AMD's 256 core EPYC Venice lands on TSMC's 2nm node

Started by Grover26, Jul 24, 2026, 12:34 AM

Previous topic - Next topic

0 Members and 2 Guests are viewing this topic.

Topic: AMD's 256 core EPYC Venice lands on TSMC's 2nm node   Views(Read 90 times)

Grover26

AMD introduced EPYC Venice at its Advancing AI event, its first Zen 6 server chip and the first HPC processor to reach volume production on TSMC's 2nm N2 node using gate all around nanosheet transistors. It tops out at 256 cores spread across eight 32 core chiplets on a new SP7 socket, with 16 memory channels and 1.6 terabytes per second of memory bandwidth alongside PCIe Gen 6 support. AMD is claiming a 70 percent performance leap over the current Zen 5 EPYC Turin lineup

First systems are expected to ship in Q3 2026, which is a fairly tight window from announcement to actual delivery for a brand new node and socket combination. Being first to volume ramp on 2nm for HPC chips is a genuine milestone regardless of how the real world benchmarks eventually shake out

This launched the same day as the AMD Anthropic partnership, and Venice CPUs are explicitly paired with MI455X GPUs in the Helios rack systems mentioned in that deal. It is clearly meant to be sold as one integrated compute platform rather than separate CPU and GPU product lines

CodeOracle

256 cores on a single socket is a number that would have sounded like science fiction a decade ago
Still figuring it all out

Joel96

First to 2nm volume ramp for HPC is a real bragging right over Intel regardless of how you feel about the rest of the announcement
404: Signature not found

PhotonBurst16

70 percent improvement claims from the vendor itself always need independent benchmarking before I get excited

Daemon92

Pairing this directly with MI455X in the Helios racks makes clear this was designed as one platform from the start
Quantum computer said maybe, so I'm calling it a win

NeverQuitZach33

1.6 terabytes per second of memory bandwidth is the number that actually matters for a lot of HPC workloads more than core count

FirstBen12

Q3 shipping target for a brand new socket and node combo is ambitious, curious if that holds

Buffer

Feels like Intel really needs a big answer here or they keep losing the server CPU narrative entirely

MellowRunner

The SP7 socket change means a whole new motherboard ecosystem, that is a real transition cost for buyers
My code works & I have no idea why

Save money on everyday spending Free cashback on thousands of retailers
View offer