AMD Epyc outpaces Intel in performance-per-watt for CDN video workloads
CDN77 analyzed the performance of AMD's Epyc 7742 (Zen 2) server processors compared to Intel's Xeon Platinum 8280, focusing on workloads critical to CDNs such as compression and encryption. The evaluation shows AMD leading in performance-per-watt, cache efficiency, and security features like Secure Memory Encryption (SME).
Key Takeaways
- AMD Epyc 7742 demonstrated superior OpenSSL and 7-zip compression performance compared to Intel Xeon Platinum 8280 benchmarks
- Rome architecture (Zen 2) reduced memory design to a single NUMA node, contributing to a 30% performance gain via improved L3 cache hit rates
- Hardware-level Secure Memory Encryption (SME) on AMD chips provides full-memory encryption with negligible performance impact compared to Intel’s 128MB-capped SGX
- Dual Intel Xeon 8280 processors exhibit nearly double the thermal design power (TDP) of a single Epyc 7742, widening the total cost of ownership gap
Why It Matters
For CDNs, efficiency in encryption and compression directly translates to lower edge operating costs and improved throughput. AMD’s shift toward chiplet-based architecture has effectively neutralized Intel’s historical dominance in high-end server deployments by offering higher core density at equivalent or lower power envelopes. While Intel is attempting to pivot with bfloat16 support in Cooper Lake to target AI-heavy workloads, AMD currently maintains a technological lead in general-purpose cloud and edge delivery. Strategists should monitor if Intel's delayed 10nm transition can successfully reclaim the performance-per-watt crown in the high-performance computing (HPC) and video processing segments.
Additional Context
The competitive landscape between AMD and Intel has intensified significantly through late 2024 and early 2025. Per Mercury Research and Reuters (November 2024), AMD achieved a record 33.9% revenue share in the server CPU market in Q3 2024, up from 31.2% year-over-year. This growth is largely attributed to the launch of the 5th-Gen Epyc “Turin” processors (Zen 5), which scale up to 192 cores in a single socket. Independent testing by Phoronix (October 2024) indicated that a dual-socket AMD Turin system can outperform Intel’s flagship Xeon 6980P (Granite Rapids) by approximately 40% across mixed data center workloads. Intel has responded with its Xeon 6 series, which bifurcates the lineup into Granite Rapids (P-cores) for high-performance tasks and Sierra Forest (E-cores) for cloud-native density. According to Intel documentation (December 2024), the Granite Rapids architecture introduces MRDIMM 8800 MT/s memory support to compete on memory bandwidth, an area where AMD has traditionally led. However, AMD's Turin Dense variants continue to hold a core-count advantage, with 192 physical cores compared to Intel's 128 P-cores, though Intel is slated to release a 288-core Sierra Forest-AP part in early 2025 to contest the efficiency crown. Financial analysts at Bank of America (June 2025) projects that AMD’s revenue share could reach 36% by the end of 2025 as hyperscale providers increasingly migrate to x86 alternatives or custom Arm-based silicon. Major cloud providers, including Amazon, Google, and Microsoft, have all introduced internal Arm designs to optimize their own specific delivery workloads. This suggests that while x86 competition remains fierce, the broader streaming infrastructure market is moving toward a more fragmented hardware ecosystem where workload-specific optimization outweighs raw brand loyalty.
Read full article at cdn77.com