NVIDIA’s Ampere Based DGX A100 Servers Leverage AMD Epyc Rome CPUs Instead of Intel Xeon

NVIDIA today unveiled its first 7nm Ampere based processor, the A100 and along with it the DGX A100 server for Data Center AI workloads. The new DGX system held a major surprise and it wasn’t related to the Ampere GPU itself, but rather the CPU paired with it. For the first time, NVIDIA’s DGX system is being powered by AMD’s Epyc Rome CPUs.

The most obvious reason for this shift is the vastly superior I/O of the Rome CPUs. Not only do the Zen 2 chips support PCIe 4 which rival Intel Xeons lack, but the number of lanes is also much higher. Furthermore, thanks to TSMC’s 7nm process, they are significantly more power-efficient than Intel’s 14nm Xeons. Other noteworthy advantages include a more competitive price-tag, increased memory support and last but not least, immensely higher compute performance.

You can read more about the Epyc Rome CPUs here.

The NVIDIA DGX A100 is a 2S system with 2x AMD Epyc Rome 7742 CPUs @ 2.25GHz, resulting in a total core count of 128. These behemoths are paired with 8x NVIDIA A100 Tensor Core GPUs and 1TB of DDR4 memory. The networking responsibilities are handled by Mellanox hardware, and the system includes a total of 15TB internal NVMe based SSD storage. It has a maximum power usage of 6.5kW, and offers as much as 5 petaFLOPS of AI performance and 10 petaOPS of INT8 mixed-precision performance.


Computer hardware enthusiast, PC gamer, and almost an engineer. Former co-founder of Techquila (2017-2019), a fairly successful tech outlet. Been working on Hardware Times since 2019, an outlet dedicated to computer hardware and its applications.

Related Articles

Back to top button