back
97 comments
While working for a broadcasting/networking integration gig, we designed, assembled and field-terminated thousands of copper and fibre runs from 1/2 ft to football field lengths.

The amount of time required to terminate a copper cable in the field is seconds, and felt a bit like art. Something about the way it reliably reacted was magical and felt "strong."

Terminating or splicing a fibre cable felt like wrestling a snake covered in melted crayons, and the failure rate was significantly higher across the board. And it wasn't just workmanship, but quality of product, terminating environment, available equipment, misuse by future operators etc.

That said, at a certain point, we as a firm learned that most purchasers would rather the low latency/small footprint of optical/fibre versus copper, maintenece/failure be damned. Though, maybe part of our willingness to push fibre came from knowing that most purchasers would in 1-2 years call us back in to replace the rack terminations with copper :)

There are 2 very different main types of "copper" in this context. Each can break down into more specific subcategories but they have a relatively common general behavior with their primary type.

First there is the BASE-T RJ45 stuff, which it sounds like you might have been working with. At 10G or higher speeds this get relatively power hungry and is not really an advantage over fiber unless you are also delivering PoE or are trying to reuse existing cabling.

This type (DAC) is a special type of pre-made cable assembly which eschews much of the advanced signalling/conversion logic. The upside is the power usage is low (often even lower than fiber) and the cost is dirt cheap. The downside is the lengths are much more limited and it's intended to be preterminated SFP-to-SFP (or the like) cable assemblies instead of modular patching/custom built.

There is indeed a latency difference, usually DAC < fiber < BASE-T, but they are all within a few microseconds (not milliseconds) of each other so you really have to be pressed to care about it (to the point you're looking at specialized low latency switches and paying extra to lay things out in a way which minimizes the number of L2 hops rather than the cost).

> it's intended to be preterminated SFP-to-SFP

I wonder why nobody's making field-terminatable DACs for custom lengths. If you've torn one down, they're not exactly complicated inside.

If they existed, would people buy them?

> The amount of time required to terminate a copper cable in the field is seconds,

This article is about Direct Attach Copper (DAC) cables which are not something you can field terminate. They use twinax copper and have special modules on the end.

You are thinking about standard RJ45 terminated cabling for 1GBASE-T or 10GBASE-T, which is a different.

> That said, at a certain point, we as a firm learned that most purchasers would rather the low latency/small footprint of optical/fibre versus copper, maintenece/failure be damned.

Direct Attach Copper has slightly lower latency than fiber, but the difference is negligible. Both have significantly better latency than 10GBASE-T through twisted pair cabling.

Yes, thanks. It was confusing me for a second because no, no almost one is terminating twinax in the server room.

I had IT company recommend to me a bunch of fiber cables for a cleanup I was doing. They had about 20 or 30 laser modules we would need, and however much fiber.

When they asked why I planned on doing copper, I told him because each run is three fucking feet.

I’m not sure if they just didn’t realize that’s not what fiber is for or they didn’t know that DAC existed.

Couple points:

1. The copper cables discussed in the article are not field terminable. And if they were, they'd be a pain in the ass.

2. Terminating fiber used to be a pain, but is now pretty easy with the right tools, fuser, and someone with basic training. Even cheap fusers do the job with very low failure rates. They now have so-called "knuckle draggers" terminating fiber.

> That said, at a certain point, we as a firm learned that most purchasers would rather the low latency/small footprint of optical/fibre versus copper, maintenece/failure be damned.

Copper bundles get real thick, real fast: I ran an OneFS cluster for many years, and we had >50 nodes, and all the cables (each node dual-connected) ran to two central switches for backend replications. Rat's nest.

I was very happy when Isilon started officially supporting active optical cable (AOC) on the backend. Really helped with airflow and keeping things tidy.

How much of that was terminating vs splicing for the optical fibre? I have not done the actual hands on work of splicing but have specified fibre runs a couple of times and all the guidance I have seen is that for reliability you want to always splice on good-quality, factory-terminated pigtails because field terminated connectors are (so I've read) very unreliable. When possible we've just run pre-terminated cables so we don't have to get somebody in to splice and that's been really reliable.
I'm pretty sure you didn't design, assemble and field-terminate DAC cables. These are not regular RJ45 or DB25 serial cables, which you can assemble in the field.

I have never heard of the possibility to field-assemble DAC cables. Usually that's when you switch to fiber

Were you not into SDI for broadcast stuff?
Pretty sure that's the copper they're talking about.
Many 10GbE DACs can be hacked to run 25GbE, which is something I had to figure out after realizing that it’s not just plug and play as it was with twisted pair cables coming from 1Gb to 2.5Gb…

https://kohlschuetter.github.io/blog/posts/2026/03/22/unlock...

Thanks! That's very useful, I might have to do the same soon.
I'm waiting for 3 DACs and a few other bits to arrive today to move closer to 10G networking at home. Moving house soon and the new place will have 2.5Gbps FTTP (both up and down) so I wanted to be prepared for that. Given my existing broadband is only 500/75Mbps FTTP I was fine with a 1GbE internal network and Wifi-6 meshing. I could have planned to move to 2.5GbE but it may have been a bottleneck at some point, so may as well push straight on to 10G.

I have a USW-Aggregation with 8 SFP+ ports arriving today too. Just have to install Intel X520-DA2 cards in two of my servers (Proxmox host and a general Linux server), and the NAS also has a 10G SFP+ port, and then connect it all up.

Most of it second hand from eBay for half the usual retail price.

Nice work, that agg switch is excelllwnt.

I went with some cheap eBay cards and slotted them into a synology and PC.

They work great and have for years.

https://www.ebay.com/itm/384094168784?_skw=connectx+mellanox...

I really have to wonder what can you use 10G for? I have 500M down from my ISP, and it is faster than I can imagine ever needing, unless I get into data-hoarding 8k movies.
My homelab has a 10G fabric (switched) for NFS, iSCSI, NVMe-OF, etc. and a 25G fabric (a mix of back-to-back and switched) for clustering (Ceph, DRDB, ZFS replication, migrating VMs).

I spun up some iSCSI-backed SQL Server a few months ago and 10G couldn't keep up with the workload, so I dropped in a pair of 100G ConnectX-4 cards with iSER (iSCSI Extensions for RDMA) support for that particular use-case.

Just because your uplink is less than 10G doesn't mean the rest of your network can't be a bit more capable. :)

True, I don't really feel limited by my existing 500Mbps down, but knowing I'll be having 2500Mbps up/down soon means I want to have the infra to handle it.

Basing things on 2.5GbE would certainly have been cheaper but some things don't support it (they either do 1GbE or 10G SFP+) so settling on 10G where possible made more sense to me. My future ISP also has a 5Gbps up/down option, but even I can't justify that right now.

My wife and kid just want their phones/laptops to work, and to be able to stream stuff to watch, they don't care about the underlying speed.

Having a faster network may make some of my work related things run a bit quicker. A few times a day I'll need to pull something big down (either an ISO or a bunch of docker images) and that can take up to 2 minutes with 500Mbps down. Having those take a fifth of that time will make it seem less of a roadblock to doing work. 2 minutes meant I went and got a cup of coffee and often got more distracted, 30 seconds should keep me at my desk and focused on what I was doing. That's not a big enough reason to justify it on its own obviously.

I also want to do offsite backups with/for various family members, so something better than 75Mbps up is going to be a huge boost. Getting 1Gbps+ out will be huge (assuming whatever is at the other end can support that).

I don't do any kind of data hoarding, I think I've got something under 4TB of data that I actually care about, and most of that are family photos/videos.

Deep down it's mostly because I'm a networking geek so it's fun to play with some new kit and make blinkenlights.

It's less "what new thing can you do" and more "what things involve noticeably waiting, how long is the waiting, and what else is impacted". E.g. updating a game on Steam practically takes slightly under half the time for me (1.2 Gbps actual rate) and has absolutely 0 impact to any other traffic in the house. If it was 10x the price to get 10x the bandwidth I wouldn't bother but it was actually about the same as my old cable modem plan.
What if you want to access your NAS at 10g+ speed? You're focusing on WAN when there is also the LAN side.
It just makes everything feel faster. I went from 500m to 2.5g thinking I would immediately go back (I really just wanted the upgrade to XGS-PON to run my own network) and then I couldn't go back. Its very much like using a higher refresh rate or a faster CPU...
I went from one dev machine to two at my desk so I connected them via 25GBe. With about 2.8GBps TCP throughput and RDMA available I don't have to think too much about task placement or cross-traffic. (specific hardware: Mellanox ConnectX 4 LX cards + a DAC cable).
For most people, 500M is probably fine. But once you have a few family members, each streaming 4K movies to their devices, and a parent that needs a video call to work seamlessly, you start to see the benefits.

10G is probably overkill, but it's also future proofing. The way things are going, loading the NYtimes will require 10G just for the advertising alone...

How many PCIe lanes are you allocating?

The card is obviously 16-lane, but it also has two ports; 40Gb total. In a server that’s fine, but if you want 10G in a desktop you’ll have a problem.

I’m probably not telling you anything new. NICs using newer PCI generations are rare as hen’s teeth. It should be possible to do this with four lanes, but isn’t…

Unless you find a 25G dual-port card, in which case the single lane my secondary slots hand out does at least suffice for 10G one way.

PCIe is also a full duplex connection so 2x10G is still just 20G instead of 40G. For PCIe 2.0 an x8 connection should get you full bandwidth on both ports simultaneously while x4 will fall just short for simultaneously usage (but still higher than 1 port). Unless you're really hankering for that full 20G, in which case a 25G NIC is definitely the better pick, that means you can just slot it in an x4 slot off the chipset on even a standard desktop PC.

Funnily enough, if you want a dirt cheap PCIe 3.0 based card the MCX353A-QCBT and MCX354A-QCBT give 1/2 ports of 40G QSFP+. They support QSFP+ to SFP+ adapters, so you can plug a 10G SFP+ into the QSFP+ port, but they don't support 4x10G breakout unfortunately. I ended up using the 2 port variant in both of my NASes - one port is 40G between the 2 for dirt cheap fast backups and the other is adapted to 10G to connect to the rest of the home network.

I've started buying Intel E810s for most purposes, even for 10G links. (SFP28 ports are generally backward-compatible with SFP+ DACs and transceivers.) The ones you can get on eBay for cheap typically run Dell firmware but it's serviceable. An E810-XXVDA2 is Gen4 x8; as long as the host slot can physically accept the card connector you only need Gen4 x1 electrical for a single 10G link or Gen4 x2 for dual 10G or single 25G.
I'm only planning on using one of the SFP+ ports on each of the cards, the dual port cards were just more common and cheaper on eBay.

The specs say they require PCIe v2.1 x8 lane.

My Proxmox server is quite old and has a Gigabyte GA-X79-UP4 mobo and has loads of spare PCI slots. One slot is taken up by a generic graphics card as the Mobo has no on-board graphics. (I think I went for this mobo because of the number of SATA ports, but it was over 10 years ago so not entirely sure.)

My general Linux server is newer and has an ASUS Prime H610M-A D4 mobo. Only two PCI slots (not used at the moment) and so the Intel X540-DA2 will use up the PCIe 4.0 x16 slot leaving just a PCIe 3.0 x1 slot. But that's fine as this machine is just a CPU (i7-13700), 64GB RAM and a 2TB NVMe. Sticking a good graphics card in it for GPU related fun had been on my list for years but I never got around to it, now the prices are just insane so I'll ignore that for now or something second hand falls into my lap.

I just went through the same process over the last few months. I had a USW agg and ran out of ports so now I have the big dog 24-port version. Mainly wanted L3 routing capability but it’s nice having more ports to lagg connections.
The limiting factor for me is that I'm renting so I can't put my own cabling in to the property. And with the new place there's no existing cabling, nor any conduits to run anything in, and chasing things into the walls/etc is going to be prohibited by the landlord or just too expensive if I'm only in this place for a year or two.

The spools of bend insensitive fibre are pretty cheap and very discreet so I'll probably have a couple of those running along skirting boards/etc in order to connect disparate areas of the house. (The ONT is ~15m away from where the majority of the equipment will live, that's the main bit I have to bridge.)

I did similar with the Mikrotik CRS305-1G-4S+IN and some surplus eBay gear. The nice thing is the NAS and my MacBook dock both have 10G and are connected - and it’s noticeable.
> Most of it second hand from eBay for half the usual retail price.

You were scammed. X520 is old enough to drive a car, the shop should pay you to get it off their hands.

I’m a huge fan of fiber optic cables instead of DAC. For one thing, the distance you can run is far greater. For another, they don’t get as hot as metal cables, particularly at high speeds. OM3 and OM4 is actually quite affordable and SFP+ is nice.

Yes on an absolute sense, DAC can be cheaper, but for a homelab or something, it’s not that much different. Also, it’s much cooler to run fiber optic to things (imho).

In a homelab scenario when asking “DAC vs. fiber” the answer is usually, “yes.” :) Basically, it’s a tradeoff as with everything else.

DACs will usually be even (slightly) lower power per port, and slightly lower latency[1] (we’re fighting over microseconds here!), with excellent durability. The tradeoff is for passive DACs you’re limited on range, cost is often higher, and they may need to be encoded for your interfaces. Moreover, the range is very limited.

Fiber (the cable) is immune to electrical noise, can run long distances, advances in wave division multiplexing extends the life of the fiber by changing what’s the fiber connects to. The downside is you pay slightly in latency for media interface changes (the electrical-to-optical conversions), the limits of bend radius of the cable to not break the cable or reduce bandwidth, and the relative complexity of field terminations compared to twisted pair. I’ve 25+ years experience with fiber, and trust me, it’s great.

————

Outside of cost, both crush twisted pair like an ant. The power consumption per port is also far lower. However, this is only going to matter if you focus on limiting power consumption (not for cost, on principle), have very high-bandwidth applications where latency matters (I do!), and/or just want field experience with things other than twisted pair.

I use DAC and fiber for some things as I try to get every scrap of capability out of my hardware. For example, I have VERY low power (silent or near silent) hardware where I can push 5GB (so ~40Gbps) / sec storage. Not just sending it over the wire, but actually committing it to disk without buffering in RAM. So I have the capability of “PCIe 3.0/4.0 x4 NVMe” speeds across the network… from the (mostly silent) storage server, to anything else that can send or ingest the data that fast. Despite the storage server having very little flash (a few TB vs 100TB+ disk). That’s harder to do with twisted pair, or at least the power consumption of the network connectivity itself starts to add up for a few virtualization cluster nodes.

———-

Generally, “DAC in the rack, fiber to out back” is a reasonable approach. Though “fiber-only” works if you want to limit complexity!

[1] Fiber and DAC tend to trade places on latency every generation or so. It’s a very close race, but they crush twisted pair.

When I did server room buildouts for VFX, I used to use DAC for short distance and fibre for anything longer than a rack.

However that was mainly because I always thought fibre was much more fragile. I mean in a sense it is, it really doesn't like being pull through a thatch of other wires. The solution of course is to do nice cabling.

Nowadays I'm not actually sure if DACs are cheaper than fibre now

> This is a bit dependent on the type of DAC, and the vendor. Generally speaking, we are moving into an era as we move above 100GbE speeds where DACs will be limited to 5m runs and shorter runs.

This is generally good enough for intra-rack runs, especially if you put your top-of-rack (ToR) switch in the middle: you basically only have to go about 1m/3' in either direction (up/down).

Worth mentioning that this is from 2021 and there are now faster speeds and more options.

We have OSFP/QSFPDD with 8 lanes and lane speeds of 100Gpbs and 200Gbps for up to 1.6Tbps on a single DAC at the bleeding edge. Although with these signal integrity limits the maximum length to about 1m.

You probably won't see this on a NIC or anything for home use any time soon though.

Surprised it doesn't mention heat - one of the big advantages.

There is also a new generation of "invisible" fiber that is interesting for home use. Basically looks like fishing line so you can run it somewhere that you just don't see unless you're really looking. Fragile tho

Too bad they didn't mention that while the cable is direct attach, the ends each have an EEPROM in them saying what the cable is. This EEPROM can be a source of immense frustration, if it's got the wrong data in it, or if both ends do not agree.
This should be higher up for anyone considering DACs for homelab use. I had a lot of issues due to this when mixing Aruba switches with Intel nics in my rack. Ended up going to fiber and SFPs.
Dacs are awesome until you encounter a full rack with nothing but 2 and 4 meter dacs. They appeal very much to the lazy engineer, and a lazy engineer can turn them into a rats nest.
To be fair, the dedicated lazy engineer can make a rats nest out of any interconnect technology.
You.. but you put a ladder on the side. Rat's nest goes in the ladder, cover goes on top of the ladder. That way cables aren't blocking access to add/remove any U's.

You make sure all of the cabling has ends that can be pulled safely through the rat nests in the ladders (EG especially RJ45 cabling needs competent clip guards) and Bob's your uncle

And they can block heat exhaustion
Dacs are great … they let you avoid having to convert to optical or having to splice your own cable.

Downside the cables are kind of heavy and sometimes folks dont like that.

Main advantage for me was saving the cost of the SPF (surprisingly pricey f^#%rs) on each end.
I picked up one of these just a few days ago for use with some SFP28 ports. They're cheap, fast, and don't use any power.
Is this connection galvanically isolated?
No.