AMD EPYC™ 9004 + 8004 Series CPU Power Management: enabling peak workload performance and power efficiency
Two of the biggest challenges IT directors face today are data center performance and power efficiency. To optimize workloads and power efficiency, IT directors must manage overwhelming and complex variables. In this short blog, I am pleased to tell you about new tools AMD has developed to help you control often vexing variables and reinvent your data center.
AMD EPYC ™ 9004 and 8004 Series CPU Power Management makes it possible for you to transform your data center and optimize workload performance and power efficiency.
It's exciting to imagine the new ways in which you will be able to fine-tune individual workloads and energy use. EPYC processors are systems on a chip, with features such as telemetry agents that can make decisions based on pre-set parameters and real-time factors, e.g., heat and core idle states.
How can EPYC 9004 and 8004 Power Management help you and your team? This blog post highlights just some of the features I am inspired by. See our white paper, “AMD EPYC™ 9004 and 8004 Series CPU Power Management,” for more detail.
Setting parameters to meet varying workload requirements
Workloads can vary based on the demands of a given application. For example, some database workloads rely on a small number of high frequency cores, others such as technical computing applications need large caches and still others require high core density. How can IT directors, in effect, calibrate their data centers, even down to the component level, to best plan for and meet the demands of each workload? Understanding the key requirements of each workload and identifying relevant key variables is the first step.
System variables and parameters
Variables
• heat/thermals
• performance
• power
• voltage
• amperage
• core idle states
• current
Using telemetry agents and pre-set parameters, our EPYC Infinity System Management Unit (SMU) - a core component of our EPYC 9004 and 8004 processors – gives you the tools to define and implement policies to control and optimize the performance and efficiency of your data center, even in real time. Our EPYC 9004 and 8004 Series processors are self-regulating meaning they will perform optimally within the limits placed by the processor itself, the server infrastructure surrounding it, and the parameters you set.
We can work with you to define and set parameters to match your workload and power efficiency needs. For example, we can show you the various power management modes that you can set. Or you can use our already powerful pre-set parameters as well.
AMD Infinity Power Management features, examples:
- Configurable thermal design power (cTDP) - customize power consumption boundaries to as low as 70W TDP on some 8004 models
- Frequency boost - increase core frequencies above the base
• Processor P-States – control the core frequency based on workload demand
- Adaptive idle control – manage multiple levels of sleep and coordinate cache management parameters which will help ensure data consistency
- I/O performance settings – make automatic adjustments, such as capping PCIe® frequencies
• Infinity Fabric™ technology – adjusts width and speed to match bandwidth and latency needs
I’ve cited a few examples of parameters above. We can also work with you and your team to assess your workloads and applications and fine-tune your data center to the component level. The hybrid, multi-die “Zen 4” architecture of our EPYC 9004 and 8004 processors enables us to design combinations of cores and I/O dies to create products, so you can choose the processor that best matches your specific workload and application requirements and help ensure their optimal and consistent performance and efficiency.
Once you set parameters – or choose to use our pre-set parameters -- the EYPC Infinity System Management Unit (SMU) runs optimally within those limits. While, for example, the SMU may demand more current, it won’t use more current if it means breaching set parameters. The SMU is alert to workload, power demands, and its predetermined boundaries. AI telemetry agents feed the SMU with real-time data, and the SMU allocates power based on these real-time observations and its set parameters.
Component level control to enable peak performance and efficiency
Workloads running on EPYC processors rely on multiple components including memory controllers, the I/O subsystem, CPU dies, the cores that run them, the Level 3 cache, and the AMD Infinity Fabric that interconnects CPUs. Setting parameters and optimizing each of these components to enable peak workload performance and processor efficiency is a highly dynamic process as variables change. Such granular control at the component level combined with telemetry agents helps ensure optimal workload/application, processor, and data center performance and efficiency.
I have just scratched the surface with this blog post, and I am looking forward to seeing what you and your team can do with our new EPYC 9004 + 8004 Series CPU Power Management tools.
Learn more here.
