How AMD Is Advancing Energy Efficiency for IT Users

On Energy Efficiency Day 2022, AMD shares how the company is helping its customers and end-users reduce their energy use and GHG emissions
Oct 6, 2022 8:00 AM ET
Article

Originally published in AMD 2021-22 Corporate Responsibility Report

At AMD, we strive to create products that improve people’s lives and help our customers and end-users reduce their own energy use and GHG emissions. We collaborate closely with our customers and partners on product design and system-level optimizations to advance environmental sustainability, including by minimizing environmental impacts and energy use.

Maximizing the computing performance delivered per watt of energy consumed is a vital aspect of our business strategy. Our products’ cutting-edge chip architecture, design and power management features have resulted in significant energy efficiency gains, and we have the track record to prove it: we achieved a 31.7x increase in performance per watt for processors in mobile devices, for example, exceeding the AMD 25x20 Energy Efficiency goal (2014-2020).1

Today, AMD RyzenTM processors are continuing to help deliver energy-efficient laptops without compromise, including our Ryzen 5800U processors with up to 43 percent better power efficiency than Energy Star 8.0 requirements.2 An enterprise that upgrades 10,000 PCs from four-year-old Ryzen 2500U processors to new Ryzen 5800U processors would save approximately 272,000 kWh of electricity and 183 metric tons of CO2e, equivalent to 3,187 tree seedlings grown for 10 years.3

In the data center, AMD EPYC™ processors power the most energy-efficient x86 servers, delivering exceptional performance and reducing energy costs.4 AMD EPYC processors drive energy efficiencies by meeting application performance demands with fewer physical servers than competitive solutions, which can result in a reduced data center footprint and associated energy use and GHG emissions. For example, to deliver 1200 virtual machines, it takes an estimated 10 2P AMD EPYC™ 7713-powered servers or 15 2P Intel® Platinum 8380-based servers. The AMD solution takes an estimated 33 percent fewer servers, uses approximately 32 percent less power and provides estimated GHG emission savings of about 70 metric tons of CO2e, equivalent to the carbon sequestration of 28 acres of forest in the United States.5

Energy efficiency is paramount when it comes to supercomputing, which is the concentration of processing power across multiple, parallel computers. The Green500 list ranks the most energy-efficient supercomputers in the world. The June 2022 update showed AMD EPYC processors and AMD Instinct accelerators power the most efficient supercomputers in the world, including four of the top five, eight of the top ten, and 17 of the top 20 most efficient. The Frontier test and development system (TDS) supercomputer secured the top spot in the Green500 June 2022 update based on optimized 3rd generation AMD EPYC processors and AMD Instinct MI250x accelerators.

Our continued ambitions are reflected in the AMD goal of a 30x increase by 2025 in energy efficiency for AMD processors and accelerators powering servers for high-performance computing and artificial intelligence training. In 2021, AMD achieved a 3.9x increase. Nearly midway through 2022, we are on track toward achieving our goal, having reached a 6.8x improvement in energy efficiency compared to 2020 using an accelerated compute node powered by one 3rd generation AMD EPYC CPU and four AMD Instinct MI250x GPUs.6 Our goal utilizes a measurement methodology validated by renowned compute energy efficiency researcher and author, Dr. Jonathan Koomey.

“The energy efficiency goal set by AMD for accelerated compute nodes used for AI training and High-Performance Computing fully reflects modern workloads, representative operating behaviors and accurate benchmarking methodology.”

Dr. Jonathan Koomey, President, Koomey Analytics

For more information on environmental sustainability programs and initiatives at AMD, please visit our website.

Footnotes:

  1. Testing by AMD Performance Labs as of April 15, 2020. Processors tested: AMD FX-7600P, AMD FX-8800P, AMD FX-9830P, AMD Ryzen 7 2700U, AMD Ryzen 7 2800H, AMD Ryzen 7 3750H, and AMD Ryzen 7 4800H. 25x20 program tracked against ENERGY STAR Rev 6.1 8/12/2014 and 3DMark® 2011 P-Score and Cinebench R15 nT. Results may vary with drivers and BIOSes. RVM-108
  2. Based on measurements by AMD labs as of February 2022 of the AMD Ryzen 7 Pro 5800U against Energy Star 8.0 requirements. Results will vary. CZM-146
  3. Estimated KWh savings based on Energy Star measurements of Ryzen 2500U vs. Ryzen 5800U as measured in AMD lab. Estimates of CO2 reduction and tree seedlings grown based on the EPA greenhouse gas equivalencies calculator https://www.epa.gov/energy/greenhouse-gas-equivalencies-calculator. Results may vary. CZM-147
  4. EPYC-028A: As of 2/2/22, of SPECpower_ssj® 2008 results published on SPEC’s website, the 55 publications with the highest overall efficiency results were all powered by AMD EPYC processors. See https://www.amd.com/en/claims/epyc3x#faq-EPYC-028 for the list. More information about SPEC® is available at http://www.spec.org. SPEC and SPECpower are registered trademarks of the Standard Performance Evaluation Corporation.
  5. MLNTCO-021: This scenario contains many assumptions and estimates and, while based on AMD internal research and best approximations, should be considered an example for information purposes only, and not used as a basis for decision making over actual testing. The AMD EPYC™ SERVER VIRTUALIZATION and GREENHOUSE GAS EMISSIONS TCO ESTIMATION TOOL tool compares the 2P AMD EPYC™ and the 2P Intel® Xeon® server solutions required to deliver 1200 total virtual machines (VM), requiring 1 core and 8GB of memory per VM. The analysis includes both hardware and virtualization software components. Hardware costs (CPU + memory + storage + chassis): The 2P AMD 64 core EPYC_7713 processor used in this solution analysis provides 128 total cores per server, each processor cost $7060 and the server uses 32 x 32GB DIMMs to achieve the minimum required memory footprint, in a 1RU server chassis that cost $2200, and requires 1 server racks. The AMD solution has a total estimated hardware acquisition cost of $217880. The 40 core Intel Xeon Platinum_8380 processor used in this solution analysis provides 80 total cores per server. Each processor cost $8666 and the server uses 16 x 64GB DIMMs to achieve the minimum required memory footprint, in a 2RU server chassis that cost $2500 and requires 2 server racks. The Intel solution has a total estimated hardware acquisition cost of $390060.

    OPERATING COSTS: The core assumptions for this analysis are as follows: Cost of power @ $0.12 with kwatts (kW) of power to each rack and a PUE (power usage effectiveness) of 1.7 and a server rack size of 42RU. Each server has 1 hard drives drawing 3 watts each. Server Admin annual salary is $85000 managing 30 physical servers with a salary burden rate of 30 percent. The VM Admin salary is $85000, with a burden rate of 30percent and managing 400 VMs.

    AMD has estimated OpEx costs as follows: a hardware admin cost of $110500, a real estate cost of $19440, and a power cost of $40208.4, for a total estimated 3 year TCO cost (hardware cost and operating expense) of $388028 with AMD. Estimated OpEx costs for Intel are: hardware admin cost of $165750, real estate cost of $38880, and power cost of $58704.

    HARDWARE TCO: This is the CapEx and OpEx directly associated with the hardware. The AMD EPYC_7713 solution requires 10 - 2P servers with a CapEx of $217880 with a total estimated 3-year TCO cost (CapEx plus OpEx) of $388028. The Intel Platinum_8380 processor requires 15 - 2P servers with a CapEx of $390060 with a total estimated 3-year TCO cost (CapEx plus OpEx) of $653394. The AMD solution has an estimated 41 percent lower hardware TCO for this virtualization solution, 1 - ($388028 ÷ $653394) = 41 percent, than the Intel solution.

    VIRTUALIZATION TCO: Analysis is based on the following estimates: 3 year Virtualization (hardware, operating, and software cost) for the Intel solution is $2005974 and $1621248 for the AMD solution. This means that the AMD solution is ~19% less expensive over three years. 1 - ($1621248 ÷ $2005974) = 19 percent. The EPYC solution 1st year TCO is $844816 and the Intel 1st year TCO is $1167418. The AMD solution 1st year TCO per VM of $704.01 where the Intel 1st yr. solution is $972.85. The AMD 1st year TCO per VM is $268.83, or ~28 percent lower than Intel. The 1st year TCO per VM is calculated by taking the 1-year TCO (hardware, software, and 1st year OpEx) and dividing it by the total number of VMs. The virtualization software used in this analysis is VMware with a VMware® vSphere Enterprise Plus w/ Production support license. This analysis uses license pricing of $5968 per Socket + Core with 3 year support. More information on VMware software can be found @ https://store-us.vmware.com/vmware-vsphere-enterprise-plus-284281000.html.

    For 1200 VMs with 1 core(s) per VM, and 8 GB of memory per VM, the Intel Platinum_8380 processor requires 15 servers, and 60 licenses. The AMD EPYC_7713 solution requires 10 servers and 40 licenses. The AMD solution requires 33 percent fewer servers than the Intel solution.

    The AMD server and virtualization software license cost are $456600, and the Intel cost are $748140. Hardware and virtualization cost are ~$291540 or ~39 percent Lower w/ AMD.

    AMD EPYC_7713 powered servers save ~154132.2kWh of electricity for the 3 years of this analysis. Leveraging this data, using the Country / Region specific electricity factors from the '2020 Grid Electricity Emissions Factors v1.4 – September 2020', and the United States Environmental Protection Agency 'Greenhouse Gas Equivalencies Calculator', the AMD EPYC powered server saves ~69.86 Metric Tons of CO2 equivalents. This results in the following estimated savings based on United States data, Greenhouse Gas Emissions Avoided of one of the following:

    15 USA Passenger Cars Not Driven for 1 year; or;
    5 USA Passenger Cars Not Driven Annually; or;
    173382 Miles Driven by Avg Passenger Car; or;
    or CO2 Emissions Avoided from:
    7894 Gallons of Gasoline Not Used; or;
    77261 Pounds of Coal Not Burned in USA; or;
    9 USA Homes' Electricity Use for 1 year; or;
    3 USA Homes' Electricity Use Annually; or;
    or Carbon Sequestered equivalent to:
    1153 Tree Seedlings Grown for 10 years in USA; or;
    84 Acres of USA Forests in 1 year; or;
    27.94 Acres of USA Forests Annually.

    The 2020 Grid Electricity Emissions Factors v1.4 – September 2020 data used in this analysis can be found at https://www.carbonfootprint.com/docs/2020_09_emissions_factors_sources_for_2020_electricity_v14.pdf and the US EPA Greenhouse Gas Equivalencies Calculator used in this analysis can be found at https://www.epa.gov/energy/greenhouse-gas-equivalencies-calculator.

    Virtualization software pricing sourced online as of 09/14/2021. Third-party names are for informational purposes only and may be trademarks of their respective owners. All pricing is in USD.

    AMD CPU pricing based on 1KU price as of January 2022. Intel® Xeon® Scalable CPU data and pricing from https://ark.intel.com as of January 2022. All pricing is in USD. Results generated by: AMD EPYC™ SERVER VIRTUALIZATION and GREENHOUSE GAS EMISSIONS TCO ESTIMATION TOOL - v10.13

  6. EPYC-030: Calculation includes 1) base case kWhr use projections in 2025 conducted with Koomey Analytics based on available research and data that includes segment-specific projected 2025 deployment volumes and data center power utilization effectiveness (PUE) including GPU HPC and machine learning (ML) installations, and 2) AMD CPU socket and GPU node power consumptions incorporating segment-specific utilization (active vs. idle) percentages and multiplied by PUE to determine actual total energy use for calculation of the performance per Watt. 6.79x = (base case HPC node kWhr use projection in 2025 x AMD 2022 perf/Watt improvement using DGEMM and typical energy consumption + Base case ML node kWhr use projection in 2025 *AMD 2022 perf/Watt improvement using ML math and typical energy consumption) /(2020 perf/Watt * Base case projected kWhr usage in 2025). For more information on the goal and methodology, visit https://www.amd.com/en/corporate-responsibility/data-center-sustainability.