AMD’s CFO believes the CPU is still the core of many server workloads

AMD’s Chief Financial Officer, Jean Hu, told the Nasdaq Investor Conference in June that she believes the CPU is still important for server workloads.  

Whilst the GPU has become the core of many AI servers, Hu stated that it’s important to remember the importance of other workloads.  

“When you look at the traditional foundational applications, your ERP system, your database and your shopping website, your Meta, Facebook, Instagram, all those things, you don't need the GPU – the CPU has the best TCO for those kind of foundational traditional applications,” she said. 

This is a good reminder. As mentioned, the flexibility of the CPU makes it ideal for handling a diverse range of workloads, from running databases and web servers to executing complex software applications.  

There are several workloads where you’ll still want to prioritise the CPU. They’re optimised for handling database queries, updates and transactions, and are generally more efficient when used for handling network traffic, firewall rules, and security protocols. They also benefit from efficient instruction pipelines, which makes them the best choice for tasks which require handling multiple jobs simultaneously, like managing virtual machines.  

Hu added that problems in the general compute side are addressed by core counts increasing into the double digits, such as in their upcoming EPYC 9005 series processors, whose Core / Thread counts reach up to 128 / 256.  

This doesn’t mean the GPU isn’t useful in non-AI workloads. A server combining CPUs and GPUs provides optimised performance for many kinds of workloads. With the CPU handling tasks that are not GPU-accelerated, the GPU is freed up for computationally intensive workloads.  

Overall, whilst the GPU has become vital to AI workloads, as AMD CFO Jean Hu shared, the CPU remains the core of many other server workloads. Which you should priortise depends on your workload and your budget.  

Our range of servers for different business workloads can be found here.