I walked into the office earlier this week to find a serious discussion raging amongst the engineering team. Concerned that the team had hit some technical roadblock, I skipped my customary walk to the coffee machine and instead grabbed a chair in the bullpen.
“What happened?” I asked. “Did we run into a major bug?”
I hung up my engineering spurs 10 years ago and have been masquerading as a product manager since, so I’m not as close to server specs as I used to be. But I thought this discussion was over back in the AMD K6 vs Intel Pentium days! I was clearly wrong.
BK and Bobby went on to give me a crash course on memory channel counts, L3 cache strategies, pre-fetching benefits, and a number of other specs that I haven’t thought about since the computer architecture course I struggled through back in college.
Later that day, the team decided to run a number of internal benchmarks to make an informed decision on what processor makes most sense. Here’s what we found:
- There are scenarios in which AMD wins, and there are scenarios in which Intel wins. Yes, Captain Obvious came by afterwards to high-five the team.
- On a per core basis, AMD is more cost effective.
If all else is equal and AMD allows us to keep our operating costs lower, AMD makes sense for our Programmable Edge platform. And as edge infrastructure providers continue to make AMD available as part of their offerings, there are plenty of reasons why we should be using EPYC for general purpose, x86 workloads.
I’ll write a follow-up blog when we have more data available. If you have already carried out a similar analysis and are comfortable sharing your learnings, we would love to hear from you.
If you’d like to keep track of what we’re up to as we build out the world’s first Programmable Edge, please do sign up for periodic updates.