAM ECC protection.
* Restricts the presets to used only P4-P9.


1.5.38
27 February 2024
-------
* Bypassed automatic execution of PCIe Configuration from Bootloader.
* PCIe Serdes Firmware integrated into MAC Firmware image and loaded during cold boot.
* PCIe initialization triggered from MAC Firmware.
* ROMBIN atlantic2 set to Antigua_1_5_xxx_487


1.3.37
31 October 2023
-------
* Signed PXE ROM v3.1.106 with EFI v2.1.13 FW 
* Added fix for Pre-Boot DMA protection


1.3.36
31 October 2023
-------
* Signed PXE ROM v3.1.105 with EFI v2.1.12 FW 
* Added boot time optimization and check for cable detect status - fix for Lenovo


1.3.34
3 February 2023
-------
* Reverted the change of loading the PCIe PHY FW to PCIe PHY SRAM (Change introduced in 1.3.29).
  The PHY FW will be loaded from PCIe PHY ROM as it was before


1.3.28
8 April 2022
------------
* Added Advertize 2.5 and 5G NBase-T speed capabilities
* Added AQC113 device ID 0x34C0 to inventory list


1.3.27
18 March 2022
------------
* OEM specific release
* exluded SerDes FW to 3.5.0
* Added configuration to enable SRAM ECC detection and correction


1.3.26
23 feb 2022
------------
* OEM specific release
* Added CFG TLV configuration changes(need your input)


1.3.25
16 feb 2022
------------
* OEM specific release
*	Fixed the unintended Wake-On-Link-Up behavior on the Mac-OS environment.
* Fixed the issue of untimely wake-up from Wake-On-Timer.


1.3.24
08 feb 2022
------------
* Updated the SerDes FW to 3.5.0
* Limited the PCIe preset mask to p4-p9 for Gen3 and Gen4


1.3.23
15 jan 2022
------------
* Introduced additional provisionable PCIe configuration mode to workaround PCIe instability observed on some platforms
  Relevance- AQC113C


1.3.22
5 jan 2022
------------
* Removed default advertisement of GARP and Unsolicited NA in Sleep Proxy mode. The driver can use the new capability mask bit in the offload-control register to control this feature
* Do not wait for 20sec (linklatchuptimeout) to enable Wakeonlinkup if the link is down at the time of suspend/sleep
* Removed the extra link flap on Wake-on-link. And when the system wakes up, initialize the link at the highest possible speed
  Relevance: fast-boot mode


1.3.21
11 nov 2021
------------
* Updated PCIe VPD data
  Relevance: images with PCIe VPD support
* Added support for drvdiag -s in PCIROM v3.1.4 UEFI driver
  Relevance: UEFI


1.3.20
16 oct 2021
------------
 * OEM specific release with specific provisioning


1.3.19
-----------
* Skipped version


1.3.18
02 july 2021
------------
* Fixed the issue of incorrect packet statistics counters reported by fw
* Removed 1G half duplex speed advertisement from the speed capability configuration
* Fixed the invalid speed advertisement 
  Relevance: AQC115C-B1
* Fixed the wrong led behavior at 2.5G speed 
  Relevance: AQC115C


1.3.17
23 apr 2021
-----------
* Added firmware support to disable NCSI via BDP provisioning
* Fixed longer link downtime during os suspend to resume transition in fast-boot mode


1.3.16
13 apr 2021
-----------
* Fixed the TFTP pxe boot failure due to failure to acknowledge Ock/Oack 
  Relevance: FW images with PCIe option rom support


1.3.15
02 apr 2021
-----------
* Added mDNS support
 
 
1.3.14
-----------
* Skipped version


1.3.13
13 feb 2021
-----------
* PHY firmware updated to v6.01.02 for Bx
* PHY change: only enable autoneg timeout when downshift is enabled if downshift was active
* Disable Reed Solomon feature in PHY firmware
* Fixed AQC113 power consumption is higher when cable is removed vs. Attached


1.3.12
08 feb 2021
-----------
* Revert additional fix for mDNS


1.3.11
05 feb 2021
-----------
* Fixed answers in mDNS query differ between when system is in sleep and when the system is awake - additional fix
* Fixed panic case: extended Lom port not open up after secure request was sent


1.3.10
01 feb 2021
-----------
* Fixed AQC113 failed when configured with 10G/full-duplex and came back up with a hot-plug of cable


1.3.9
28 jan 2021
-----------
* Fixed additional fix - remove questions from response (mDNS)


1.3.8
26 jan 2021
-----------
* Fixed AQC113: link became inactive and still fails with hot-plug of cable
* Fixed additional fix for regression test failure
* Fixed AQC113: answers in mDNS query differ between when system is in sleep and when the system is awake
 
 
1.3.7
20 jan 2021
-----------
* Fixed AQC113 ping failed after stress test when system is in sleep
* Fixed [Lom] ping6 extLom is successful after the system panics without pinging basic Lom


1.3.6
10 jan 2021
-----------
* Fixed AQC113 PTP tx ring appears to stop working

	
1.3.5
6 jan 2021
----------
* Fixed AQC113 A0 : built-in became 10base after cable was plugged in switch
* Fixed [Lom] extended mode enabled before 30 sec window
* Fixed [Lom] extended Lom port should be closed after 300 seconds, not 500+ seconds
* Added provision registers 3660 and 3664 in config
* PHY firmware updated to v6.01.01 for Bx


1.3.4
18 dec 2020
-----------
* Fixed AQC113 extended Lom port should be closed after 300 seconds, not 500+ seconds - "shutdown" case
* Fixed AQC113 extended Lom port kept open after Lom unenrolled
* Fixed AQC113 ping6 ext Lom drops packets when system is shutdown (power on failed)


1.3.3
15 dec 2020
-----------
* Fixed AQC113 (both A0/B0) FW does not respond to mDNS query while sleeping
* Fixed AQC113 did not wake after tcp keepalive expires
* PHY firmware updated to v6.0.15


1.3.2
26 nov 2020
-----------
* Fixed AQC113 (both A0/B0) FW does not respond to mDNS query while sleeping (side-issue with uncast request)
* Fixed thermal shutdown functionality is not provisionable
* Remove support for 10m_HD and 100m_HD


1.3.1
09 nov 2020
-----------
* Added support for B1 devices


1.2.3
27 oct 2020
-----------
* Fixed AQC113 Lom0 doesn't generate interrupts for link change events


1.2.1
14 oct 2020
-----------
* Added support for B0 devices


1.1.27
16 sept 2020
------------
* Fixed MAC firmware hang while pc wakeup from suspend state
* Fixed unexpected wake when thunderbolt ethernet adapter is connected
* Fixed regression: ipv4 ping to nic proxy not functioning


1.1.24
22 july 2020
-------------
* Fixed configuration for GPIO
* Fixed calculation read bytes for smbus transaction of Lom
* Fixed interrupt line for Lom if rx packet available
* Fixed Lom interruption pin set to "asserted" after initializing


1.1.19
26 june 2020
-------------
* No changes


1.1.18
25 june 2020
-------------
* Fixed processing of mDNS resource records
Known issue:
* GPIO pin for reset function should be pulled up instead of pulled down


1.1.12
10 june 2020
-------------
Bug fixes:
* Fixed possible datapath stuck after hot-reset
* Fixed link negotiation issue caused after reset for AQC113C 7x7 devices
* Fixed chip stuck which may be caused by soft reset
* Fixed issue preventing chip reset when reset requested while loading ITI


1.1.5
6 may 2020
-----------
Bug fixes:
* Fixed issue that cause PCIe link drops
* Fixed issue when MAC calculated PHY hash before PHY is ready
Enhancements:
* FW provide packet to driver that woke up the system
New features:
* SNR data were added to driver interface


1.0.32
16 march 2020
-------------
Bug fixes:
* PHY reset stability improvements
* Fixed PHY FW stuck caused by fast reset
* Fixed ipv6 offload issue which may prevent receiving fragmented packets
* Fixed VPD corruption caused by multiple VPD read/write operations
Enhancements:
* Hot reset alarm support added
* MSM settings may be applied without link drop-off


1.0.25
14 february 2020
-------------
Bug fixes:
* FW no longer wake host after first link down/up immediately after hibernation
* Fixed SerDes initialization which cause disappearance from PCIe
* Fixed wake on link up
Changes:
* Disabled half-duplex by default for 10M, 100M, 1G speeds


1.0.22
31 january 2020
-------------
Enhancements:
* Disabled half duplex on 10M, 100M, 1G speeds by default
Changes:
* Disabled run-time power optimizations


1.0.19
24 january 2020
-------------
Enhancements:
* Added "coma" power mode
Bug fixes:
* Fixed disappearing from PCI after hibernating
* Fixed data at mDNS response for ipv4 "A" request
* Fixed missing response for mDNS PTR+TXT request
* Fixed chip licensing, which hadn't got settings from efuse high word


1.0.13
10 january 2020
-------------
Enhancements:
* Reduced minimal timeout for TCPKA from 60 to 1 sec
Bug fixes:
* Fixed link up in sleep proxy mode
* Fixed missed ipv6 probes after TCPKA timeout
* Fixed wake up caused by receiving TCPKA probe with wrong flags
* Fixed data at mDNS reply


1.0.11
30 december 2019
-------------
New features:
* Added chip licensing functionality
* Added thermal monitor interface
* Added cable diagnostics
Bug fixes:
* Fixed flow control configuration. Incoming pause frames cause traffic slow down as intended.
* Fixed forced reset of PHY FW caused by MAC FW
* Fixed incorrect data at probe packet which is generated by TCPKA offload
* Fixed invalid src ipv6 at mDNS reply
* VPD mailbox was removed from FW
* Fixed wake up counter
* Fixed resending probes by TCPKA after receiving correct ack on first probe


1.0.3
20 december 2019
-------------
Initial release

FW version 1.5.45 for Marvell AQC113 Family of Controllers
==============================================

1.5.45
?
-------
* Updates the PCIe Gen3 x4 advertisement information.
* Tunes the PCIe memory timing for 0.7 V power supply.
* Sets hysteresis to 50 mV.
* Provides SRAM ECC protection.
* Restricts the presets to used only P4-P9.


1.5.38
27 February 2024
-------
* Bypassed automatic execution of PCIe Configuration from Bootloader.
* PCIe Serdes Firmware integrated into MAC Firmware image and loaded during cold boot.
* PCIe initialization triggered from MAC Firmware.
* ROMBIN atlantic2 set to Antigua_1_5_xxx_487


1.3.37
31 October 2023
-------
* Signed PXE ROM v3.1.106 with EFI v2.1.13 FW 
* Added fix for Pre-Boot DMA protection


1.3.36
31 October 2023
-------
* Signed PXE ROM v3.1.105 with EFI v2.1.12 FW 
* Added boot time optimization and check for cable detect status - fix for Lenovo


1.3.34
3 February 2023
-------
* Reverted the change of loading the PCIe PHY FW to PCIe PHY SRAM (Change introduced in 1.3.29).
  The PHY FW will be loaded from PCIe PHY ROM as it was before


1.3.28
8 April 2022
------------
* Added Advertize 2.5 and 5G NBase-T speed capabilities
* Added AQC113 device ID 0x34C0 to inventory list


1.3.27
18 March 2022
------------
* OEM specific release
* exluded SerDes FW to 3.5.0
* Added configuration to enable SRAM ECC detection and correction


1.3.26
23 feb 2022
------------
* OEM specific release
* Added CFG TLV configuration changes(need your input)


1.3.25
16 feb 2022
------------
* OEM specific release
*	Fixed the unintended Wake-On-Link-Up behavior on the Mac-OS environment.
* Fixed the issue of untimely wake-up from Wake-On-Timer.


1.3.24
08 feb 2022
------------
* Updated the SerDes FW to 3.5.0
* Limited the PCIe preset mask to p4-p9 for Gen3 and Gen4


1.3.23
15 jan 2022
------------
* Introduced additional provisionable PCIe configuration mode to workaround PCIe instability observed on some platforms
  Relevance- AQC113C


1.3.22
5 jan 2022
------------
* Removed default advertisement of GARP and Unsolicited NA in Sleep Proxy mode. The driver can use the new capability mask bit in the offload-control register to control this feature
* Do not wait for 20sec (linklatchuptimeout) to enable Wakeonlinkup if the link is down at the time of suspend/sleep
* Removed the extra link flap on Wake-on-link. And when the system wakes up, initialize the link at the highest possible speed
  Relevance: fast-boot mode


1.3.21
11 nov 2021
------------
* Updated PCIe VPD data
  Relevance: images with PCIe VPD support
* Added support for drvdiag -s in PCIROM v3.1.4 UEFI driver
  Relevance: UEFI


1.3.20
16 oct 2021
------------
 * OEM specific release with specific provisioning


1.3.19
-----------
* Skipped version


1.3.18
02 july 2021
------------
* Fixed the issue of incorrect packet statistics counters reported by fw
* Removed 1G half duplex speed advertisement from the speed capability configuration
* Fixed the invalid speed advertisement 
  Relevance: AQC115C-B1
* Fixed the wrong led behavior at 2.5G speed 
  Relevance: AQC115C


1.3.17
23 apr 2021
-----------
* Added firmware support to disable NCSI via BDP provisioning
* Fixed longer link downtime during os suspend to resume transition in fast-boot mode


1.3.16
13 apr 2021
-----------
* Fixed the TFTP pxe boot failure due to failure to acknowledge Ock/Oack 
  Relevance: FW images with PCIe option rom support


1.3.15
02 apr 2021
-----------
* Added mDNS support
 
 
1.3.14
-----------
* Skipped version


1.3.13
13 feb 2021
-----------
* PHY firmware updated to v6.01.02 for Bx
* PHY change: only enable autoneg timeout when downshift is enabled if downshift was active
* Disable Reed Solomon feature in PHY firmware
* Fixed AQC113 power consumption is higher when cable is removed vs. Attached


1.3.12
08 feb 2021
-----------
* Revert additional fix for mDNS


1.3.11
05 feb 2021
-----------
* Fixed answers in mDNS query differ between when system is in sleep and when the system is awake - additional fix
* Fixed panic case: extended Lom port not open up after secure request was sent


1.3.10
01 feb 2021
-----------
* Fixed AQC113 failed when configured with 10G/full-duplex and came back up with a hot-plug of cable


1.3.9
28 jan 2021
-----------
* Fixed additional fix - remove questions from response (mDNS)


1.3.8
26 jan 2021
-----------
* Fixed AQC113: link became inactive and still fails with hot-plug of cable
* Fixed additional fix for regression test failure
* Fixed AQC113: answers in mDNS query differ between when system is in sleep and when the system is awake
 
 
1.3.7
20 jan 2021
-----------
* Fixed AQC113 ping failed after stress test when system is in sleep
* Fixed [Lom] ping6 extLom is successful after the system panics without pinging basic Lom


1.3.6
10 jan 2021
-----------
* Fixed AQC113 PTP tx ring appears to stop working

	
1.3.5
6 jan 2021
----------
* Fixed AQC113 A0 : built-in became 10base after cable was plugged in switch
* Fixed [Lom] extended mode enabled before 30 sec window
* Fixed [Lom] extended Lom port should be closed after 300 seconds, not 500+ seconds
* Added provision registers 3660 and 3664 in config
* PHY firmware updated to v6.01.01 for Bx


1.3.4
18 dec 2020
-----------
* Fixed AQC113 extended Lom port should be closed after 300 seconds, not 500+ seconds - "shutdown" case
* Fixed AQC113 extended Lom port kept open after Lom unenrolled
* Fixed AQC113 ping6 ext Lom drops packets when system is shutdown (power on failed)


1.3.3
15 dec 2020
-----------
* Fixed AQC113 (both A0/B0) FW does not respond to mDNS query while sleeping
* Fixed AQC113 did not wake after tcp keepalive expires
* PHY firmware updated to v6.0.15


1.3.2
26 nov 2020
-----------
* Fixed AQC113 (both A0/B0) FW does not respond to mDNS query while sleeping (side-issue with uncast request)
* Fixed thermal shutdown functionality is not provisionable
* Remove support for 10m_HD and 100m_HD


1.3.1
09 nov 2020
-----------
* Added support for B1 devices


1.2.3
27 oct 2020
-----------
* Fixed AQC113 Lom0 doesn't generate interrupts for link change events


1.2.1
14 oct 2020
-----------
* Added support for B0 devices


1.1.27
16 sept 2020
------------
* Fixed MAC firmware hang while pc wakeup from suspend state
* Fixed unexpected wake when thunderbolt ethernet adapter is connected
* Fixed regression: ipv4 ping to nic proxy not functioning


1.1.24
22 july 2020
-------------
* Fixed configuration for GPIO
* Fixed calculation read bytes for smbus transaction of Lom
* Fixed interrupt line for Lom if rx packet available
* Fixed Lom interruption pin set to "asserted" after initializing


1.1.19
26 june 2020
-------------
* No changes


1.1.18
25 june 2020
-------------
* Fixed processing of mDNS resource records
Known issue:
* GPIO pin for reset function should be pulled up instead of pulled down


1.1.12
10 june 2020
-------------
Bug fixes:
* Fixed possible datapath stuck after hot-reset
* Fixed link negotiation issue caused after reset for AQC113C 7x7 devices
* Fixed chip stuck which may be caused by soft reset
* Fixed issue preventing chip reset when reset requested while loading ITI


1.1.5
6 may 2020
-----------
Bug fixes:
* Fixed issue that cause PCIe link drops
* Fixed issue when MAC calculated PHY hash before PHY is ready
Enhancements:
* FW provide packet to driver that woke up the system
New features:
* SNR data were added to driver interface


1.0.32
16 march 2020
-------------
Bug fixes:
* PHY reset stability improvements
* Fixed PHY FW stuck caused by fast reset
* Fixed ipv6 offload issue which may prevent receiving fragmented packets
* Fixed VPD corruption caused by multiple VPD read/write operations
Enhancements:
* Hot reset alarm support added
* MSM settings may be applied without link drop-off


1.0.25
14 february 2020
-------------
Bug fixes:
* FW no longer wake host after first link down/up immediately after hibernation
* Fixed SerDes initialization which cause disappearance from PCIe
* Fixed wake on link up
Changes:
* Disabled half-duplex by default for 10M, 100M, 1G speeds


1.0.22
31 january 2020
-------------
Enhancements:
* Disabled half duplex on 10M, 100M, 1G speeds by default
Changes:
* Disabled run-time power optimizations


1.0.19
24 january 2020
-------------
Enhancements:
* Added "coma" power mode
Bug fixes:
* Fixed disappearing from PCI after hibernating
* Fixed data at mDNS response for ipv4 "A" request
* Fixed missing response for mDNS PTR+TXT request
* Fixed chip licensing, which hadn't got settings from efuse high word


1.0.13
10 january 2020
-------------
Enhancements:
* Reduced minimal timeout for TCPKA from 60 to 1 sec
Bug fixes:
* Fixed link up in sleep proxy mode
* Fixed missed ipv6 probes after TCPKA timeout
* Fixed wake up caused by receiving TCPKA probe with wrong flags
* Fixed data at mDNS reply


1.0.11
30 december 2019
-------------
New features:
* Added chip licensing functionality
* Added thermal monitor interface
* Added cable diagnostics
Bug fixes:
* Fixed flow control configuration. Incoming pause frames cause traffic slow down as intended.
* Fixed forced reset of PHY FW caused by MAC FW
* Fixed incorrect data at probe packet which is generated by TCPKA offload
* Fixed invalid src ipv6 at mDNS reply
* VPD mailbox was removed from FW
* Fixed wake up counter
* Fixed resending probes by TCPKA after receiving correct ack on first probe


1.0.3
20 december 2019
-------------
Initial release

Perform the first update

The requirements should be:

• Secure Boot must be disabled in UEFI
• Everything in Windows Defender must be turned off (including the “Driver Block List”)
• Other antivirus software must also be disabled

To update to version 1.5.38, simply download the archive linked above, extract it and run the “startprg.bat” file it contains as an administrator. Once the process is complete, the command prompt windows that opened will close automatically, and you should restart your PC.

Using “kickstart2.exe” you can check before or after the update to see which firmware is currently running on the controller:

Script modifications

To update to version 1.5.45, you need to make changes to the script provided by station-drivers.com, as the IDs listed there are incorrect:

For the modified updater to work with this HP FlexIO v2 module, the four ID blocks must be adjusted to match those of the module. Actually, only two of them need to be changed. To do this, in the “Check if device matches required IDs” section of the “flashUpdateScript.ps1” script file, enter the following:

    # Configuration 1 and 2
    If (($ids[1] -eq "0x1D6A") -and ($ids[3] -eq "0x1043") -and ($ids[4] -eq "0x8890")) {
        If (($ids[2] -eq "0x04C0") -or ($ids[2] -eq "0x14C0")) {
            $deviceSupported = $true
        }
    }

    # Configuration 1 and 2
    If (($ids[1] -eq "0x1D6A") -and ($ids[3] -eq "0x1043") -and ($ids[4] -eq "0x8890")) {
        If (($ids[2] -eq "0x04C0") -or ($ids[2] -eq "0x14C0")) {
            $deviceSupported = $true
        }
    }

by replacing it with:

    # Configuration 1 and 2
    If (($ids[1] -eq "0x1D6A") -and ($ids[3] -eq "0x103C") -and ($ids[4] -eq "0x88FB")) {
        If (($ids[2] -eq "0x04C0") -or ($ids[2] -eq "0x14C0")) {
            $deviceSupported = $true
        }
    }

    # Configuration 1 and 2
    If (($ids[1] -eq "0x1D6A") -and ($ids[3] -eq "0x103C") -and ($ids[4] -eq "0x88FB")) {
        If (($ids[2] -eq "0x04C0") -or ($ids[2] -eq "0x14C0")) {
            $deviceSupported = $true
        }
    }

It’s best to cross-check the IDs with the hardware IDs in Device Manager; they must match. Then save the script, close it, and run the “startprg.bat” file, which is also included.

Perform the update to version 1.5.45

The first thing I noticed was a message saying that the installed driver (from Windows Update) was too old. Once I updated it by installing the newer v3.1.11 version from the Marvell website, the update went through without any problems:

After restarting the PC, everything is working as it should. That was a nerve-wracking process – all that tinkering beforehand would have been for nothing if the update had gone wrong. Now the controller is ready for the future on Debian / Proxmox VE.

Der Beitrag <h5>Mini-PC Server #7: </h5><h3><b>Firmware Update for the Marvell / Aquantia AQC113 Controller</b></h3> erschien zuerst auf flohs blog.

First and foremost, I had to make sure that the module actually worked, both electrically and in terms of software. I’ve read in several places online that there are indeed models (e.g. the ProDesk series) where the module physically fits but doesn’t work on the UEFI side because the necessary drivers don’t seem to be integrated there. To address this, I temporarily replaced the CPU cooler with an old Intel cooler, with some thermal paste in between. This allows the module to fit:

When offline, the network adapter is recognized as such, and Windows automatically installs the appropriate driver once an Internet connection is established. However, the module should also be recognized and function properly with the driver directly from Marvell; moreover, the driver version from Marvell is always more recent (Marvell Public Drivers -> Windows -> AQC113):

Download the drivers from Marvell

Device Manager

Once everything has been installed correctly, you can check in Device Manager to see if everything has been configured or recognized as expected:

You can also learn a lot about the module using HWiNFO:

The module is connected internally via the FlexIO v2’s proprietary slot using two PCI Express 3.0 lanes, which should easily support the full 10 Gbps (one PCIe 3.0 lane theoretically provides 8 GT/s, so two lanes provide 16 GT/s. This theoretically corresponds to 1.969 GB/s, whereas 10 GbE can transfer a maximum of approximately 1.25 GB/s).

In summary, it appears that everything is functioning properly at the electrical and software levels. It is therefore truly “only” a mechanical issue that prevents the revised, newer module from fitting into the Mini PCs. A completely baffling decision on HP’s part.

Mechanical adjustments

On closer inspection, it should be possible to adapt the mounting point. Ideally, this would be a CNC-machined part that incorporates the factory-installed spring pressure beneath the screw to ensure the cooler continues to apply the necessary contact pressure to the CPU. However, I wanted to resolve the issue practically and quickly, so I built an adapter out of an aluminum angle profile, right on the actual component.

First, the cooler needs to be made mountable at all; to do this, I first took stock of what was possible:

The third bolt connection has to go; unfortunately, there’s no way around it. There isn’t much room, but at least there’s some.

Cooler modification

First, I removed the thermal pads to prevent metal shavings from getting inside them. Next, I unscrewed the tinplate cover of the cooler (PH0) to see what the inside looked like. Then I modified the “corner” to fit:

Now, at least, the cooler and the network module can be placed in the right spot:

Build a mount

I used a leftover piece of an aluminum angle profile and modified it accordingly. I had a suitable screw with the “UNC 6#-32” inch thread – which is (unfortunately) common in the IT sector – in my parts stash. This is the same thread used, for example, on the side mounting holes of 3.5-inch hard drives:

To determine the position of the second hole for mounting the cooler, I temporarily screwed the component into the mini PC. On the underside, where the bracket touches the circuit board, I used PVC electrical tape to provide insulation and protection:

The Torx screw facing the cooler is an M3 stainless steel countersunk screw, designed to apply at least some pressure to the processor via the countersunk head. I felt that securing it using the slotted hole in the bracket and a pan-head screw was too “unreliable” – if it slips, nothing will hold it in place. I also filed down the UNC screw at the original mounting point, since the ill-fitting module directly above the screw is unfortunately also full of components from below:

Now I was able to put everything back together and run some initial tests. I used Arctic MX-4 as the thermal interface material to check whether the modification actually works or if the cooler fits so poorly that I need to build a second, more stable version. I applied a layer of white PVC electrical tape to the filed-down screw head, just as I did on the underside of the bracket; a standard sheet of printer paper fits just barely between the SMD components and the PVC. It works, but it’s still very tight:

Now we need to determine whether the cooling capacity is still sufficient so that we can then update the network controller’s firmware first.

Der Beitrag <h5>Mini-PC Server #6: </h5><h3><b>Can’t be done? No way!</b></h3> erschien zuerst auf flohs blog.

This post from ServeTheHome essentially presents the ideal networking solution for a Proxmox mini-PC host: A suitable HP FlexIO v2 module for 10 Gigabit Ethernet, featuring the highly power-efficient AQC113 controller from Marvell / Aquantia (HP part number 56Q71AA), which also supports the intermediate speeds of 2.5 and 5 GbE!

After initial excitement about the existence of this module, disillusionment set in: it’s expensive. Roughly on par with Synology – around €120–150 for a module. On top of that, after reading the comments under this STH post, it became clear that there is a first and second revision of this module, and the second revision – which is now almost exclusively available – NO LONGER fits into the Mini PCs!

What’s really annoying, though, is that the part number (56Q71AA) hasn’t changed at HP; both revisions have the same number. So it’s impossible to specifically buy the first revision, which would simply fit but apparently had certain issues and was therefore revised, without seeing actual photos of the item.

Too bad.

A 10 GbE port and the built-in 1 GbE Ethernet port would be really great. Especially since our connection to the media files also runs through a 10 GbE switch, which is connected to the DiskStation DS1819+, which has also recently been upgraded to 10 GbE. So this module would be technically perfect for our use case; the hard drives and SSDs in the NAS or mini-PC could operate at full capacity for the first time ever.

So what is the problem with the newer version – what exactly doesn’t fit?
The module covers a mounting point on the CPU cooler in such a way that it can no longer be properly secured or even installed at all. Electrically, it should still work, as the connector and mounting screws are in the same position. So it should be a purely mechanical issue.

Well done, HP. But why?

After taking a closer look, I decided to try and get the whole thing working anyway. 10 GbE is just too good to pass up.

At Jacob Elektronik, I found the module as a B-grade item for just under €100 and took the risk of receiving the “wrong” module. That’s exactly what happened – Revision 2 arrived, and it doesn’t fit. Next, I looked for ways to make the whole thing work together anyway.

Der Beitrag <h5>Mini-PC Server #5: </h5><h3><b>The (almost) impossible 10 GbE NIC</b></h3> erschien zuerst auf flohs blog.

Our largest Synology NAS, a DS1819+, is equipped with a PCIe 3.0 x8 expansion slot (electrically connected with four lanes), which, along with its eight 3.5-inch hard drive bays, was one of the reasons we chose this specific model at the time. We purchased our unit right after its release in mid-2019. That was almost seven years ago.

Since then, the RAM has been upgraded from the factory configuration of a single 4 GB DDR4-2133 module to 2 x 8 GB DDR4-2133* modules for a total of 16 GB of memory and dual-channel operation and the storage configuration has been changed several times. Currently, eight 18 TB drives from Western Digital* are installed, which originally came from external 3.5″ WD Elements* hard drives.

By chance, I’ve finally carried out the long-overdue network upgrade:
The original plan was to use one of the compatible Intel network cards as an alternative to the very overpriced Synology-rebranded cards, since these are inexpensive and widely available in used condition. Typical candidates with 10 Gbps over RJ45 (copper) would be the X540-T2* with two ports or the X540-T1* with just one port, or the significantly better but also more expensive X550-T2* or X550-T1*. Both have the well-known issue of running warm to hot. With the X540, instability and crashes are not uncommon without active cooling, especially since it only supports 10, 100, 1,000, or 10,000 Mbps; 2,500 and 5,000 Mbps are missing. The chip is too old for that (these intermediate speeds were only introduced later).

Given these circumstances and the fact that more than 1 Gbit is currently more of a “nice-to-have” than an absolute necessity for us, I have postponed the network upgrade indefinitely.

Until a few days ago, when I spotted the Synology “E10G18-T1*” network card at the bottom of Mindfactorys “Mindstars” section, a single unit for €99 including shipping (Review). Since I’d also had a €50 Mindfactory gift card sitting around for a while, it was clear: €99 for a network card guaranteed to be compatible without any tinkering or issues isn’t a bad deal; the standard price for this model has hovered around €150 since it hit the market. The fact that I ended up paying €49 for it instead made the deal even more irresistible to me. So off I went and bought the card.

It arrived today, fast as expected in its original packaging:

This card is based on the “AQC107” from Aquantia (a networking equipment manufacturer that has since been acquired by Marvell). Compared to older Intel chips, this is a very energy-efficient network controller (~6 watts). However, an even more energy-efficient successor from the same manufacturer is already available: the “AQC113” (~3.5 watts).

100 / 1,000 / 2,500 / 5,000, and 10,000 Mbit are supported, everything you could possibly want.

Overview and RAM

Fully equipped, the DiskStation 1819+ weighs a few pounds, so it’s a good idea to remove the hard drive bays (which are almost prohibitively expensive right now) along with the drives. I combined the upgrade with a cleaning since NAS devices run continuously, they tend to get dusty quickly, and it was time to do it again:

To access the RAM (in my case, for cleaning), you need to turn the DS1819+ upside down. Then, remove two PH2 Phillips-head screws to open the compartment:

Install the expansion card

To access the PCIe slot, the metal cover must be removed. To do this, remove all the PH2 Phillips-head screws circled in red; then slide the metal cover toward the back to unhook and carefully remove it. The Phillips-head screw circled in blue secures the card/slot bracket:

Now the card can be installed very easily:

Finished?

Yes, that’s all there is to it. Now just put everything back together in reverse order, and there you go:

In the network configuration, the additional RJ45 port is now correctly displayed as “LAN 5.” That is the strength of the expansion options offered by Synology itself: plug and play. As an alternative to the model I installed, there are also other options:

Der Beitrag Synology: Upgrade the DS1819+ to 10 Gbps over copper – E10G18-T1 erschien zuerst auf flohs blog.

This “rack-mounted switch for small businesses” from Linksys (then still a Cisco subsidiary) will initially be used temporarily in our home until the network is fully set up. But even later, 48 ports with a maximum of 1 GbE* each will still be useful for a large number of devices. Fortunately, the technical data can still be found at Cisco (archive.org). The user manual is elsewhere also available.

First, the device had to be reset, cleaned, serviced, and reconfigured, as it came from a company liquidation and had been lying around for years (condition unknown). This model series is equipped with “WebView” which allows for convenient configuration via a web browser. Unfortunately, this only works in older browser versions – but at least it works.

Condition

The black plastic front panel is cracked on both sides (this should be easy to repair with superglue*) and the fan(s) have audible bearing damage. After testing all 48 RJ-45 ports, they all seem to work fine with both 100 Mbit and 1 Gbit. I did not test the four SFP ports on the right-hand side due to a lack of modules but these can also “only” handle GbE and share their bandwidth with specific RJ-45 ports as labeled, so they are of little interest at the moment. I have sealed them with 3D-printed PLA caps to protect them from dust (model from Thingiverse by jbanion):

Apart from dirt and scratches, the rest of the device is in relatively good condition; there is no visible corrosion or mechanical damage to the relevant components.

Hardware

Open the device

First, the switch must be opened: this is not as easy as it sounds. The rack mounts are each secured with four PH2 screws on each side and must be unscrewed. Then, using a soft, flat tool such as a guitar pick*, the front panel must be loosened; it is pushed on all around and clipped in place. Next, use a spatula* or a similarly thin but harder tool to gently pry open the gap in the metal housing on one side first; both parts of the housing are held together with metal clips. Work carefully, as the brackets are fragile.

Once the gap has widened, you can work your way across until the entire width is detached. Then do the same on the opposite side, but this is much easier because you can push and press. Finally, pull the upper part of the housing out of the rear plastic cover. It is quite tightly fitted.

Remove the fans

The fans are clamped into special brackets using bent metal: To remove them, it is advisable to use a flathead screwdriver to gently bend up the flat side of the construction (at the edge of the lower part of the housing) and remove the respective fan by tilting it sideways toward the circuit boards.

Unfortunately, all three of the bearings in the installed fans are defective and are not standard components: 40 x 40 mm size with a depth of 20 mm, 5 volts DC with 0.4 watts power consumption, not speed controlled. The OEM manufacturer is Sunon, but the specific model was too expensive for me because it is hardly available and officially only for commercial customers:

As an alternative to the built-in model “KDE0504PKV3”* I used the slightly quieter and significantly cheaper model “MF40200V3-1000U-A99”* also from Sunon. Other fans with these dimensions and specifications are available from Delta Electronics, for example, or from no-name suppliers (not a good idea, as a switch is a 24/7 device!). The replacement I used does not come with a plug, so I simply cut off the plugs from the original fans, soldered them to the new ones, and insulated them (heat shrink tubing*). Here is a comparison of the technical characteristics:

Remove and clean the motherboard

To remove the motherboard, the ribbon cable for console access and the power supply connector (caution: potential residual voltage!) must be removed. Then the three spacers and the remaining PH2 screws must be unscrewed. Now you can remove the circuit board from the lower part of the housing. Here, all components have already been cleaned and are ready for reassembly:

The sticker on the flash chip suggests that this unit was already shipped with the latest firmware version (v1.2.2d) and therefore no update will be necessary.

Configuration?

I connected the SRW2048 via a power cable* to power and to our home network via a network cable* connected to port 1. After several attempts, the switch was not assigned an IP address. I suspected that the integrated HTTP/HTTPS server, which should provide the WebView configuration, was disabled.

This leaves access via the serial console port on the rear panel. This is where the next problem arises: I don’t have the double female DB9 cable (null modem*) that was originally supplied. However, I do have several console cables for newer Cisco switches, which are designed as female DB9 serial to RJ-45. Using an RJ-45 coupler*, I then put together a double female cable:

Next, you need a computer with a serial port (or a serial-to-USB adapter, which I don’t have). The docking stations for the HP EliteBook 8540p have one, so I used mine and installed PuTTY and Tftpd64 (for a possible firmware update that doesn’t exist) on it.

The correct settings are:

Here is a screenshot of PuTTY (only the correct speed needs to be entered):

The initial contact

If the switch is already running and you only establish the connection then, you will initially only see an empty terminal. However, after pressing “Return” you will see that everything is correct and a response will appear:

To access the boot menu and configuration, leave the serial connection as it is and power cycle the switch, i.e., briefly disconnect and reconnect the power. After all LEDs light up briefly as with every boot, you can now watch the firmware POST live and interrupt it with “ESC” to enter the boot menu. This is particularly important if you do not know the access data and want to reset it, or if you are trying to solve other problems:

Once the boot process is complete, you will be greeted with this login screen. By default, the username is “admin” and the password is either blank or also “admin” . In my case, I was able to log in with a blank password field:

Finally, access!

For documentation purposes, I took lots of screenshots of everything, divided into menu categories. Maybe these will help someone in the future figure out what works and what doesn’t. As mentioned before, the latest firmware version v1.2.2d is installed.

System Configuration Menu

System Information

Management Settings

User & Password Settings

This is what it looks like with the default settings.

Security Settings

IP Configuration

Here, “DHCP client” and “HTTP” and/or “HTTPS” must be set to “ENABLE” otherwise no IP address will be assigned for WebView or it will not be available!

File Management

Port Status

You can use the arrow keys to cycle through all available ports.

Port Configuration

Here, too, you can navigate using the arrow keys.

System Mode

Help

Configuration options via WebView

After configuring everything as desired, I restarted the system using the menu option. And lo and behold: the switch now receives an IP address from the router in the network as configured and can also be accessed on this address. However, nothing works in the latest version of Firefox on Windows 11 and the view is very unusable, but in general, “WebView” should now work.

So I booted up the EliteBook with Windows XP and everything works as it should in Internet Explorer 8:

I have also documented the possibilities of WebView below:

Login

Setup

Port Management

VLAN Management

Statistics

ACL

Security

QoS

Spanning Tree

Multicast

SNMP

Admin

LogOut

This should ensure that every configuration option is documented in detail.

Der Beitrag Recommissioning of a 19″ GbE switch: Linksys SRW2048 erschien zuerst auf flohs blog.

We have planned and already purchased two generously sized network cabinets (12SZBIT42U88RAGM from ZPAS), each with 42 U of installation space and a width and depth of 80 cm. One is on the ground floor to distribute the “incoming Internet” sensibly (the plan is to use fiber optic from Deutsche Giganetz, which has already been laid and is active, but cables from Vodafone or classic (V)DSL would also be possible, as well as mobile and satellite Internet, which are available almost everywhere) and to create a cool place for the storage servers and NAS.

This article will focus on this very cramped built-in cabinet. The second cabinet will be located in the DG and will be responsible for computing and most of the switching, in a specially partitioned and separately air-conditioned area with more than enough space in every direction.

⧉ IT-Budget

The problem

In Germany there are a multitude of regulations governing virtually everything in the electrical sector, with dependencies on everything else. In some cases, this leads to complicated situations, such as the one described here:

There must be at least 1.20 meters of unobstructed space in front of the main meter cabinet. The same applies to the house connection box (HAK), which, like the meter cabinet and the network cabinet, is located in a room measuring less than 7 square meters. That sounds cramped and it is: the somewhat unfavorable L-shape of the room further exacerbates the problem.

It is not possible to maintain both distances at the same time; unfortunately, in old buildings, you can’t always choose. Since the meter cabinet needs to be accessed frequently and the HAK virually never needs to be accessed, the priorities are clear – but what if you need to create the distance to the HAK (temporarily)?

What to do?

Fortunately, there is an M12 thread at each corner of the base plate of the cabinets, which is even open on the inside and therefore does not significantly restrict the screw-in depth. The network cabinets come with feet for levelling:

⧉ IT-Budget

If I were to install these simple feet* as originally planned, the cabinet, with its considerable weight once fully equipped, would never be able to be moved again. On the one hand, this is good because the Cat. 7A installation cables* are not designed for movement anyway, but on the other hand, it is bad because it would be impossible to move the whole thing temporarily.

There are a variety of casters available as an alternative, with or without brakes, made from a wide range of materials. The simplest solution would be to install casters with M12 threads* instead of the feet. The disadvantage is that these “stand in the floor,” as PVC will be laid in this room and will certainly give way somewhat over time. At the same time, standing permanently in the same position is probably also not particularly good for the casters’ bearings.

It exists, the optimal solution:

After doing some research, I came across several types of heavy-duty casters that combine both:
a caster and a foot, which is even rubberized. The original manufacturer of this type of combination base appears to be G-DOK with the brand name “FOOT MASTER®” (Product overview as PDF, archive.org), with corresponding patents and prices. Since most things are quickly copied these days, it’s not hard to find a much more affordable and equally suitable clone (note the copied but modified “FOOT MASIER” instead of “FOOT MASTER®”):

Von diesem gibt es dann sogar wie im Original noch eine leicht weiterentwickelte Form als Nachbau, u.a. mit Weldinger-Branding* – für diese Variante habe ich mich letzendlich entschieden: Statt sich wie bei ersten Variante mit dem Drehrad zur Höhenverstellung abmühen zu müssen (das mit mehr Gewicht und mehr Standzeit garantiert wesentlich schwerer zu verstellen sein wird) kann man den integrierten Fuß einfach “hoch- oder runterrätschen”, bei hoher Gewichtsbelastung dann auch z.B. mit einem Schraubendreher als Verlängerung zur Kraftverstärkung:

As with the original, there is even a slightly more advanced replica version available with Weldinger branding*, among other offers – I ultimately decided on this variant: Instead of having to struggle with the rotary wheel for height adjustment (which will definitely be much harder to adjust with more weight and after some time), as with the first version, you can simply “ratchet up or down” the integrated foot, and for heavy loads, you can also use a screwdriver as an extension to increase the force:

As with the previous swivel wheel version, the weight rests permanently on the rubberized foot, which in turn can be raised or lowered internally by means of a thread, ensuring that it stands securely. If movement is necessary, all four feet can simply be “ratcheted down” until the roller stands securely and the foot is far enough off the ground.

This was also copied, but not quite as brazenly as the previous version and is marked with “FOOT CASTOR”.

This solution is sold for workbenches and laboratory equipment, but the feet are also very well suited for my intended use. The maximum load capacity of 1000 kg with four feet is also well dimensioned, which is also the limit of the cabinet. The only thing I didn’t like was the yellow color of the hexagon for countering the M12 thread, so I painted it glossy black to match the rest of the foot:

Der Beitrag Casters or feet on network cabinets – why not both? erschien zuerst auf flohs blog.

We have been operating several Synology NAS devices for many years, which have been discussed previously. One of them in particular – the DS1819+ – has an interesting feature: you can install an expansion card.

In terms of networking, this model comes with four RJ45 sockets, each of which can handle a maximum bandwidth of 1 gigabit. Link aggregation would allow you to use 4 gigabits, but there are obstacles and you would have to occupy four ports on the switch. Not ideal.

But Synology does offer expansion cards:

Anyone who has ever searched for any of these models in price comparisons will notice that they are BRUTALLY overpriced and actually only consist of components from other manufacturers (similar to Synology-branded hard drives and SSDs!).

Third Party?

For a long time, I was unaware that Synology directly tests third-party expansion cards for compatibility and that there is even an official list. It’s no coincidence that this list isn’t that easy to find:

What is compatible or works and what is not is largely determined by the device drivers available in DSM, and that is in the hands of Synology. Since our basic system will be based on RJ45 for the foreseeable future, the two older Intel models, X540-T2* and X550-T2*, look interesting to me.

The X540-T2 is available used starting at €40, while the X550-T2 tends to start at €70 – 80, as it is generally the better choice: Significantly lower power consumption plus support for the now widespread 2.5 and 5 Gbit modes (the X540-T2 can only do 10 / 100 / 1,000 / 10,000, nothing in between!).

Great, so you can use cards that are much cheaper than those from Synology!

Yes, and it gets even better: User “bb-qq” has something really cool up his sleeve in one of his repositories:
DSM drivers for Realtek network adapters that run via “USB 3.1 Gen.1” (i.e., USB 3.0) and are available in China and generally very inexpensive!

The following Realtek chips are supported (as of 01/2026):

• RTL8152/RTL8153:
  1.0 Gbit
   
• RTL8156:
  2.5 Gbit
   
• RTL8157:
  5 Gbit
   
• RTL8159:
  10 Gbit

2.5 or 5 Gbit make the most sense here; 10 Gbit would be severely slowed down by the lack of bandwidth because of USB 3.1 Gen.1. This way, Synology NAS devices in particular can be expanded with faster networking. These devices do not have an expansion card slot but do have at least one USB 3.1 Gen.1 interface, which is almost always the case!

The Github page also lists exact models with compatibilities and incompatibilities so that you can buy the right one right away. However, installing the drivers is something for advanced users; SSH and CLI should not be foreign words.

For the inexperienced, it is recommended to use one of the Intel network cards mentioned above, preferably used or refurbished.

Der Beitrag Alternative networking options for Synology NAS erschien zuerst auf flohs blog.

I have always been a big fan of AVM and its products in the consumer sector. I thought their philosophy was spot on: sophisticated hardware and long-term, high-quality software support. Almost like Synology and Apple. Plus a really well-equipped ecosystem.

Over the last three years, there have been increasing reports of defects in some Fritzboxes, with the Wi-Fi failing but still showing as active according to FRITZ!OS. Almost always the 2.4 GHz band first, often both bands later (5 GHz as well). The former flagship FritzBox 7590, which we have been using since its release, was also affected, shortly before the expiry of the 5-year warranty voluntarily granted by AVM.

It took me a while to figure out why only some Wi-Fi devices were still working. If I didn’t use a separate, hidden SSID for each band, I would probably have spent even longer searching (the default setting sends the same SSID for both bands). In addition, we run everything we can via Ethernet.

AVM replaced our unit some time ago, i thank them for the excellent service. It only took a few days and in the meantime I used my old FritzBox, a venerable 7390. Shortly afterwards, I heard about a similar case among my friends. Same defect, also replaced under warranty.

This topic came up again for me because I have another case to complain about in my family. Similar symptoms – this time, both Wi-Fi bands are gone, while the software says they should be there. This unit is just too old to be replaced by AVM and will soon have to be replaced by a FritzBox 7590 AX V2*.

Possibilities

Günter Born reported on this problem some time ago. His blog post also contains lots of useful tips on repairs and companies that deal with the problem. Heise also picked up on the topic later.

The fault can be repaired but many Fritzboxes will probably end up as electronic waste anyway. This is annoying, especially since 90% of the device continues to function without any problems. Alternatively, the WLAN can be bypassed in tandem with a repeater from AVM*, which is connected to the Fritzbox via a LAN bridge. Not ideal, but practical.

Consequences

Mistakes can happen, but what matters is how they are dealt with. AVM is making every effort to replace affected units. To be honest, I have to say that my confidence in the products has been somewhat shaken. I had never had any problems with AVM products before and the company had long been my standard recommendation in my private sphere.

The fact that Imker Capital Partners is now managing the company’s fortunes does not bode well for the future.

Der Beitrag Hardware problems with Fritzboxes, especially with the very popular 7590 model erschien zuerst auf flohs blog.

The mountains of data accumulated over the years made increasingly extreme storage solutions necessary. The situation could be aptly described as “organically grown,” but neatly organized.

We operate three original NAS devices from Synology (as mentioned earlier in this article) and two “devices that also run a type of DSM”. Everything is local; we do not use Synology services. The sole task is to provide energy-efficient storage space.

Synology NAS

“Synology” NAS

All of these desktop NAS devices will sooner or later be replaced by one or more 19″ rack storage servers. This saves space and is generally more efficient. At this point, at the latest, the network will also be upgraded to 10 Gbit.

Der Beitrag Overview of our network attached storages erschien zuerst auf flohs blog.