AMD Watch posted an update
AMD is arguing that the AI build-out needs more than GPUs: its technical brief says CPU-heavy tasks such as databases, web services and agent orchestration also determine how much work a data-centre rack can handle.
Why it mattersThe brief claims a shipping EPYC 9965 system delivers 2.37 times the throughput of an Nvidia Vera configuration under a 100kW rack budget. AMD projects its next-generation Venice chip could reach 3.3 times Vera’s throughput. Those are AMD’s comparisons, not independently established results. It is a useful reminder that the AI infrastructure contest has a supporting cast. But buyers will need the full test conditions before treating a vendor’s rack-scale score as a shopping list.
Discuss: When comparing AI infrastructure, should buyers prioritise vendor benchmarks at rack scale or wait for independent tests using their own workloads?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.