AMD has released preliminary performance data for its upcoming Epyc Venice server processors based on the Zen 6c architecture. In the SPECrate 2026_int_base benchmark, the 256-core Epyc 9996 outperformed the 192-core Epyc 9965 by approximately 78 percent. Accounting for the 33 percent increase in core count, the per-core performance gain is estimated at roughly 34 percent.
The company claims the Epyc 9996 is more than twice as fast as the Intel Xeon 6980P and the Nvidia Vera processor in the same integer throughput test. Additionally, a 96-core configuration of the Venice chip showed a 20 percent lead in per-core performance over the 88-core Nvidia Vera. AMD noted that these figures are based on varying system configurations and compiler versions.
In memory bandwidth testing using the STREAM benchmark, the 96-core Epyc 9996 demonstrated an 18 percent higher throughput compared to the Nvidia Vera. On a per-core basis, this advantage narrowed to 8 percent. These tests were conducted with a 600-watt power limit for the AMD processor.
AMD also evaluated the new architecture across various cloud and enterprise workloads, including database management, Java applications, and cryptographic tasks. Furthermore, the company showcased performance in agentic AI tasks, such as NGINX web server operations and vector search via FAISS, while noting that these results are preliminary and subject to further official verification.
