Funny this one. TSMC has announced that AMD EPYC processors are now powering workloads for its global microchip manufacturing business. The company not only fabricates wafers but is also heavily engaged in research and development.
From Europe to Asia to the United States, TSMC’s 50,000+ employees needed powerful IT services to communicate and collaborate seamlessly – all while reducing total cost of ownership. That’s exactly what TSMC achieved with AMD EPYC processors. TSMC had to deliver and balance server core density within its data centre environment. “Our existing data centres are limited by space and power,” says Simon Wang, Director of Infrastructure and Communication Services Division at TSMC. TSMC needed to find products that could directly be introduced into its standard template, and then deployed across its data centres. It designed and implemented the virtualisation of the computing, storage and network, and all of that in our standard template of service architecture. TSMC has two VM [ed. virtual machine] configurations for our workloads depending on their complexity.
But running a lot of VMs on one server isn’t optimal. “If I had my server supporting 20 machines, then when the server goes down, 20 machines will take the hit,” says Wang. “That is why we have two types of VM server configurations to reduce the risk.”
How to reduce total cost of ownership
TSMC partnered with HPE and AMD, choosing the HPE DL325 G10 platform running the 2nd Generation AMD EPYC 7702P CPU with 64 cores, a base 2GHz clock frequency and up to 3.35GHz boost clock.
“For our general workloads, memory density is a clear advantage of AMD EPYC processors”, says Wang. With AMD EPYC processors, fewer resources were required for each virtual instance in the HCI implementation, while achieving better performance. There were cost savings from having fewer physical servers due to the per-processor core density, reducing management and power costs as well as the space required in the data centre.
The reduced number of servers will provide increased room to expand in the data centre as the company grows further. Switching from dual to single socket also meant hardware, operating and software cost savings as well as reduced power consumption.
Expanding deployment to manufacturing
“We're now introducing EPYC servers into our manufacturing teams,” says Wang, looking at the new EPYC 7F72 with 24 cores and a base 3.2GHz clock frequency. “That's the CPU we're considering for R&D, because of the high clock rate. This team doesn't necessarily need more sockets or cores. For R&D, if we use two sockets, that might create unexpected effects, because it means one CPU needs to communicate with another CPU, which will create overheads. So, for R&D, we choose a one-socket CPU and the high clock-rate will be an important advantage.”
For automation with the machinery inside the fab, each machine needed to have one x86 server to control the operation speed and provision of water, electricity and gas, or power consumption. These machines are very costly, at potentially billions of dollars, while the servers that control them are much cheaper. Wang and his team need to ensure they have a high availability in case one rack is down, then they can use another rack to support the machine. With a standard building block, TSMC can generate about 1,000 virtual machines, which can control 1,000 fab tools in the cleanroom. This will mean a huge cost saving without sacrificing failover redundancy or reliability.
TSMC is planning to implement two more data centres in 2021.