Announcements
We ıntegrate ınformatıon ın lıfe

  • DOLAR
    %0,05
  • EURO
    %0,17
  • ALTIN
    %0,05
  • BIST
    %0,02
Response from AMD to Nvidia: Zen 6 with 256 Cores Pushes the Boundaries

Response from AMD to Nvidia: Zen 6 with 256 Cores Pushes the Boundaries

AMD shared the performance assumptions of its 256-core Zen 6 architecture EPYC Venice processors against Nvidia Vera. Here are the details.

AMD shared the first official performance estimates for the new generation EPYC Venice processors, which will feature Zen 6 architecture. The company claims that its 256-core flagship model offers 3.3 times higher performance than the Nvidia Vera processor at the scale of a server rack with a 100kW power budget.

This data is based on the total capacity of an entire rack, rather than performance on a single socket or system. AMD emphasizes that the results it shares aim to provide a guiding comparison rather than directly measured shelf tests.

Performance Based on Modeling and Claims

AMD states that it followed a comprehensive modeling technique instead of direct physical tests when reaching these results. The company calculated the number of nodes that would fit into the 100kW budget based on the processor TDP values ​​and the power consumption of other components, and then multiplied the performance of these nodes with the claims.

AMD, which does not have physical access to the Nvidia Vera processor, scaled the results obtained from the Nvidia Grace chip by 1.63 times with the information published by Phoronix to obtain these numbers. The 256-core EPYC Venice results were generated using the EPYC 9965 processor with a 1.7x scaling factor and in-house testing.

These calculations bring with them the fact that systems cannot be scaled linearly. It is known that in real-world scenarios, factors such as medium relations, thermal constraints and power limits play a determining role on performance.

Data Center Focused Comparisons

AMD shares these performance claims through a series of tests focused on general-purpose data center missions. The benchmarks the company uses include SPEC CPU 2017 integer throughput, SPECjbb 2015-based server-side Java, NGINX web server load, Redis in-memory workloads, and TPROC-C database performance on MySQL.

It is noteworthy that these data are a response to the publication of Vera results supported by Nvidia on Phoronix. AMD is preparing to give much more detailed information about Zen 6 architecture, Venice processors and corporate roadmap at the Advancing AI event to be held next month.

How do you think this estimated performance data from AMD will affect the competition in the information center market?

Social Media Share:

TOGETHER FOR A LOOK

Can you share with us your comment?