Google & Twitter adopts AMD EPYC Rome processors

AMD EPYC Rome processors

After posting impressive result with their 2nd gen EPYC (Rome) server processors, AMD have found its first large customers. And as a bonus, its two customers instead of one. The search giant Google and the micro blogging site Twitter. Google and Twitter both will be upgrading some of their server farms to Epyc Rome from Intel Xeon.


As per reports Google will be offering Rome processors through its Google Cloud virtual machines. At the same time Twitter will shift some of its data centers to EPYC Rome to reduce energy consumption. As Epyc Rome in a 2-socket (2P) configuration offers up to a staggering 128 cores & 256 threads, which should allow more dense servers that not just save space but reduce electricity bill on the long run.


But this is just a tiny step for AMD as Intel continues absolute dominance in the server space with a total share of above 90%. With EPYC Rome and its successors, AMD believes it can finally reach double digit figure in a few year’s time.


Unlike consumer processor like Ryzen, server processors require a lot more work from customer side as they need to optimize their applications, software, toolchains, etc. before shifting to a completely new architecture. AMD allowed them to do just that with 1st gen EPYC, while at the same time demonstrating that they can hold against Intel onslaught. With 2nd gen EPYC, AMD made a healthy 10-20% performance lead over Intel’s offering at lower cost forcing Intel client to consider AMD. And with a lower cost of ownership and less power draw, server giants had no other option than to adopt EPYC one and for all.




AMD achieved this impressive 64 cores per processor by way of chiplets. Instead of one large chip, they broke the chips into 8 core chips (or chiplets) that are then interconnected by AMD Infinity fabric. This allowed them to mass produce at way lower cost as manufacturing a single large die is both complicated and prone to manufacturing defects, which can render the processor useless. Furthermore, the performance light parts like memory controller & PCIe lanes are packed into a 9th chip called the IO die which is manufactured in 14nm process unlike 7nm used for core chips. This development allows AMD to keep the price further low and focus on what’s important. The and end result being, they can add whatever number of chiplet they want to get the kind of processor core configuration they want.

Leave a Reply