AMD Details Epyc Venice Zen 6 Server CPUs at Advancing AI Event
AMD has officially unveiled the Zen 6 generation of Epyc server CPUs, codenamed Venice. Sporting a new socket and up to 256 Zen 6c cores, these new chips are going to be far faster than their predecessors, TechPowerUp reports. AMD even calls them the best CPUs for agentic AI, taking the fight directly to Nvidia and its new Vera chips, as well as stomping Intel’s ambitions as firmly into the dirt as it can.
AMD’s next-generation server chips have been brewing for some time, and we’ve heard lots of leaked and rumored details over the many months build up. But now we have it all officially from the red horse’s mouth.
The chips are split into four distinct ranges: the Epyc 9006 Venice family with socket SP7 designed for mainstream servers; Epyc 9006 on socket SP8 for enterprises for optimized performance per node; Epyc 9006X on SP7 for high-performance computing with large cache; and Epyc 9007 LP Verano, for AI hoisting duties and paired with LPDDR5X memory.
It’s all confusing enough that I’m glad I’m not making the purchasing decisions at the gigantic enterprises that actually make up the customer base for these kinds of chips. The hardware itself, though, is exciting and intimidating in equal measure.
Credit: AMD
In the Venice family of new Epyc chips, there’s a 96-core/192-thread version with full Zen 6 cores, up to a 256-core/512-thread version with Zen 6c high-density cores. All the chips enjoy a new 16-channel DDR5 interface, with DDR UDIMM support for up to 8,000MT/s memory, or MRDIMMs at up to 12,800MT/s.
Each chip type also supports PCIe 6 with a new enhanced Infinity Fabric, and advanced power delivery and improved power efficiency through unified power provisioning (UPP), utilization-based power scaling (UBPS), and frequency-adaptive spectrum technology (FAST).
Unsurprisingly, AMD’s first-party results paint these chips as the best of the best, but we’ll need to wait for third-party results to confirm. Notably, however, AMD’s charts show the 9996 with 256 cores and a 600W TDP delivering up to twice the throughput of Intel’s 6th-generation Xeon 6980P CPU with 128 cores and a 500W TDP.
Perhaps more crucially, it shows it delivering up to 2.2x the throughput as Nvidia’s 88-core Vera CPUs, and up to 1.2x the throughput when pitting it against the 96-core, more modest Venice CPU.
With AMD now claiming over 50% of the server CPU market, it’s well and truly overtaken Intel. The question now becomes whether it can make inroads on Nvidia’s CPU dominance in the server space, and ward off its attempts to enter the consumer/prosumer arena with its N1x chips.
