Bringing a 2010 Dual-Opteron Server Back to Life

The question

My home storage server is a small, modern box. It does the job, but it only has 15 GB of memory, and every few weeks I run into that limit. Then I remembered the old server sitting in the corner: a dual-socket AMD machine from 2010 that I was told had 192 GB of RAM. Could a computer that old replace the one I use every day?

I got it from a wholesaler with the thought it could replace my server i have , which i cannot update to a faster more memory machine. it does its job but i want more.

Meet the contenders

• Processor: My current server: Intel Core i3-13100 (2023); The old server: 2 x AMD Opteron 6180 SE (2010)
• Cores: My current server: 4 cores, 8 threads; The old server: 24 cores
• Memory: My current server: 15 GB; The old server: 192 GB installed, 184 GiB visible, with error-correcting (ECC) memory
• Board: My current server: Consumer-class; The old server: ASUS KGPE-D16 server board
• Built-in video engine: My current server: Yes (Intel QuickSync); The old server: No
• Power use: My current server: A few tens of watts (estimate); The old server: Roughly 150-250 W at idle (estimate)

On paper the old server wins on memory and core count and loses on almost everything else. A 15-year-old board has no modern CPU instructions, no video acceleration and a much higher power bill. I made a chart of the comparison before I touched the machine. The rest of this post is about whether that chart survives contact with reality.

Getting in

The first problem was that I had lost the password. The box answered only on its SSH port. It was running a version of Linux from around 2014, and I could not log in with my usual key, so I decided to boot it from something I control and look at the disks myself.

I tried two ways in:

1. A USB stick with an Ubuntu image, which I wrote on my main PC.
2. Booting over the network (PXE) from a small boot server I run on my main storage box, using a tool called iVentoy.

PXE gave me the most trouble, in two separate ways.

First, the network was handing the old server the wrong boot file name. The boot server simply did not have a file by that name. I only found that out by asking the boot server directly for each file it offered, and then correcting the name.

Second, even with the right name, the old server printed PXE-E61: Media test failure, check cable. This board has two network ports, and the network boot step was using the first one, which had no cable plugged in. When I moved over to the second port, it worked straight away.

While I was fighting with that, I also found that the server could boot its own installed system from its hard drives. At the boot menu I saw an Ubuntu 14.04 install, which is the 2016-era system I had been talking to over SSH, and a second, newer Ubuntu next to it. Trying to get a recovery shell on the newer one only produced a loop of failing NVIDIA driver messages, which was not going to get me anywhere.

What I found inside

Opening it up in software told me a lot.

• The board: an ASUS KGPE-D16, a dual-socket server board from around 2010.
• The memory: 12 modules of 16 GB DDR3, which is 192 GB. Linux sees 184 GiB of it, and the memory reports multi-bit ECC with no errors so far.
• The disks: an Adaptec 6805 hardware RAID card. The operating system never sees individual drives. It sees three arrays of about 3.6 TB each, and the boot screen reported one of them as degraded, meaning a disk had failed. That matters, because a card like this hides the raw disks that a file system like ZFS wants to manage itself.
• The data: nothing on the arrays mattered to me. I never got the old password back. Since the arrays held nothing I needed, I put a fresh Ubuntu 26.04 on one of them so I could run proper tests.

I also noticed a problem with cooling. One of the board's fans read 0 RPM, another turned at only a few hundred RPM, and the board temperature sat at about 55 degrees C even at idle. The processors themselves stayed cool, but I would not leave this machine running for months without fixing the fans.

The test results

I ran the same tests on both machines.

• CPU speed, one thread (sysbench events/s): Current server: 1,467; Old server: 242
• CPU speed, all threads (sysbench events/s): Current server: 5,942 (8 threads); Old server: 5,780 (24 threads)
• Memory speed (sysbench MiB/s): Current server: 86,335; Old server: 11,697
• Disk reads (read-only, three RAID volumes): Current server: not tested; Old server: 191 / 307 / 352 MB/s
• Memory error checking: Current server: Not confirmed; Old server: multi-bit ECC, 0 errors
• Processor temperatures: Current server: [___]; Old server: 30-37 C idle, 44-56 C at full load
• Missing CPU features: Current server: none that matter here; Old server: no AES, AVX, AVX2 or SSE4
• Stability: Current server: n/a; Old server: all 24 cores at full load for 2 minutes, no failures

The headline is that 24 old cores add up to about the same total speed as 8 threads on the new chip, 5,780 against 5,942. But any single job runs about six times slower on the old machine, and its memory is about seven times slower.

I did not measure power at the wall, noise, Immich machine learning or video transcoding. The old processors have no AVX or video engine, so I would expect trouble with the last two. The power figures in the table above are estimates from published data, not my own readings.

The verdict

The old server loses. The memory is real, and 192 GB of ECC RAM is a lot of headroom. But the processors are far slower for any one task, the missing instructions rule out some of the software I run, the RAID card hides the disks I would want ZFS to manage, the cooling needs attention, and I expect the power bill to be many times higher. My main server stays the main server.

The old machine is still useful as a lab box, for running many virtual machines at once and for experiments where speed does not matter.

What I would do differently

• When network boot fails, test the boot server directly and check which network port it is using before blaming the network.
• Read the storage screen in an installer before pressing Done. It will happily erase a disk, and it does not know which one matters.
• Only consider machines that have DDR4 or better and made with UEFI boot , DDR3 is too old and too expensive to run now.