author here! it&#x27;s not on top of USB4NET, no (RXE can already do that, it&#x27;s compared in the benchmarks). it&#x27;s built with the same tb primitives as the networking stack obviously, just assembled differently to emulate a verbs device instead of a nic. happy to answer any other q!

Yes, the description from TFA does not match the traditional Thunderbolt networking protocol, whose performance may be as low as that of a 10 Gb&#x2F;s Ethernet interface.The description from TFA matches what the poster above you said about a new Linux device driver that allows access to the raw Thunderbolt protocol for transferring data between computers. This appears to be an independent implementation of the same principle as in the device driver that will be merged in the mainline Linux.While the official Linux device driver makes the raw Thunderbolt appear like a file, which can be written and read to transfer data, this implementation emulates an Infiniband interface, which presumably was simpler to use for distributing work over multiple GPUs.They actually mention that with traditional Thunderbolt networking on the same computers, they had obtained only 9 Gb&#x2F;s, i.e. more than 5 times slower than what they obtained with raw Thunderbolt.

&gt; This is using Thunderbolt networking as transport,Are you sure? It doesn&#x27;t sound like it in some places in the text, e.g.:&gt;&gt; a kernel driver that sits alongside thunderbolt-net, allocating DMA rings from the controller&#x27;s NHI port in the same waybut I don&#x27;t have the domain knowledge to tell…

This is using Thunderbolt networking as transport, which incurs a bit of overhead.But starting with the upcoming Linux v7.2, there&#x27;s a new feature called USB4STREAM to use raw Thunderbolt packets as transport with minimum overhead and a super simple user interface:<a href="https:&#x2F;&#x2F;lore.kernel.org&#x2F;r&#x2F;20260511102744.1867485-1-mika.westerberg@linux.intel.com&#x2F;" rel="nofollow">https:&#x2F;&#x2F;lore.kernel.org&#x2F;r&#x2F;20260511102744.1867485-1-mika.west...</a>Release of v7.2-rc1 is predicted for Jul 5, that&#x27;s when this will first be available as a tarball. Until then you have to clone from thunderbolt.git&#x2F;next:<a href="https:&#x2F;&#x2F;git.kernel.org&#x2F;pub&#x2F;scm&#x2F;linux&#x2F;kernel&#x2F;git&#x2F;westeri&#x2F;thunderbolt.git&#x2F;log&#x2F;?h=next" rel="nofollow">https:&#x2F;&#x2F;git.kernel.org&#x2F;pub&#x2F;scm&#x2F;linux&#x2F;kernel&#x2F;git&#x2F;westeri&#x2F;thun...</a>Or alternatively linux-next:<a href="https:&#x2F;&#x2F;git.kernel.org&#x2F;pub&#x2F;scm&#x2F;linux&#x2F;kernel&#x2F;git&#x2F;next&#x2F;linux-next.git&#x2F;" rel="nofollow">https:&#x2F;&#x2F;git.kernel.org&#x2F;pub&#x2F;scm&#x2F;linux&#x2F;kernel&#x2F;git&#x2F;next&#x2F;linux-n...</a>Press coverage:<a href="https:&#x2F;&#x2F;www.phoronix.com&#x2F;news&#x2F;Intel-Linux-USB4STREAM" rel="nofollow">https:&#x2F;&#x2F;www.phoronix.com&#x2F;news&#x2F;Intel-Linux-USB4STREAM</a>

Thanks for saying this. I&#x27;m as &#x27;get off my lawn&#x27; about AI as any oldster at this point, but if all project were this up front about what and how they&#x27;re doing things I&#x27;d have a lot fewer reasons to grumble.

Kudos for the idea and being fully open to the state of this project (AI code, expect breakage)!

Fascinating. Infiniband is already fascinating, running it on something else is more fascinating.

thunderbolt-ibverbs: We have InfiniBand at home