Netflix considers switching from Intel to AMD EPYC in its datacenters

AMD has been making all the right moves lately with its EPYC line of processors, but now a huge name might be changing over from Intel Xeon CPUs to AMD EPYC CPUs in its datacenters: Netflix.

Netflix has servers in place that can hit 100Gbps of bandwidth, but as you can expect from a streaming giant like Netflix it is always looking forward. The next goal from there is a 200Gbps server, so Netflix has the option of going to a dual-socket Intel Xeon set up or a single-socket AMD EPYC solution.

If Netflix wants to hit 200Gbps of bandwidth it's going to need some beefy I/O, with the server infrastructure Netflix has deployed right now being a mix of Intel Xeon CPUs in both Broadwell and Skylake/Cascade Lake families. Broadwell-based Xeons have around 60GB/sec of memory bandwidth and 40 PCIe 3.0 lanes (with 32GB/sec I/O bandwidth).

Moving up to the Skylake and Cascade Lake-based Xeons which have 90GB/sec of memory bandwidth and 48 lanes of PCIe 3.0 with 38GB/sec of I/O bandwidth. If Netflix want to get close to 200Gbps, they're going to have to add a helluva lot more Intel Xeon CPUs... or switch to AMD EPYC.

A single-socket AMD EPYC 7502P solution has 120-150GB/sec of memory bandwidth (much higher than the 60-90GB/sec on the Intel Xeon solutions) and 128 lanes of PCIe 4.0 with 200GB/sec of I/O bandwidth (much, much higher than the 32-38GB/sec on the Intel Xeon solutions).

Netflix has considered moving to a dual Intel Xeon CPU solution, but a single AMD EPYC solution is much cheaper. If it wanted to go to a dual AMD EPYC solution in the future, then it's an even more attractive platform.