Whenever companies try to sell a product, especially a new processor, it can be difficult to sift through the promotional jargon and objectively interpret the specs. We saw this recently when Intel benchmarked its i7s against Apple’s M1.
AMD’s EPYC 7003 architecture
When AMD released its newest family of server processors yesterday, the company was quick to call one CPU from the family “the world’s highest-performing server processor.” In this article, we’ll evaluate this claim and compare it against similar Intel processors.
The EPYC 7003 Family and the “Highest-performing Server Processor”
While AMD released an entire series of EPYC 7003 CPUs, the third generation of the EPYC family, a clear highlight of the announcement was the EPYC 7763 processor.
New EPYC family features
The EPYC 7763 features 64 x86 based cores, supporting up to 128 threads at a base frequency of 2.45 GHz and a boost frequency up to 3.5 GHz. The CPU also comes with an increased TDP, reaching up to 280 W at maximum.
Basic architecture of the third-generation EPYC family
Many of the processor’s improvements seem to focus on memory, with improvements in branch prediction, misprediction recovery time, and load/store bandwidths all being big talking points. The processor also offers up to 256 MB of L3 cache, which is bolstered by a new, more cohesive cache design allowing for 32 MB of L3 to be shared for each 8-core group.
This allows each core to access more cache than in previous generations (Gen 2 offered 16 MB per 4-core group), resulting in faster memory access and an overall decrease in memory access latency.
Security Etched at the Silicon Level
In addition to these new performance-based features, the new processor family also focuses on security based on AMD’s Infinity Guard platform.
Layered security built into the EPYC 7003 family
One of the implemented security measures is secure encrypted virtualization, a process in which each virtual machine is encrypted with one of 509 unique keys that are known only to the processor, making data more secure and encryption harder to break. This is coupled with secure nested paging, a feature that restricts virtual machine page reading privileges to prevent attacks such as memory remapping.
Beyond these features, the processors also implement AMD’s Secure Memory Protection, Shadow Stack, and Secure Boot processes to fortify hardware security at all levels of device operation.
AMD Benchmarks Against Intel
While AMD did benchmark its new family against Intel processors, most of the tests were against Intel Xeon Gold 6258R processors.
However, the AMD benchmarking did show that AMD’s processors outperformed the Xeon Gold 6258R processors in a number of categories: the report claims EPYC 7763 has 106% faster 2P floating-point performance for HPC and cloud workloads compared to Xeon Gold and 117% faster enterprise performance compared to Xeon Platinum.
The results also showed that an EPYC 7763 system is capable of running 112% more sessions than a Xeon-based system.
Intel Xeon Gold 6258R vs. EPYC7763
The results, in AMD’s calculation, are that the new EPYC family can provide improved performance while requiring fewer server racks. Specifically, AMD estimates that the new EPYC family can afford customers 49% fewer servers, resulting in lower cost, reduced area, and less power consumption overall.
A Worthwhile Candidate for the Data Center?
While it’s difficult to ascertain an objective estimation of the EPYC 7003 CPUs compared to similar processors around the industry, it is clear that the new series could be a worthwhile consideration for designers in data center sectors.
Namely, the high core count, new memory architectures, and improved hardware security may make the processors a candidate for customers seeking to bump performance at a lower cost.
All images used courtesy of AMD
What performance improvements would you like to see at the circuit level for power-hungry applications like the data center? Share your thoughts in the comments below.