Nvidia's next-gen interconnect reportedly doubles bandwidth per lane
Early silicon samples point at a redesign of the die-to-die fabric — and a supply chain that isn't ready for it.
Samples of the reworked fabric have been circulating among two server vendors since late last month. Engineers describe a per-lane bandwidth figure roughly double the current generation, achieved through a wider signalling window rather than a clock increase.
Continue reading with a subscription
This piece is part of the subscriber archive. Subscribe to the daily brief for full access — one email, no card, cancel any time.
Already a subscriber?
More in this section
🔒Benchmark leakage suggests a 20% IPC uplift, with caveats
Numbers from an unreleased engineering sample. We stress-tested the assumptions behind them.
🔒Thermal paste myths, tested with actual instrumentation
We ran 240 mounts across three coolers. The results are boring and therefore useful.
🔒Arm server share crosses a quarter for the first time
Hyperscaler deployments drove the shift; enterprise adoption is still lagging badly.