I'm thrilled to talk about an important addition to the AMD EPYC server processor family: 4th Gen AMD EPYC™ 4004 Series Processors. Large enterprises, demanding data centers, supercomputers, and hyperscalers have come to rely on the robust performance and scalability of AMD EPYC 7000, 8000, and 9000 Series Processors. By contrast, AMD EPYC 4004 Series Processors targets dedicated hosting providers and small businesses seeking cost-effective solutions for entry-level server workloads. These processors feature low core counts, Thermal Design Power (TDP) as low as 65 watts, and competitive pricing while maintaining the performance, scalability, and reliability expected from the AMD EPYC product line. This blog delves into some of the workloads for this market segment and the performance benefits offered by AMD EPYC 4004 processors.
Technical Specification
The AMD EPYC™ 4004 Series Processors offer a powerful solution with boost frequencies reaching up to 5.7 GHz. These processors feature configurations ranging from 4 to 16 "Zen 4" cores, spanning up to 2 Core Complex Dies (CCDs), and support 8 to 32 threads with Simultaneous Multi-Threading (SMT) enabled. Each CCD provides up to 32 MB of shared L3 cache, totaling up to 64 MB per processor. Equipped with Gen 3 Infinity Fabric architecture, the AMD EPYC 4004 processors support die-to-die bandwidth of up to 32 Gbps and can accommodate up to 192 GB of ECC-enabled DDR5-5200 RAM. They also feature up to 28 PCIe® Gen 5 lanes, with additional lanes available based on system vendor specifications (please consult your vendor for detailed information). Notably, these processors are designed with low TDP ratings, ranging from 65 to 170 watts.
All AMD EPYC 4004 models utilize the reliable AM5 socket, providing flexible deployment options for various computing needs. The 12-core AMD EPYC™ 4484PX and 16-core AMD EPYC™ 4584PX take advantage of AMD’s 3D V-Cache™ die stacking technology, which effectively doubles the maximum L3 cache to 128 MB per processor.
Features are only meaningful when they enhance performance and efficiency. Hosting providers and small businesses need robust systems that can handle critical workloads while keeping acquisition and operating costs manageable. Servers powered by the high-performance AMD EPYC 4004 processors, combined with the streamlined memory and I/O features, deliver compelling cost-to-performance ratios for essential workloads. Let’s delve deeper into the impressive performance and value these processors offer by comparing the capabilities and potential cost savings of the 16-core 4th Gen AMD EPYC 4004 processors against the competition.
Target Use Cases
The AMD EPYC™ 4004 processor offers versatility, performance, and cost-effectiveness for a diverse array of computing tasks, from general business applications to compute-intensive workloads. Here are some key use cases:
- Infrastructure Management: Helps organizations optimize IT operations, enhance efficiency, and maintain a robust infrastructure
- Web Hosting: Ideal for hosting websites, middleware applications and edge operations
- Database Management: Effective for managing small to medium-sized databases
- Virtualization: Capable of running multiple virtual machines to handle various workloads
- Gaming: Delivers exceptional performance for game servers and multiplayer gaming environments.
The AMD EPYC 4004 Series Processors provide a powerful solution for businesses looking to upgrade their server capabilities while keeping costs manageable, making them suitable for a broad range of applications.
Competitive Pricing
Small and medium businesses must strike a balance between affordability and performance, particularly when selecting processors for server builds. The cost of processors is a significant consideration in this regard. Here, I will illustrate the exceptional cost-effectiveness of AMD EPYC 4004 processors compared to their competitors, using the following
- 16-core AMD EPYC 4584PX: $699 (which is ~$43.69 per core)
- 8-core Intel® Xeon® E-2388G*: $606 (which is ~$75.75 per core)
- 8-core Intel Xeon E-2488*: $606 (which is ~$75.75 per core)
In other words, the cost per AMD EPYC 4584PX processor core cost is only ~58% of the Intel Xeon costs per core. These prices underscore how AMD EPYC 4004 processors offer competitive affordability alongside leadership performance capabilities, making them a compelling option for small and medium businesses looking to optimize their server infrastructure investments.
General Purpose Workload Performance
SPEC CPU® 2017 is a widely recognized industry standard benchmark for evaluating the performance of general-purpose workloads. SPEC CPU® 2017 measures compute-intensive tasks by stressing processor(s), memory subsystems, and compilers across various computer systems. This benchmark comprises 43 tests categorized into four suites, with this blog focusing on SPECrate® 2017 Integer. Figure 1 shows that a single-socket 16-core AMD EPYC 4584PX system delivers a ~1.73x SPECrate® 2017_int_rate_base performance uplift versus a single-socket 8-core Intel Xeon E-2488 system. The same AMD EPYC 4584PX also yields a ~1.50x performance/the same Intel Xeon E-2488 processor.[3]
Figure 1: 16-core SPECrate® 2017 int_rate_base performance and performance/CPU$
Power Efficiency Advantage
Energy costs pose a significant challenge for both large-scale data centers and small- to medium-sized businesses. The SPECpower_ssj® 2008 benchmark provides a standardized method for assessing the energy efficiency of volume server-class computers, enabling customers to compare energy efficiency across various server configurations. 4th Gen AMD EPYC 4004 processors deliver leadership SPECpower_ssj® 2008 power efficiency. Figure 2 shows a 16-core AMD EPYC 4584PX system exhibiting ~1.81x greater energy efficiency than the same Intel system.[4] Here again, the AMD EPYC4004 processor delivers a significant ~1.57x performance/CPU dollar uplift.[5]
Figure 2: 16-core SPECpower_ssj@ 2008 performance and performance/CPU dollar
Server Side Java Performance
Java is a versatile language used across a wide range of environments, from edge devices to enterprise applications. The SPECjbb® 2015 benchmark evaluates server-side Java® applications by simulating a corporate IT environment handling point-of-sale transactions, online activities, and data mining tasks. The widespread use of Java makes this benchmark relevant to JVM vendors, hardware manufacturers, Java developers, researchers, and academia. SPECjbb® 2015 measures performance using max-jOPS for maximum throughput without strict response time requirements and critical-jOPS for maximum throughput under response time constraints. Figure 3 shows a single-socket 16-core AMD EPYC 4584PX system achieving ~2.02x the performance of a single-socket 8-core Intel Xeon E-2388G system on the SPECjbb® 2015 composite max jOPS metric and ~2.59x the performance of the same Intel processor on the SPECjbb® 2015 composite critical jOPS metric at a performance/CPU dollar result of up to ~2.24x.[6][7]
Figure 3: 16-core SPECjbb® 2015 composite max-jOPS and critical-jOPS performance and performance/CPU dollar
Database Application Performance
MySQL™ is one of the most widely adopted open-source database management systems, utilized by both small and large enterprises for various applications. We used TPROC-C to perform transaction processing performance comparison. Figure 4 illustrates a single-socket 16-core AMD EPYC 4584PX system achieving ~1.50x performance and ~1.30x the performance/CPU dollar of the same Intel processor.[8][9]
Figure 4:16-core MySQL TPROC-C TPM performance and performance/CPU dollar
Media Processing Performance
Media processing performance is crucial as it increasingly emerges as a key edge workload driven by the growing demand for high-quality video content management, distribution, and collaboration tools. FFmpeg serves as a robust multimedia framework, facilitating encoding, decoding, transcoding, streaming, filtering, and playback of video content across both legacy and modern formats and standards. Figure 6 illustrates a single-socket 16-core AMD EPYC 4584PX system achieving average FFmpeg encode speed-ups of ~2.13x (8 jobs @ 2 threads per job), ~2.25x (4 jobs @ 4 threads per job), and ~2.45x (2 threads @ 8 cores per job) compared to the same Intel system at a processor cost only ~15% higher than the Intel processor.[10]
Figure 5: 16-core FFmpeg transcoding performance
A single-socket 16-core AMD EPYC 4584PX system achieves average FFmpeg encode processor performance/CPU dollar uplifts of ~1.85x (8 jobs @ 2 threads per job), ~1.95x (4 jobs @ 4 threads per job), and ~2.12x (2 threads @ 8 cores per job) compared to the same Intel system.[11]
Figure 6: 16-core FFmpeg transcoding performance/CPU dollar
Conclusion
The 4th Gen AMD EPYC™ 4004 Series Processors offer an ideal solution for dedicated hosting providers and small businesses seeking cost-effective options for entry-level server workloads, including infrastructure management, hosting services, middle-tier applications, and gaming. While large enterprises, data centers, supercomputers, and hyperscalers have benefited from the robust performance of the AMD EPYC 7000, 8000, and 9000 Series, the 4004 Series is specifically designed with low core counts and TDP as low as 65 watts, all at competitive prices.
Endnotes
1. EPYC-18: Max boost for AMD EPYC processors is the maximum frequency achievable by any single core on the processor under normal operating conditions for server systems.
2. SPECrate® 2017 int_rate_base results @ 8 cores:
- 1P Intel Xeon E-2488, Score 104, https://spec.org/cpu2017/results/res2024q1/cpu2017-20240129-40767.html
- 1P AMD EPYC 4584PX, Score 180, https://spec.org/cpu2017/results/res2024q2/cpu2017-20240422-43140.html
- 1P Intel Xeon E-2488, Score 104, https://spec.org/cpu2017/results/res2024q1/cpu2017-20240129-40767.html
3. Relative processor performance/$ calculated as relative uplift/(AMD EPYC 4004 1Ku price/Intel Xeon 1Ku price). For the AMD EPYC 4584PX vs. Intel Xeon E-2488, this is 1.73/(699/606)=~1.73/~1.153=~1.50x. These results reflect only processor costs at the listed 1Ku prices. Your actual costs may vary widely due to variables such as but not limited to the number of processors purchased and system configuration.
4. SPECpower_ssj® 2008:
- 1P Intel Xeon E-2488 (8 cores), Score 11182, https://spec.org/power_ssj2008/results/res2024q2/power_ssj2008-20240325-01382.html
- 1P AMD EPYC 4584PX (16 cores), Score 20240, https://spec.org/power_ssj2008/results/res2024q2/power_ssj2008-20240422-01404.html
5. Relative processor performance/$ calculated as relative uplift/(AMD EPYC 4004 1Ku price/Intel Xeon 1Ku price). For the AMD EPYC 4584PX vs. Intel Xeon E-2488, this is ~1.83/(699/606)=~1.81/~1.153=~1.57x. These results reflect only processor costs at the listed 1Ku prices. Your actual costs may vary widely due to variables such as but not limited to the number of processors purchased and system configuration.
6. SPECjbb® 2015 composite max-jOPS and critical-jOPS (16 cores):
- 1P AMD EPYC 4584PX, Scores max-jOPS 73234 & critical-jOPS 57443, https://www.spec.org/jbb2015/results/res2024q2/jbb2015-20240502-01263.html
- 1P Intel Xeon E-2388G, Scores max-jOPS 36185 & critical-jOPS 22201, https://spec.org/jbb2015/results/res2022q4/jbb2015-20221019-00879.html
7. Relative processor performance/$ calculated as relative uplift/(AMD EPYC 4004 1Ku price/Intel Xeon 1Ku price). For the AMD EPYC 4584PX vs. Intel Xeon E-2488 (max-jOPS), this is ~2.02/(699/606)=~2.02/~1.55=~1.75x. For the AMD EPYC 4584PX vs. Intel Xeon E-2488 (critical-jOPS), this is ~2.59/(699/606)=~2.59/~1.55=~2.24x. These results reflect only processor costs at the listed 1Ku prices. Your actual costs may vary widely due to variables such as but not limited to the number of processors purchased and system configuration.
8. E4K-014: A 16-core AMD EPYC™ 4584PX processor delivers an average ~1.50x MySQL TPROC-C Transactions Per Minute (TPM) uplift versus an 8-core Intel® Xeon® E E-2488 running 800 warehouses and 128 virtual users. The HammerDB benchmark tool was used to build and generate the TPROC-C workload. The HammerDB TPROC-C workload is an open-source workload derived from the TPC-C™ Benchmark Standard and as such is not comparable to published TPC-C results, as the results do not comply with the TPC-C Benchmark Standard. AMD system configuration: 1P 16c AMD EPYC 4584PX CPU, 4 x 32GB DDR5 5200 running at 3600 Mt/s, Intel Corporation I210 Gigabit Network Connection (rev 03), 1 x 1TB NVMe®, BIOS 3.13, SMT=ON, NPS=1, OS Ubuntu® 22.04.4 LTS (kernel 5.15.0-105-generic), OS options= vm.swappiness=1, vm.dirty_ratio = 40, net.ipv4.tcp_rmem=4096 87380 16777216, net.ipv4.tcp_wmem=4096 65536 16777216, fs.file-max = 6815744, fs.aio-max-nr = 1048576, MySQL v8.2.0, HammerDB v4.8. Intel system configuration: 1P 8c Intel Xeon E E-2488 CPU , 2 x32 GB DDR5 4800, Intel Corporation I210 Gigabit Network Connection (rev 03, 1 x 1TB NVMe®, BIOS 1.1, Hyperthreading=ON, OS Ubuntu 22.04.4 LTS (kernel 5.15.0-105-generic) options= vm.swappiness=1, vm.dirty_ratio = 40, net.ipv4.tcp_rmem=4096 87380 16777216, net.ipv4.tcp_wmem=4096 65536 16777216, fs.file-max = 6815744, fs.aio-max-nr = 1048576, MySQL v8.2.0, HammerDB v4.8.
9. Relative processor performance/$ calculated as relative uplift/(AMD EPYC 4004 1Ku price/Intel Xeon 1Ku price). For the AMD EPYC 4584PX vs. Intel Xeon E-2488, this is ~1.52/(699/606)=~1.52/~1.15=~1.30x. These results reflect only processor costs at the listed 1Ku prices. Your actual costs may vary widely due to variables such as but not limited to the number of processors purchased and system configuration.
10. E4K-016: Testing by AMD Performance Labs as of June 6, 2024. A 1P 16c AMD EPYC 4584PX system delivers average FFmpeg encode uplifts of ~2.13x (8 jobs @2 threads/job), ~2.25x (4 jobs @ 4 threads/job), and ~2.45x (2 threads @ 8 cores/job) versus a 1P 8c Intel Xeon E E-2488 system encoding the video tearsofsteel-4k.y4m (size, 172.8 GB, duration (secs) 12:14.2, frames 17620, format Raw y4m, resolution 4096x1714, container y4m, codec Raw, encoder libvpx_vp9). Tears of Steel 4K. © Blender Foundation | Mango.blender.org. AMD system configuration: 1P 16c AMD EPYC 4584PX, 4 x 32 GB 5200 Mt/s, NIC I210 Gigabit Network (2 ports), OS disk 1 x 1 TB Samsung SSD 980 PRO NVMe, Data disks 2 x 3.7 TB Samsung SSD 870 NVMe, BIOS 3.13, SMT=ON, NPS=1, OS Ubuntu 22.04.4 LTS (kernel 6.5.0-35-generic), default OS options. Intel system configuration: 1P 8c Intel Xeon E E-2488 CPU, 4 x 32 GB DDR5 3600, NIC I210 Gigabit Network (2 ports), OS disk 1 x 500 GB Micron 7450 MTFDKBA480TFR NVMe, Data disks 2 x 4 TB INTEL SSDPF2KX038T1 NVMe, BIOS 1.1, Hyperthreading=ON, OS Ubuntu 22.04.3 LTS (kernel 6.5.0-060500-generic), default OS options. Results may vary due to factors such as BIOS settings and hardware configurations.
11. Relative processor performance/$ calculated as relative uplift/(AMD EPYC 4004 1Ku price/Intel Xeon 1Ku price). For the AMD EPYC 4584PX vs. Intel Xeon E-2488, this is ~2.13/(699/606)=~2.13/~1.15=~1.85x (8 jobs @ 2 threads/job), ~2.25/(699/606)=~2.25/~1.15=~1.95x (4 jobs @ 4 threads/job), and ~2.45/(699/606)=~2.45/~1.15=~2.12x (2 jobs @ 8 threads/job).