- Venice CPU double bandwidth with up to 256 zen nuclei 6
- 2027 AI Rack is expected to deliver 144 GPU with summer chips
- AMD plans the summer and MI500 deployment with 36 manure infrastructure
In his recent AI advanced event, AMD revealed some of the first technical details about its next -generation Zen 6 processor, called “Venice”.
Built in a 2 Nm process and in a packaging of up to 256 Zen 6 nuclei, an increase of 33% with respect to the EPYC ‘Turin’ processor of current generation, AMD seeks to raise the bar for the optimized calculation power AI -AI -AI -AI -AI -Opi.
It is expected that more than the double memory bandwidth compared to current EPYC CPUs, reaching 1.6Tb/s per plug. Venice is also compatible with Pcie Gen6, which offers faster communication between CPUs and GPUs.
Pushing the envelope
“Venice extends our leadership in each dimension that matters in the data center,” said Lisa his, CEO of AMD. “More performance, better efficiency and an outstanding total cost of property.”
“It offers 70% more calculation performance than our current generation leadership CPU,” he added. “And to continue feeding the MI400 with data at full speed, even at rack scale, we have doubled both the GPU and the memory bandwidth and the optimized venice to function at higher speeds. Now we have just recovered Venice in the laboratories and looks fantastic.”
While Venice heads AMD’s 2026 platform, his idea also gave us an idea of what the chips giant is working for 2027.
“We are already deeply in the development of our 2027 shelf that will push the envelope even more to performance, efficiency and scalability with our next summer series GPU and the GPUs of the MI500 series, so many things from AMD,” he revealed.
This new system is based on the “helios” shelf that is shown by 2026, which houses 72 GPU Mi400 in 18 racks. By 2027, AMD is planning an even more expansive configuration with 36 racks.
Assuming a similar configuration, the next -generation AI shelf could contain 144 GPU. The summer platform is expected to maintain the high bandwidth and the efficiency of performance established by Venice, probably with more memory and computation for shelf.
AMD has not officially appointed its 2027 rack, but we hear that informally reference is made in industry circles as a continuation on a larger scale of Helios.
Through Tom hardware
