AMD just released the first real performance numbers for Venice, its next generation EPYC server processor, and the results are not close. Under a 100 kilowatt rack power limit, Venice scores 3.30 on the SPEC CPU 2017 benchmark using Nvidia's Vera processor as the 1.0 baseline. Intel's Xeon 6980P scores 1.46. AMD's own previous generation Turin scores 2.37. Venice does not just beat the competition, it more than doubles Intel and more than triples Nvidia within the same power budget.
The way AMD framed this test is worth paying attention to because it reflects how data center buying decisions actually get made. A benchmark that measures a single chip at peak performance in isolation tells you almost nothing useful if you are a cloud operator trying to figure out how much computing capacity fits inside a 100 kilowatt rack. Power is the real constraint in modern data centers, not physical space or even cost per chip. AMD built its comparison around that reality rather than chasing headline single-chip numbers, and the result is a benchmark that reads like a procurement argument rather than a product demo.
Venice runs on a 2nm process and carries 256 cores built on Zen 6 architecture. The three processors in AMD's comparison carry 256, 128, and 88 cores respectively, and AMD is transparent about that gap. More cores under the same power envelope is precisely the point. When a data center operator is constrained by how many kilowatts a rack can draw, the processor that delivers the most work per watt wins the contract. Venice is engineered around that exact argument.
The single core numbers are interesting for a different reason. Even averaged across individual cores, the 256-core Venice still holds a 27 percent advantage over Nvidia Vera. Drop Venice down to its 96-core configuration and that advantage shrinks to 11 percent. The single-core gap is real but modest. The rack-level gap is enormous. AMD's strategy is clearly to let the core count and power efficiency combine into an argument that is difficult to counter at the infrastructure level.
For Intel the 1.46 score against Venice's 3.30 is the kind of gap that does not get closed with a firmware update. The Xeon 6980P is Intel's current server flagship and it is being outperformed by more than double within the same power budget. AMD has been taking data center market share from Intel steadily since the first EPYC generation, and Venice suggests that trajectory is not slowing down.
Nvidia's position in this comparison is different because Vera is built on ARM architecture through the Olympus platform rather than x86, and CPU workloads are not where Nvidia is trying to win the data center argument. Nvidia's real play is GPU compute for AI training and inference. The fact that AMD is using Vera as its baseline rather than Intel says something about where AMD sees the real competitive battle shifting as AI infrastructure spending dominates data center investment.
Venice has not launched publicly yet and these numbers come from AMD's own testing, which means independent verification is still ahead. But the methodology is sound and the metric being measured is the one that actually matters to the buyers writing the cheques.
