{"slug": "mac-thunderbolt4-real-latency-numbers", "title": "Mac Thunderbolt4 real Latency numbers", "summary": "A Level1Techs forum user posting as boatman reported that USB RDMA on Macs is \"just good software\" and not real RDMA hardware offload, after being unable to obtain ib_read_lat, ib_write_lat or ib_send_lat latency numbers on a two-node Mac Thunderbolt setup. The user measured 7.8uS over TCP on Strix Halo with busy polling enabled via net.core.busy_poll=100 and net.core.busy_read=100, and set thresholds of greater than 5uS for software RDMA, under 2uS for genuine hardware RDMA.", "body_md": "Hello All,\n\nNot posted in a long time. but I cant find answers anywhere else.\n\nI have been following the AI clustering topics about USB4.\n\n**I need someone to help get me real latency numbers for Mac Thunderbolt RDMA.**\n\nCan someone run ib_read_lat, ib_write_lat and ib_send_lat or some equivalent on a 2 node setup.\n\nI need to know if it is real RDMA or just really good software RDMA. My understanding right now is that there is no way that thunderbolt can bypass the CPU/kernel and go to the GPU vRAM with a zero-copy. There is a limit to how low they can get without adding a RDMA chip to the USB port, which apple might do.\n\nIf results are >5 than it is just software RDMA.\n\nIf results are lower than they must have RDMA hardware.\n\nIf it is <2uS then we are really cooking.\n\nNotes:\n\nI have tested both `thunderbolt-ibverbs` and `OdinLink-Five`. and they are good but most of the improvements they get in latency is from enabling busy polling. I am running it with Strix Halo and getting 7.8uS over TCP. I want to know if Apple is just doing the same trick.\n\n`sudo sysctl -w net.core.busy_poll=100`\n\n`sudo sysctl -w net.core.busy_read=100`\n\nI look forward to getting real numbers for latency.\n\n \n\n \n[boatman](https://forum.level1techs.com/u/boatman)\n2\n \nUpdate: I was not able to get numbers but I can confirm that the USB RDMA on Macs is just good software. not real RDMA hardware offload.", "url": "https://wpnews.pro/news/mac-thunderbolt4-real-latency-numbers", "canonical_source": "https://forum.level1techs.com/t/mac-thunderbolt4-real-latency-numbers/253734#post_2", "published_at": "2026-09-16 05:41:08+00:00", "updated_at": "2026-09-16 06:06:26.392433+00:00", "lang": "en", "topics": ["ai-infrastructure", "ai-chips"], "entities": ["Apple", "Mac", "Thunderbolt", "USB4", "Strix Halo", "Level1Techs", "boatman", "thunderbolt-ibverbs"], "alternates": {"html": "https://wpnews.pro/news/mac-thunderbolt4-real-latency-numbers", "markdown": "https://wpnews.pro/news/mac-thunderbolt4-real-latency-numbers.md", "text": "https://wpnews.pro/news/mac-thunderbolt4-real-latency-numbers.txt", "jsonld": "https://wpnews.pro/news/mac-thunderbolt4-real-latency-numbers.jsonld"}}