Author talks about 4k video, but AI tells me that for "4K production mastering (ProRes 422 HQ, 4:2:2, 10-bit)" the rate is under 2Gbps.
I'm not saying it's not needed, - just curious.
So for instance you have two or more machines that have USB 4 ports at 20 to 80 Gbps and you plug in the appropriate USB C cables to connect them.
What do you need to make the connections appear as a (point to point) part of a regular TCP/IP network? Seems much cheaper than Ethernet if you already have the ports.
His iperf3 results show that it's able to go full beans on a single 25g port. A cross platform SMB file copy is kind of a meaningless benchmark in this context.
IMO a pretty cool product would be a network switch providing native thunderbolt interfaces to multiple upstream hosts. That wouldn't magically fix this RDMA issue, though. This is just a separate thought. Seems like the kind of thing Mikrotik should make.
I found that SMB on macOS is just bad. If I try to heavily load it (e.g. by doing mass conversions in my Audible library), it can cause stalls and disconnections.
I managed to find a set of parameters (in /etc/nsmbd.conf) that makes it work more-or-less reliably by trial-and-error: dir_cache_off=yes, file_ids_off=yes, notify_off=yes, mc_on=no.
It's still terrible, though. It's so bad that Linux in a Parallels VM works _better_.
See also NVMe drives? They draw enough idle power to drain a laptop battery overnight, and will easily hit 60+C in a bad enclosure. That’s only a 10Gbps module. The USB4 enclosures are all heatsink.
Latency was an order of magnitude higher than twisted pair but the throughput was greater than 10Gbps and I didn’t have to spend on 10GbE capable servers and network equipment. I vaguely remember that they had to be proper Thunderbolt not just USB-C but that could have changed it’s been quite a while.
MacOS has supported RDMA over thunderbolt since March 19th.
And your computer already speaks the PCIe and thunderbolt. It has that probably on CPU. Here we have to receive USB, convert it to Ethernet. On much less up to date process nodes than your CPU.
Third, until very recently for even 10Gbit it meant using a usb4<->pcie bridge chip which itself was hot, to provide a generic pcie connection, which then talked to a conventional network chip, so it was a double hop: USB <-> pcie <-> nic. I don't know if 25Gbit to USB direct convert is available yet but we've seen a wave of vastly cheaper 10Gbit adapters come out that are single chip USB<->nic, probably not pcie at all, that are much much cooler. 25Gbit is probably still not here yet, maybe?
With ethernet it's for lots of reasons but a big one is not creating ground loops between distant equipment. If you didn't isolate the network cable would be sinking potentially amps of current between distant devices and that just isn't sustainable. You don't have this problem attaching USB between two devices.
Edit: https://www.jeffgeerling.com/blog/2025/15-tb-vram-on-mac-stu...
Obviously it's possible to move this to a NAS, not use a mac, etc. etc. - but just wanted to chime in with my little use case of moving multi-gig-files around. :-)
I was just looking for a thunderbolt-sfp solution. Super timely!
That'd be the rate for 1 stream at 1x playback. Multiple streams because of overlays/crossfades/etc... along with playback at >1x (such as during rendering in particular) will change that significantly.
But for all certified Thunderbolt 4 and 5 ports, this kind of networking support is a requirement. (It is optional with Thunderbolt 3.)
The rest is up to the operating system.
I believe MacOS supports it via the usual network control panel. I’ve used it for transferring data from an older laptop to a newer one. The cable is more expensive than fiber, last time I checked.
I'm in a never-ending pursuit of cutting time when transferring large files though — AI models is one thing, but I have 100-200 GB projects I transfer across for backup or import (using an external TB5 drive capable of 3-5 GB/sec), and if I can do that in 1 or 2 minutes per project instead of 3 or 4, that's a noticeable speedup.
Definitely overkill for 98% of my workflow though. 10 Gbps is fine.
I do wonder what Apple's custom-built servers are running for networking. I wish they still made Mac Pro-level chassis specifically for better IO. Thunderbolt 5 is nice, but far inferior to a PCIe x16 slot (which can hit 400 Gbps networking) nowadays.
The perf is amazing

I've been using the built-in 10 Gigabit Ethernet on my Mac Studio for a few years. It works fine: I can edit 4K video straight off my NAS over the network, and run backups at around 1 GB/sec.
But... I want more. I upgraded my rack and my NAS to 25 GbE a couple years ago, and wanted to upgrade my main workstation, too.
I looked up 25G networking options for the Mac, and they're all crazy expensive:
The problem is Macs all require Thunderbolt adapters; you can't just plug an inexpensive(ish) PCIe card into a Mac (RIP Mac Pro).
I stopped looking until I saw this blog post. Christian Kohlschütter found a cheap 25G Thunderbolt adapter that uses a server-pulled OCP 2 NIC with a little Thunderbolt 3 adapter board. And it works on any computer with Thunderbolt.

Back in January, it was only $160, which was insta-buy territory for me. Since that time, the Amazon listing jumped to $299, which still might be good compared to the Sonnet... but you might have to go digging through some Chinese sites to find a non-marked-up version now.
This blog post is a companion to the following YouTube video:
Once I had pulled some new fiber to my desk (where I only had Cat6A cabling before), I tested the bandwidth using iperf3 between my Mac and my NAS. That's when I ran into two problems:
The version of iperf3 I was running on the NAS was too old. Without multi-threading, it maxed out at 15 Gbps.
The 25G NIC enclosure was getting hot. Painful to the touch.
I could solve the first problem easily: I compiled the latest version of iperf3. That got me to 20 gigabits, since more than one CPU core could hand the transfers on my NAS. As Christian mentioned in his blog post, 20 Gpbs single direction and 25 Gbps bidirectional is about the limit for the Thunderbolt 3 chipset being used (even if you plug into a Thunderbolt 5 port).

But the second problem was more tricky.
The burning-hot enclosure wasn't thermally bonded to the OCP 2 network card, meaning the NIC chips were cooking.
They had tiny heatsinks on them, but OCP 2 NICs are meant to be inside servers with high pressure fans, not in a little passively-cooled enclosure.
It was acting like a little oven.
Christian mentioned he slapped a couple giant heatsinks on the enclosure. That brought the chip down to a temperature that wouldn't cause NIC dropouts, but it was still getting pretty hot.
I wanted to make sure things were stable, and that meant active cooling.
My first idea was to stick on these low-profile heatsinks and set this speed-controlled USB fan in front. I had to remove the enclosure's barely-ventilated front plate to get more airflow inside, but the back plate also created a ton of resistance.
It would still get hot, and the fan was just loud enough to be distracting, even on its lowest setting.
A Prusa rep had reached out around this time asking if I had any use for their new Prusament PLA in Noctua Brown and Beige... and I decided to switch tracks once they offered to send a spool of each. I purchased a Noctua NF-A8 80mm 5V fan, and an NA-FC1 Speed Controller to silence it.
I designed a fan duct for the 25G Thunderbolt NIC enclosure, and printed it in Noctua beige PLA.
Then I printed this airflow-optimized 80mm fan grill, and screwed that on the front of the 80mm fan.
I was able to use the screws from the 25G NIC enclosure (I removed the front plate entirely), and the fan screws and extra cable that came with the Noctua 80mm fan, to secure everything together.
I chose to splice the braided fan extension cable Noctua includes, and taped it down inside the enclosure with kapton tape for some strain relief.
I soldered the cut end of the fan cable into these through-holes on the Thunderbolt-to-OCP adapter PCB to get the needed 5V power (well, 4.8V, but it's close enough):

The fan only used about 0.5W of power, so I don't think it'll cause any brownout conditions on the NIC itself (which uses 4-5W total at idle).
After final assembly, this is what it looks like:

I plugged it in, re-tested with iperf3, and checked the temperature. It was sitting at less than 36°C after 10 minutes, with the fan on low. And this being a Noctua fan, I couldn't hear it at all under the desk.
Like earlier, it maxes out around 20-25 Gbps, because of the slower Thunderbolt 3 connection.

Testing Samba file copies between my NAS and my Mac, I got around 1.4 GB/sec read, and 1 GB/sec write1.
That's only marginally better than the built-in 10G Ethernet. It's an improvement, sure... but was all the work pulling fiber, designing a fan cowling, paying $200 for all the parts, and assembling everything worth it?
Maybe. At least I got this blog post out of it.