• Welcome to ASR. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Repurposing an outdated Xeon workstation into a quiet, high efficiency, high speed (10Gbps) TrueNAS server

My NAS is an old Dell T20 server which I recently upgraded to an i7-4790 - this server has 32G RAM, a couple of 4TB disks on an LSI MegaRAID SAS 2108 raid controller.

I only have an Intel 82572EI gigabit card connected to an HP switch.

The speed of this is quite adequate for me - I can't think when I'd need 10G that's for sure.

I'm running Debian along with a couple of VMs (KVM) on this NAS as well. I don't find myself wanting more speed (yet!!).
I ran a Xeon (i5 equivalent) version of the T20 to for many years, in fact it's still in my garage. I bought it in the days when server manufacturers were desperate to get 'techies' to run their kit at home and I remember that the cashback was such that, effectively, I paid for the CPU and everything else was free!

The T20 was used as an ESXi host running Synology DSM (AKA Xpenology) in a VM with the HBA and 4 disks passed through. I managed to fit 6x HDD and 2x 2.5" SSDs in there by hanging Dell 2 drive cage (pinched from a Dell workstation) off a 3.5" to 2.5" adapter mounted in the 'floppy' slot...

IMG_20190308_190345 (Medium).jpg
 
Last edited:
Repurposing an outdated Xeon workstation into a quiet, high efficiency, high speed (10Gbps) TrueNAS server
Part-02: Selection, flashing, and implementation of HBA Card in the TrueNAS NAS server


Hello, dear ASR friends and colleagues,

As you are likely aware, selecting and implementing an HBA card (Host Bus Adapter/multi-HDD controller card) that is robust, stable, reliable, silent (or near-silent), and physically cool-running is a crucial—perhaps the most critical—element of building a NAS server.

After deciding to use TrueNAS SCALE (now Community Edition, Ver. 25.04.0), a free highly reliable NAS OS with a wealth of community support/information, as my NAS operating system, I carefully researched the selection and implementation of such an HBA card. Discussions, suggestions, recommendations, and assistance from Copilot proved invaluable throughout this process.

Consequently, I decided to use the IBM M1050 (an OEM version of the LSI SAS9200-8i) as the HBA for my Xeon-based TrueNAS server; although it is an older-generation model, it has a proven track record within TrueNAS environments. A key reason for this choice is the card's history of widespread use in large-scale enterprise networks. During routine maintenance or system upgrades, fully functional M1050 units are often released in large quantities into the secondary market (specializing in used professional networking equipment), creating frequent opportunities to purchase genuine, guaranteed-working units at low prices.

In this context, it is vital to select a genuine IBM unit—complete with proper identification markings and serial numbers—rather than a knock-off or counterfeit. This is because flashing a genuine IBM M1050 from its default RAID mode to IT (Initiator Target) mode requires the unit to be authentic and possess a valid serial number. TrueNAS relies on its own ZFS RAIDZ implementation (I am using RAIDZ2 with five HDDs), requiring the HBA to operate in IT mode—essentially functioning as a simple hub controller for multiple AHCI HDDs.

Fortunately, in early June, I found a listing for an M1050, which appeared to be a genuine IBM unit, at a Japanese online shop specializing in professional networking gear. Since the website only displayed a photo of the front of the unit, I contacted the shop owner requesting high-resolution images of both the front and back. They promptly sent me the detailed photos, allowing Co-Pilot and me to verify the ID and serial number, and we could confirm that it was indeed a genuine product. Furthermore, the shop informed me that they had tested the unit in both Linux and Windows 11 environments and guaranteed its full functionality, so I purchased it at an affordable price. The unit is shown in Fig. 09 below.
Fig09_20260806b.JPG


In this post, I dare not fully describe the method of flashing the IBM M1050 (an OEM version of the LSI SAS9200-8i) into IT-mode. If you would be seriously interested in “how to”, you would please Google search by keywords “SAS9200-8i IT-mode”. I could successfully flash it to IT-mode using FAT32-formatted 32 GB USB memory in UEFI Shell configuration while having the M1050 card on PCIe16 slot in my ASUS X99-E WS motherboard.

Since this HBA card has two (2) mini-SAS SFF-8087 ports so that max. 8 (eight) SATA HDDs can be connected (I will use five HDDs in a RAIDZ2 configuration), I need two of SFF-8087 to 4x SATA breakout cable; I could purchase, at an affordable price, two (2) 10Gtek MiniSAS SFF-8087 to 4x SATA 80 cm cables at Amazon Japan shown in Fig.10 below.
Fig10_20260806b.JPG


As you are aware, adequate and quiet(!) cooling of the HBA card is crucial for the stable operation of a TrueNAS server. Even though IBM M1050 HBA Card has one seemingly efficient heatsink on it, I decided to add much more efficient cooling on it using dual 120x120x12 mm super-thin quiet fans in push-pull airflow to the entire HBA card.

After my rather intensive web search, I purchased three (3) of such 120x120x12 mm quiet fans shown below in Fig.11; furthermore, I control the two fans using two DC rpm fan controllers for optimal balance of cooling and quietness. One of the three is reserved for renewal/replacement.
Fig11_20260806b.JPG


I also could purchase, at an affordable price, a dual-12cm-fan-stay-on-PCIe-slot kit, made in Japan, as shown below in Fig.12.
Fig12_20260806b.JPG


Then, two of the 120x120x12 mm super-thin quiet fans were nicely assembled in push-pull configuration with the fan-stay as you find below in Fig.13.
Fig13_20260806b.JPG


The ASUS P9X79 WS motherboard I am repurposing for the TrueNAS server features six PCIe16 slots. Fortunately, there is a gap (open space) equivalent to one-PCIe-slot-width between the top slot (PCIe16_1, Xeon CPU direct), where the HBA card is installed, and the second slot (PCIe16_2). Since the large Lian-Li chassis includes mounting points for slot screws in this gap area as well, I was able to securely install dual 120x120x12 mm fans there. This setup allows for efficient, quiet cooling of the HBA card above using a push-pull airflow configuration, as schematically shown below in Fig.14. As mentioned earlier, the fan speeds are controlled by independent DC fan controllers, enabling an optimal balance between cooling performance and quiet operation.
Fig14_WS2297.JPG


I have already detailed the fan setup and airflow for the entire TrueNAS server in Fig.03 of the Part-01 post.
Fig03_20260806b.JPG

The IBM M1050 (LSI SAS9200-8i OEM) HBA, accordingly selected, flashed, and installed, properly recognizes the connected five HDDs (6+4+4+4+6 TB), as shown in Fig. 15 below, and as I shared in my previous post (Part-01), it is delivering the expected performance silently and stably within this Xeon TrueNAS server project.
Fig15_20260806b.JPG


Please be reminded that, as I shared in my Part-01 post, these HDDs for RAIDZ2 storage pool are just for my present test and evaluation objectives, and all the HDDs will be replaced by 8TB WD RED HDD hopefully within the coming year.
 
Last edited:
Pretty impressive; I am surprised those old controllers are still going! I think Broadcom (Avago/LSI) no longer supports them. Flashing FW can be "interesting" on older controllers but not all that hard if you have the SW and I2C interface to do it.
 
Flashing FW can be "interesting" on older controllers but not all that hard if you have the SW and I2C interface to do it.
IIRC those LSI controllers have a linux tool for flashing, no I2C was required. Just a matter of running a command.

I have flashed a few of these, but some ended up unstable, likely issues with the "newer" firmware. Flashing back to the previous version fixed the problem.
 
IIRC those LSI controllers have a linux tool for flashing, no I2C was required. Just a matter of running a command.

I have flashed a few of these, but some ended up unstable, likely issues with the "newer" firmware. Flashing back to the previous version fixed the problem.
Depends on what you are flashing, but I think you are correct for the FW. There are in-band tools for the major OS'. I was thinking of something else, sorry, been a while. I was focused on the analog side so flashing FW often meant grabbing one of the guys who knew how to do it, and we often ran custom FW to perform certain tests.
 
I think there is some over engineering. I would just let Linux LVM/ZFS handle the disks directly.

Next I wouldn't even bother with Jumbo frame because some people will look for a problem to solve. The internet works on 1500. Modern network cards do all the I/O packet processing. The only large MTU we use is in the Data Center and Storage. And not all manufacturers support the same max MTU size.

Been running TrueNAS as a hosted environment for 2 years on top of ProxMox. 22TB of storage on a 2X10GB lag. Max out out ~600MB/s with the 4 spinners. So I can't come close to saturating the network bus.
 
If you fancy jumbo frames, check your LAN switch is happy with them. Then do a test between devices and see if jumbo frames help anything at all - Xfer speed, latency etc when the LAN is concurrently supporting other users. On my LAN I did this and jumbo frames were a waste of time, when transfering say 18GB disk images to NAS, but your mileage might vary - lots of links in a chain. As said, 1Mbps of FLAC files is so trivial as to be almost background chatter on any home LAN.
 
Hello, and thank you, @Jinjuku and @Redacted for your kind comments.

At least in my present Xeon TrueNAS project, the 10Gbps I/O with MTU=9000 (and MTU=9014 in client Windows side) works very stable and efficient even though I intentionally use odd (different size, different brand) mixture of five HDDs (6+6+4+4+4 TB) in ZFS RAIDZ2 configuration for current "test and evaluation" platform, thanks to Xeon E5-2697 v2 (12-core, 24-thread) and the 64GB RAM (still non-ECC ones) plus Cat6A cables and the new silent cool and efficient all-10GbE switching hub on "dedicated" data transfer 10Gbps LAN, which is completely separated from my usual home internet-access 1Gbps LAN; each of NAS server and client PCs has 1GbE NIC plus 10GbE NIC.

Please stay tuned, I will further share some details of each major "component" in the TrueNAS Server, 10GbE switch, and 10GbE NICs in client PCs, as well as present (tentative) backup policy and practice for the ZFS RAIDZ2 SMB share storage in which I have already transferred — quite smoothly in quiet manner — total of ca. 7.6TB small- to large-size various data files including many VHDx disk images of 120GB to 550GB in size.
 
Last edited:
Repurposing an outdated Xeon workstation into a quiet, high efficiency, high speed (10Gbps) TrueNAS server
Part-03: An “intermezzo” of the post series: PCIe adapter for M.2 NVMe SSD and cooling fans on its heat sink


This Part‑03 post serves as a short intermezzo focusing on the PCIe M.2 adapter and its cooling setup.


Hello, dear ASR friends and colleagues,

I assume many of you following this thread are also aware of another discussion hosted by @TonyJZX entitled “is anyone affected by the rampocalypse? gpupocalypse? nandpocalypse?”, where I briefly participated last week in my post #23.

That thread — and my short comment there — explains why I shall, I should, be waiting until the prices would come down to reasonably affordable levels for ECC RAMs (8GB x 8 = 64GB), M.2 NVMe SSD (1TB or 2TB) for RAIDZ2 caching, as well as six (6) new 8TB server HDDs (5 x 8TB for RAIDZ2, one for possible replacement), all to be implemented/installed after my intensive test/validation procedures; maybe "the timing for purchasing" will not come to the reality within a year or two, or even further year(s) ... :facepalm:

Even prior to my purchasing of M.2 NMVe SSD, however, I would like to decide the PCIex16 slot utilization and occupancy for M.2 SSD adapter, and for other PCIe cards, since the repurposing ASUS P9X79 WS motherboard does not have CPU-direct M.2 SSD socket.

Fortunately, I was able to find — at an affordable price — an excellently built PCIe adapter for M.2 NVMe SSDs: the M2.H-PCIE by Kurōtoshikō (Expert Oriented, CFD group), available on Amazon Japan.
Fig16_20260813.JPG


The package of M.2H-PCIE includes fairly large aluminum top heatsink, three kinds of double heat-spreading sheets of different thicknesses, screw driver and screws, and even the low-profile bracket. The overall build quality is solid, with precision exceeding my expectations.

As many of you already know, proper and sufficient cooling for an M.2 NVMe SSD is crucial for stable and fast operation, especially when used as ZFS RAIDZ2 caching device in a TrueNAS server. I was “again” intending, therefore, addition of some kind of fan cooling on the heatsink of the M.2H-PCIE adapter.

For many years, I have known this nicely built, very quiet (almost silent) 40x40x10 mm fan: the AINEX TYPHOON CFZ-4010LB. I decided to use this tiny but almost silent fan in my present “Xeon TrueNAS project”, and I have purchased five of it, and later additional three, to be installed on heatsinks of not only the M.2H-PCIE adapter but also several of the 10Gbps NICs (network interface cards, to be shared in my posts onward) in this Xeon True NAS server project.
WS2148.JPG

After the installation of CFZ-4010LB fans inside the TrueNAS server chassis, I cannot hear them at all from outside even in its full 12 DCV rotation with no fan-controller rpm adjustment.

Using Φ3.0x16mm tapping screw with thin washers and spring washers, I was able to attach two AINEX CFZ-4010LB in a push-pull airflow configuration onto the top heatsink of the M.2H-PCIE (Fig.18).
Fig18_20260813.JPG


Then, thus “prepared” M.2H-PCIE adapter was installed on PCIex16_2 (second) slot which is directly connected to Xeon CPU as shown in Fig.19 below.
Fig19_WS2298.JPG


I have already detailed the fan setup and airflow for the entire TrueNAS server in Fig.03 of the Part-01 post.
Fig03_20260813.JPG

Of course, I will share the "effectiveness" of M.2 NVMe SSD (1TB or 2TB) caching after purchasing and installing it in the "Xeon TrueNAS server project", hopefully within a year when the prices of M.2 NVMe SSDs come down to reasonably affordable levels.

I believe that I do not need the most advanced M.2 NVMe SSD for ZFS RAIDZ2 caching; a previous-generation model should perform perfectly well. I continue carefully watching/following the possible large quantity release of such outdated M.2 NVMe SSD — in perfectly healthy shapes — from enterprise industries upon their regular renewal/renovation for advanced AI(?) compatibilities.


By the way, another reason I am in no rush to purchase an M.2 NVMe SSD for caching is that, even without one, the TrueNAS OS and its ZFS-RAIDZ2 system efficiently and dynamically utilize free space of the 64GB RAM as a cache—complementing the distributed processing capabilities of the Xeon E5-2697 v2’s 12 cores and 24 threads.

Thanks to this, read and write operations for various small- to medium-size data files (such as large volumes of high-resolution raw and JPEG photos) via SMB proceed much faster than I had anticipated, as shown in Fig.20 below.
Fig20_20260813.JPG



Please stay tuned; in my coming Part-04 post, you will find some details of the selection and implementation of two-port 10Gbps server NIC (network interface card) in present “Xeon TrueNAS server project”.
 

Attachments

  • Fig17_20260813.JPG
    Fig17_20260813.JPG
    1.3 MB · Views: 46
Last edited:
I used to run a similar system as a home NAS. As I got older I realized I would rather spend my time on other pursuits so migrated over to a Synology NAS.

The turnkey appliance approach frees me from the ongoing system administration.
 
I fully agree with both points. Just a note - the setup described above uses the wonderful software ZFS raid.

ZFS maintainers tell you specifically not to use RAID controllers. I appreciate the OP's enthusiasm but they are getting a bit over their ski's. But it's about their learning at this point which is great.
 
ZFS maintainers tell you specifically not to use RAID controllers.
Yes, IMO that's why OP talks about the IT (= HBA/JBOD) firmware. I second him, I have great experience with the LSI controllers he uses (non-hw-raid, of course), 8 ports, 16 ports. I used to run a machine with 24 hot-swap SATA drives on two 16-port LSI controllers + ZFS raid for over ten years. Inexpensive, very reliable, great linux support (i.e. TrueNAS).
 
My small server “rack” includes a ASUS NUC i5-1240P and 64GB of RAM (glad I don’t have to buy that now ;) ). It’s running Proxmox with Home Assistant and some LXC containers for file and media, and whatever else I might need at some point.

Fast storage comprises of two internal SSDs (2 and 4 TB, one NVME, one SATA) for OS’es images, ZFS cache and mergerFS cache.

Main storage is two refurbished enterprise Seagate Exos X18 drives in an OWC USB-C enclosure. These drives are special because they are dual actuator disks, and therefore can reach double the speeds normal drives do. They need special formatting though, which puts them in a kind of RAID10 setup. This works quite well, especially for reads, were you get close to 800 MB/sec from only two disks over USB-C. Writing isn’t as fast. For this I use part of the SDDs with MergerFS. New files will be written to the SDD first and get moved to the HDDs every night. Works very well. ZFS caching is mostly pointless for the kinds of workloads I (and probably most people have), both ZIL and L2ARC. MergerFS is much more effective.

Networking is via a Thunderbolt 10 gbps adapter.

All of this is more than fast enough, and has been very stable despite the cables and boxes. Nowadays though, there are some more nicely integrated enclosures that would reduce the clutter. But prices are through the roof right now :rolleyes:

In terms of power efficiency, my system should be pretty good :) probably around 55W all-in.
 
Hello, @Jinjuku, @Eric Larson, @phofman and @voodooless for your thoughts, sharing experiences, and invaluable discussion plus comments.

As you are well aware, I am in "learning" and "test/validation" stage of my "Xeon TrueNAS project"; any of your follow-up comments and information are, will be, much welcome.

Yes, my TrueNAS project is based on software ZFS RAIDZ2 using HBA in IT-mode; not based on hardware HBA RAID system.

By the way, you would please be reminded about the "requirement list" I shared in my very first post Part-01; I am not targeting the most SOTA performance in my present NAS building project. The average practical I/O throughputs I shared in the above post #29i.e. sequential read ca. 750 MB/s, sequential write ca. 650 Mb/s — are already more than enough in my present DIY NAS project.

On the other hand, "almost silent and physically cool operation" is also one of the critical targets in my case, since I like to have all the Xeon TrueNAS server and 10GbE switching hubs in my upstairs office room —actually under my large office table, side-by-side with my client PC-workstations— for easy access and monitor while I sit on my office chair where I do all the business and hobby works including preparation and writing posts for ASR! I have other five (5) client PCs in my office including one notebook PC and one small NUC PC.

For other client PCs in different rooms upstairs and first floor, I will soon have new 10GbE silent NICs to be connected by newly DIY-ed Cat6A wiring from the 10GbE switching hubs under my office desk to establish independent data-transfer-dedicated 10Gbps LAN.

This means that the 10Gbps NICs in the TrueNAS server and in client PCs needs to be "silent and cool", and the 10GbE switching hubs under my offince desk also need to be "silent and cool".

Along with this context, therefore, I will continue my present Part-0x post series sharing my efforts on "balancing" the high (acceptable) I/O throughput and acceptable quietness (less than ca. 25 dBA) in office environment.
 
Hello, @Jinjuku, @Eric Larson, @phofman and @voodooless for your thoughts, sharing experiences, and invaluable discussion plus comments.

As you are well aware, I am in "learning" and "test/validation" stage of my "Xeon TrueNAS project"; any of your follow-up comments and information are, will be, much welcome.

Yes, my TrueNAS project is based on software ZFS RAIDZ2 using HBA in IT-mode; not based on hardware HBA RAID system.

By the way, you would please be reminded about the "requirement list" I shared in my very first post Part-01; I am not targeting the most SOTA performance in my present NAS building project. The average practical I/O throughputs I shared in the above post #29i.e. sequential read ca. 750 MB/s, sequential write ca. 650 Mb/s — are already more than enough in my present DIY NAS project.

On the other hand, "almost silent and physically cool operation" is also one of the critical targets in my case, since I like to have all the Xeon TrueNAS server and 10GbE switching hubs in my upstairs office room —actually under my large office table, side-by-side with my client PC-workstations— for easy access and monitor while I sit on my office chair where I do all the business and hobby works including preparation and writing posts for ASR! I have other five (5) client PCs in my office including one notebook PC and one small NUC PC.

For other client PCs in different rooms upstairs and first floor, I will soon have new 10GbE silent NICs to be connected by newly DIY-ed Cat6A wiring from the 10GbE switching hubs under my office desk to establish independent data-transfer-dedicated 10Gbps LAN.

This means that the 10Gbps NICs in the TrueNAS server and in client PCs needs to be "silent and cool", and the 10GbE switching hubs under my offince desk also need to be "silent and cool".

Along with this context, therefore, I will continue my present Part-0x post series sharing my efforts on "balancing" the high (acceptable) I/O throughput and acceptable quietness (less than ca. 25 dBA) in office environment.

You are preaching to the choir on most of this. My ProxMox is a dual 2690v4, 256GB DDR3 ECC, Solar Flare 2X10Gbe SFP PCIe 8x gen 3 $18, some no name X99 mainboard 6x SATA and 2 NVME M.2 slots with a quality power supply ($220 for those two items).

The CPU's were dirt cheap, RAM was free, NIC was $19, SFP's were pulls, and my 24P copper, 4P SFP+ switch was from work that they didn't want back.
 
Last edited:
Repurposing an outdated Xeon workstation into a quiet, high efficiency, high speed (10Gbps) TrueNAS server
Part-04: Selection and implementation of 10 GbE NIC (Network Interface Card) in the TrueNAS server


Hello, dear ASR friends and colleagues,

Just as with the HBA card implementation I shared in my post #22 Part-02, the selection and implementation of a 10GbE NIC (Network Interface Card) is a key theme at the heart of this “Xeon 10GbE TrueNAS Server Project”. My approach and practices here are nearly identical to those for the HBA.

Specifically, just as with the HBA, for the 10GbE NICs, my selection criteria and requirements are as follows;
  • Select highly reliable models with a proven track record of use in the enterprise network sector over many years,
  • Frequently available at low prices in the used market, where units are released in large quantities due to regular equipment upgrades in the enterprise sector and are guaranteed to be in complete working condition,
  • They have two 10GbE SFP+ ports that can serve as mutual backups,
  • They are fully recognized by the stable kernel drivers in TrueNAS OS 25.04.0,
  • They have a proven track records of use within the TrueNAS Community, and a wealth of related information has been accumulated,
  • Can be connected to a switching hub with SFP+ 10GbE ports using an SFP+ to AOC transceiver,
  • Can be connected to a switching hub with 10GbE RJ45 ports using an SFP+ to RJ45 transceiver,
  • There is practically no difference in I/O throughput when comparing AOC and RJ45 connections,
  • By adding an appropriate quiet fan onto the heatsink, near-silent cool operation (below 50°C of NIC chips and adapters) is possible.
Based on the above “requirement list”, I did rather intensive web search, and again Microsoft Co-Pilot helped a lot for gathering information and giving suggestions for the selection; I decided, consequently, selecting Intel X520-DA2 NIC — having 2-port SFP+ — in my present “Xeon 10GbE TrueNAS Server Project”.

Fortunately, in late June, I found two Intel X520-DA2 cards with a functional warranty listed on a Japanese web shop specializing in pre-owned professional networking equipment. I requested high-resolution photos of both sides of the two NICs from the shop owner, and after careful inspection by myself and Co-Pilot, we were confident that they were genuine Intel products (Fig. 21).
Fig21rev060.JPG


The NICs with above complimentary goods (free!) arrived to my desk enhancing my belief that the NICs have had been working at enterprise sector.

Since I will use the NICs using SFP+ to RJ45 transceiver module and/or SFP+ to AOC transceiver module, I “again” attached the AINEX OMEGA TYPHOON CFZ-4010LB near-silent 40x40x10mm fan onto the heatsink of the NICs using Φ3.0x16mm tapping screw with thin washers and spring washers, just like I did for M.2 NVMe PCIE adapter in my post #29 Part-03.

In this case for X520-DA2 NICs, I attached one AINEX CFZ-4010LB in push airflow configuration so that enabling cooling airflow onto not only the NIC processor heatsink but also to the inserted SFP+ adaptors, as shown in Fig.22 below.
Fig22rev060.JPG


Thus prepared one of the two X520-DA2 NICs was installed onto the PCIEx16 Gen3 slot (slot #4) of the ASUS P9X79 WS M/B as shown in Fig.23 below.
Fig23_WS2299.JPG


I have already detailed the fan setup and airflow for the entire TrueNAS server in Fig.03 of the Part-01 post.
Fig03rev060.JPG

I then decided to compare the 10GbE I/O throughput between an AOC (active optical cable) connection and an RJ45-Cat6A copper cable connection, using the respective transceiver (adapter) modules connected to the X520-DA’s SFP+ ports. The reason for this is that, if there is no significant difference in speed between the two, I plan to set up all 10GbE LAN connections using Cat6A copper cables and a silent switching hubs with all 10GbE RJ45 ports.

To conduct this comparison, I needed a switching hub equipped with both 10GbE RJ45 ports and 10GbE SFP+ ports. Thanks to the generosity of ELECOM Co., Ltd., I had the opportunity, for free, to use a test/evaluation unit of ELECOM EHB-SX2B08F — featuring eight (8) 10GbE RJ45 ports and two (2) 10GbE SFP+ ports (Fig.24 below) — in my office for three weeks starting in early July. This unit has a relatively quiet fan that runs continuously, but since the purpose is a preliminary throughput comparison and evaluation of RJ45 vs. AOC, the faint fan noise is not an issue even if it is audible.
Fig24rev060.JPG


As mentioned above, even though I already have two AOC cables came as complimentary goods with one of the X520-DA2 NICs which can be used in the comparison, I further needed SFP+ to RJ45 transceiver adapter for Cat6A copper cable connection from the NIC to RJ45 port of ELECOM EHB-SX2B08F; having nice recommendation by Co-Pilot, I purchased two (2) unit of SFP-10GM-T-30 — Intel compatible SKU (Product ID) 178041 — (Fig.25 below) at FS.COM web shop in Japan; this unit automatically recognizes and supports 10M/100M/1G/2.5G/10Gbps multi-rate LAN link.
Fig25rev060.JPG


When I connected one end of a 2-meter Cat6A RJ45 cable to the RJ45 port on the EHB-SX2B08F, connected the other end to the SFP-10GM-T-30, and then inserted it into the SFP+ port of the Intel X520-DA2 on the TrueNAS server, TrueNAS OS 25.04.0 perfectly and correctly recognizes the 10GbE connection and the jumbo frame setting with MTU=9000, as shown in Fig. 26 below; for getting detailed information on the NIC recognition, I used the shell commands of TrueNAS OS.
Fig26rev060.JPG


Of course, TrueNAS OS 25.04.0 also perfectly recognizes the 10GbE connection, including MTU=9000, when I connected from X520-DA2 into SFP+ port of EHB-SX2B08F using an AOC cable.

Thanks to the entire airflow in the TrueNAS server and local near-silent airflow given by the small AINEX CFZ-4010LB fan, the temperature of SFP-10GM-T-30 at regular workload is always below 45 degree-C (113 F).


Consequently, I became ready for comparative throughput evaluation from a client Windows 11 Pro PC to TrueNAS server SMB shared storage via AOC connection vs. RJ45 connection in the setup shown in Fig.27 below; in this setup, again, I intentionally used a “stone-age” generation outdated client Windows PC.
Fig27_rev0100.JPG


In the client Windows PC, I used the 10GbE NIC BUFFALO LGY-PCIE-MG3 (Realtek RTL8127 NIC chip) installed on the top PCIE-x16 Gen3 slot with MTU=9014 jumbo frame setting; I will touch on some details of this new client 10GbE NIC, including the latest Windows driver for it, in my coming Part-05 post.

Of course, for the I/O throughput comparison across these four connection paths, the relevant port on the X520-DA2 used for the comparison is set to TCP/IPv4 192.168.7.1/255.255.255.0, while the other port is set to a meaningless address (such as 192.168.99.99/255.255.255.0 which is not used at all on the system).

I believe the four results given by CrystalDiskMark 9.0.3 x64 are self-explanatory for your easy understandings; there is no practically significant difference between AOC connection and RJ45 connection, nor between two SFP+ ports of X520-DA2, quite impressive at least for me.

I could conclude, therefore, that the RJ45-Cat6A connection from the TrueNAS server’s X520-DA2 — either of the two SFP+ ports having SFP-10GM-T-30 transceiver module — can be utilized in my present “Xeon 10GbE TrueNAS project” connecting into silent 8-port and 5-port 10GbE switching hubs; as for such 10GbE silent switching hubs, I will touch on my coming Part-05 post.


Now, I can easily think about your possible simple question of “Where has the second set of Intel X520-DA2 + SFP-10GM-T30 gone?”

Yes, even though I can carefully “store” the second X520-DA2 NIC set — as the backup NIC set for TrueNAS server — in my electrically-active desiccator box together with my cameras and lenses, I could not ignore my curiosity of utilizing it in one of my client PCs, actually the X99-E WS PC-workstation running on Windows 11 PRO 25H2, which I always use for my business and hobby including preparations and writings of ASR posts. If this works as intended, it also means I will be able to reduce the number of 10GbE NICs by one, to be purchased for client PCs.

Since I do not play PC games at all, it was/is quite feasible for me to move the VGA card, actually outdated NVIDEA GEFORCE GTX 970, from PCIe_x16-1 (top slot) to PCIe_x16_3 (third slot), and I installed the Intel X520-DA2 with AINEX CFZ-4010LB fan plus SFP-10GM-T-30 transceiver onto the PCIe_x16_1 (top slot).

The accumulated information on X520-DA2 taken from TrueNAS and other NAS community told me that Windows11 25H2 does not automatically recognize X520-DA2 as 10GbE NIC, but the Intel driver involved in Windows 10 should work nicely even though the “driver details” in “device manager” may show some odd responses. I applied, therefore, went through those procedures.

As sown in Fig.28 bellow, X520-DA2 plus SFP-10GM-T-30 with Cat6A connection to 10GbE switching hub can be properly recognized as 10GbE NIC on Windows 11 PRO 25H2 by using manually installed Intel driver 4.1.254.0; please note that the jumbo packet setting MTU=9014, Max. 8 Queues RSS, as well as 10 GbE Full Duplex speed are properly set and confirmed by the PowerShell command of get-netadapteradvanced property “Ethernet 11”; in this specific case, “Ethernet 11” is the #2 SFP+ port of X520-DA2.
Fig28rev060.JPG


Interestingly, due to some compatibility mismatch of the old driver and Windows 11 Pro 25H2, the #1 SFP+ port of X520-DA2 often may not properly appear (or appear belatedly) in the “device manager” – “network adapter”- “properties” - “details of driver” in Windows 11, even though it works perfectly fine like #2 SFP+ port.

In any way, above Fig.28 contains also the benchmark results by CrystalDiskMark 9.0.2 x64 representing sequential I/O throughput from the Windows 11 PRO 25H2 PC-workstation (X99-E WS M/B, Xeon E5-2630 v3 CPU, 64GB memory) to the SMB share storage of the TrueNAS server, both using X-520-DA2 NIC with SFP-10GM-T-30 module for RJ45-Cat6A connection to 10GbE switching hub. The results are significantly better than my anticipation.


All the results and descriptions shared in this post nicely guided me establishing 10GbE data-transfer-dedicated LAN with Cat6A cables as shown in Fig.29 below in present “Xeon 10GbE TrueNAS server project” utilizing Intel X520-DA2 NICs.
Fig29rev060.JPG



Please stay tuned; my coming Part-05 will focus on client 10GbE NICs and silent 10GbE RJ45 switching hubs.
 
Last edited:
Repurposing an outdated Xeon workstation into a quiet, high efficiency, high speed (10Gbps) TrueNAS server
Part-05: Selection and implementation of 10 GbE NICs (Network Interface Cards) in client Windows PCs, and silent cool-running 10GbE switching hubs


Hello dear ASR friends and colleagues,

Over the past two or three years—and particularly since the start of 2026—the market has seen a succession of new PCIe-based 10GbE NICs (Network Interface Cards) designed for home and SOHO (Small Office/Home Office) use. These cards are characterized by their reasonable pricing, low power consumption, low heat generation, fan-less operation, and high efficiency.

In tandem with this trend, numerous 10GbE switching hubs targeting the home and SOHO markets have also been released this year; these, too, offer reasonable pricing, low power consumption, and low heat generation, featuring either fan-less designs or "semi-fan-less" operation (where the fan does not spin at all under typical home or office conditions).

To put it perhaps a bit boldly, we are clearly entering a phase where 10GbE LANs are becoming practical for home/SOHO use; this shift is one of the key motivations behind my present "Xeon 10GbE TrueNAS server project."


10GbE NIC Selection for Windows Client PCs
Let me begin this post, then, by focusing on the selection and implementation of the 10GbE NICs in Client Windows 11 PCs.

I believe that Realtek’s development of the RTL8127 NIC processor chip—and the subsequent launch of numerous NICs utilizing it—has been a driving force in this sector. The resulting product development competition and falling prices have created a favorable situation for us, the end consumers.

I have been closely following the market developments surrounding RTL8127‑based NICs. Key factors for me include not only price but also driver maturity on Windows 11, the availability of proper warranties and rapid repair/replacement services from domestic vendors, and the accumulation of real-world usage reports.

Furthermore, while many RTL8127-based NICs recommend installation in a PCIe Gen4 (or later) slot to achieve peak performance (10 Gbps), I am particularly interested in the performance limits achievable when using a PCIe Gen3 slot.

The reason for this is that, for my Xeon 10GbE TrueNAS project, achieving sequential read and write speeds between 650 MB/s and 790 MB/s—as measured by CrystalDiskMark benchmarks—when transferring data from a client PC to the TrueNAS SMB share storage would be more than sufficient.

When setting up 10GbE LANs in home or SOHO environments today, it is extremely rare for 10GbE NICs to act as the throughput bottleneck; as you can easily imagine, the actual bottlenecks typically lie elsewhere—such as in client PC HDD I/O or, in my case, the I/O performance of a TrueNAS ZFS-RAIDZ2 SMB share storage spreading in five HDDs.

While a properly configured M.2 NVMe SSD for ZFS caching can dramatically improve I/O for small-to-medium-sized data files—potentially pushing transfer speeds to the limits of the 10GbE connection—my usage pattern differs. Once I have initially stored large volumes of photos, audio files, and documents (Word, Excel, PowerPoint, etc.) on the TrueNAS SMB storage, subsequent periodic additions are not massive in scale (though I remain very interested in the technical benefits of M.2 NVMe SSD caching).

Since the primary use after such initial transfer involves transferring and storing large files—such as VHDx virtual disks and/or PST email archives—the benefits of M.2 NVMe caching are not always strictly necessary.

Taking all these factors into account, I decided to purchase two units of the Buffalo LGY-PCEI-MG3—a product from a domestic Japanese vendor—in early July to evaluate them as part of my ongoing Xeon-based 10GbE TrueNAS server project, as shown in Fig.30 below.
Fig30_rev0100.JPG


As is widely known, the Realtek RTL8127 NIC chip is characterized by high performance as a 10GbE NIC, combined with low power consumption, minimal heat generation, and silent (fan-less) operation. In my case—prioritizing long-term stable operation—I opted once again to install an AINEX CFZ-4010LB (40x40x10mm) near-silent fan onto the heatsink in a "push" airflow configuration, using Φ3.0x16mm tapping screws, washers, and spring washers.

Windows 11 recognizes this NIC without any issues, even when installed in a PCIe x1 Gen3 slot—a slot type commonly found even on older motherboards—despite the manufacturer's recommendation to use PCIe Gen4 or later slots. Of course, it also functions perfectly in PCIe x4 or x16 slots.

While Buffalo provides a slightly older Windows 11 driver for the LGY-PCIE-MG3, I followed recommendations found online and downloaded the latest Windows 11 driver (ver.1127.30.50.508, 2026/05/08) directly from the Realtek website. As shown in Fig. 31 below, the device appears correctly in Windows Device Manager, allowing for advanced settings such as Jumbo Packet MTU=9014 and 10GbE Full Duplex.
Fig31_rev0100.JPG


I was also able to verify these details using PowerShell commands (Fig.32).
Fig32_rev0100.JPG


So far, both the driver and the hardware have proven extremely stable, robust, and entirely trouble‑free.

I have already shared details regarding the I/O throughput performance—which exceeded expectations (660 – 792 MB/s for sequential read/write)—achieved when connecting this "cool-running" 10GbE-NIC-equipped Windows client PC to TrueNAS SMB storage in my previous posts #1 (Part-01), #29 (Part-03) and #38 (Part-04).


Why RTL8127 Is Not Yet Suitable for TrueNAS
By the way, at the initial stage of present “Xeon 10GbE TRueNAS Project”, I seriously considered using the LGY‑PCIE‑MG3 in the P9X79 WS TrueNAS server, but I found the current TrueNAS OS kernel driver library is still not compatible with RTL8127 NIC chip; I mean that the LGY-PCIE-MG3 could not be automatically recognized by TrueNAS OS ver.25.04.0. This is another reason for my selection of Intel X520-DA2 as stable and robust NIC in the TrueNAS server at this timing.

Within the TrueNAS community, many discussions and requests have already argued for early incorporation of proper kernel driver for RTL8127-based NICs in TrueNAS OS; for that, smooth collaboration between TrueNAS free version development team and Realtek technical team will be essential, I believe.

Even at present, there are a few DIY approaches preparing non-official driver for RTL8127 and install it on TrueNAS OS through the so-called “back door” procedures. The driver installed by such path, however, will not be incorporated into TrueNAS OS upon the version-up of the OS.

Consequently, at least for me, I decided not to pursue incorporation of RTL8127-based NICs in TrueNAS server since even the driver will be incorporated in the kernel on coming-soon versions of TrueNAS, the driver would likely require several years to fully mature; for the time being, I will stick to the stable and robust Intel X520-DA2 2-port 10GbE NIC as shared in my post #36 (Part-04).


Silent 10GbE Switching Hubs (LXW‑10G5 / LXW‑10G8)
Now, let us turn to the silent, all‑RJ45 10GbE switching hubs used in my current “Xeon 10GbE TrueNAS project”, which form the backbone of the new 10GbE LAN configuration.

Reasons for selecting the LXW-10G5 included the fact that Buffalo has thoroughly verified its compatibility and interoperability with the company's own LGY-PCIE-MG3 10GbE NIC, and that detailed specifications—including a guarantee of operation—have been published, as shown in Fig. 33.
Fig33_rev0100.JPG

(If you have difficulty in reading the details of the specifications, please simply PM me; I will send you the HiRes image of above Fig.33.)

Although the LXW-10G5 is equipped with a small cooling fan, it spins at full speed for only about 10 seconds upon startup (powering on); thereafter, in typical home or SOHO environments, it does not rotate at all, allowing for completely silent operation. Even under prolonged high-load conditions in my SOHO (where the room temperature is approximately 25°C), the chassis temperature never exceeds 45°C and the fan has never rotated.


A natural question here might be: “At what temperature does the internal fan actually start spinning?”

I personally do not have the courage to perform such extreme heat‑stress (“torture”) tests myself, but I did find an evaluation report online that includes such “brutal torture”; I hope your web browser translates the article correctly into your language. It will likely give you LOL, just as it did me.
https://internet.watch.impress.co.jp/docs/column/shimizu/2020963.html

I did open the chassis to take a look inside, but with my respects for Buffalo Inc. and due to some possible concerns regarding Intellectual Property Rights (IPR), I hesitate to share "my own" photos of the interior here.

You can find the interior photos, however, on this web article.
https://zigsow.jp/item/373690/review/386008

As you can see in the photos from that article, the LXW-10G5 features three fairly large aluminum heatsinks mounted on the PCB via thermal interface pads. Furthermore, the back of the board is connected to a large, ca. 4mm-thick aluminum heat spreader—also via a large thermal pad—ensuring an efficient thermal dissipation design that works from both the top and bottom surfaces. Solid contact between the thermal pads, the PCB, and the aluminum heat spreaders is consistently maintained by several spring-pressure mechanisms; these design and heat-spreading performance are impressive for me.

As shared in my post #36 (Part-04), in July, I had the good fortune of receiving a free three‑week evaluation unit of the Elecom EHB‑SX2B08F — having 8-port 10GbE RJ45 plus 2-port 10GbE SFP+ — designed primarily for enterprises and public facilities, for three weeks. During that period, I tested and evaluated the tandem connection between Buffalo LXW-10G5 and Elecom EHB-SX2B08F using 1m Cat6A cable; as easily anticipated, LXW-10G5 was working perfectly fine in 10GbE operation on such tandem two switching hubs connection.

Buffalo Inc. features 8-port 10GbE version, LXW-10G8, within the same new 10GbE silent switching hub series. The specifications of LXW-10G8 are almost identical to that of LXW-10G5, and hence I do not share the details of LXW-10G8 here in this post; LXW-10G5 is powered by 12 VDC adapter, but LXW-10G8 has internal PS unit which to be connected directly to AC 100V electricity outlet.


Finalized 10GbE LAN Configuration
Based on all the information shared in this post, I have finalized the configuration for the new data-transfer-dedicated 10GbE LAN setup within my ongoing "Xeon 10GbE TrueNAS server project": I will use the Buffalo LXW-10G5 and LXW-10G8 as silent switching hubs, and for all the Windows 11 Pro 25H2 client PCs, I will install the Buffalo LGY-PCIE-MG3 into the PCIe x1 (or x4, x16) Gen3 slot. However, as noted in my post #36 (Part-04), the X99-E WS workstation PC is the exception; like the Xeon TrueNAS server, it will utilize the Intel X520-DA2 as its 10GbE NIC.



Please stay tuned; my coming Part-06 post will be the second intermezzo in the post series focusing very briefly on the thin one-slot-thickness simple VGA card in the Xeon TrueNAS server.
 
Last edited:
Repurposing an outdated Xeon workstation into a quiet, high efficiency, high speed (10Gbps) TrueNAS server
Part-06: An “intermezzo” of the post series: one-slot thickness simple VGA card in the True NAS server, and summary of optimized PCIe slots utilization


This Part‑06 serves as an intermezzo in the series.

It focuses on (1) selecting a quiet single‑slot VGA card for BIOS‑level maintenance, and (2) summarizing the optimized PCIe slot utilization of the P9X79 WS TrueNAS server.


Hello, dear ASR friends and colleagues,

In the “requirement list” for my present “Xeon 10GbE TrueNAS Server Project” shared in the first post #1, I included the following item:

Need for a simple, single‑slot VGA card equipped with a thin, quiet cooling fan, used only for rare occasions such as motherboard BIOS tuning, HBA firmware updates, or fundamental TrueNAS OS changes performed outside its web GUI.

This is because the Xeon E5‑2697 v2 CPU has no iGPU, and the ASUS P9X79 WS motherboard provides no onboard video output.

As many of you may know, it has become increasingly difficult to find a brand‑new, single‑slot VGA card in today’s market.

Most consumer CPUs now include an iGPU, and most consumer motherboards provide HDMI/DP/DVI outputs—eliminating the demand for such standalone, low‑profile VGA cards.

My web search indicated that, among Japan‑domestic vendor products, the only one that completely fits my present requirement is GF-GT710-E1GB/LP —an NVIDIA GEFORCE GT 710 VGA card — by Kurōtoshikō (Expert Oriented, CFD group). Unfortunately, however, Kurōtoshikō has no longer provided this VGA card which means I cannot purchase a brand-new one.

Fortunately, in late May, I found a new‑old‑stock unit—meaning the package had once been opened but the card itself had never been used—at an affordable price on a pre‑owned market for which I quickly placed my purchase order. The unit that arrived was indeed confirmed to be “new‑old‑stock,” as shown in Fig.34 below.
WS2322.JPG


I installed the VGA card onto the PCIe x16_6 (bottom) Gen3 slot of the M/B ASUS P9X79 WS as shown in Fig.35 below; it works more than fine for my present objective, and the near-silent fan cannot be heard at all from outside of the chassis.
WS2321.JPG



Summary of Optimized PCIe Slots Utilization
Please note that the placement of the four PCIe cards in the PCIe slots was not a matter of chance; it was determined with great care to optimize the system as a 10GbE TrueNAS server built on the ASUS P9X79 WS motherboard. Let me summarize, therefore, the PCIe slot utilization as follows:
  • Slot-1 (top), PCIe x16 Gen3 — Server HBA IBM M1050 (LSI SAS9220-B OEM) flashed to IT-mode
    —— one-slot-width spacing —— thin 120x120x12mm dual push-pull quiet fans cooling the HBA
  • Slot-2 , PCIe x16 Gen3 — PCIe-M.2 caching SSD adapter with dual 40x40x10mm push-pull near-silent fans
  • Slot-3 , PCIe x16 Gen2 — Open for airflow
  • Slot-4 , PCIe x16 Gen3 — Intel X520-DA2 dual-port 10GbE NIC with 40x40x10mm near-silent fan
  • Slot-5 , PCIe x16 Gen2 — Open for airflow
  • Slot-6 (bottom), PCIe x16 Gen3 — GeForce GT 710 VGA (Kurōtoshikō GF-GT710-E1GB/LP) with near-silent fan
And, for your convenience of hyperlink jumps;
Post #22 focusing on Server HBA IBM M1050 (LSI SAS9220-B OEM) flashed to IT-mode
Post #29 focusing on PCIe-M.2 caching SSD adapter
Post #38 focusing on Intel X520-DA2 dual-port 10GbE NIC
Post #40 (this post) focusing on GeForce GT 710 VGA (Kurōtoshikō GF-GT710-E1GB/LP)
Post #39 focusing on 10 GbE NICs in client Windows PCs, and silent cool-running 10GbE switching hubs

All the four (4) PCIe cards in the TrueNAS server are installed onto the PCIe x16 Gen3 slots, and the two PCIe x16 Gen2 slots are open for efficient airflow. The entire airflow has been shared several times as shown in Fig.03:
Fig02rev2_revt030.JPG


Please stay tuned; my coming Part-07 post will focus on present status of the TrueNAS ZFS-RAIDZ2 SMB share storage and complete backup platform for it.
 
Last edited:
Back
Top Bottom